Romance Languages and Linguistic Theory 2010 : Selected Papers from 'going Romance' Leiden [1 ed.] 9789027272478, 9789027203847

Citation preview

Romance Languages and Linguistic Theory 2010

Romance Languages and Linguistic Theory (RLLT) The yearly ‘Going Romance’ meetings feature research in formal linguistics of Romance languages, mainly in the domains of morphology, syntax, and semantics, and, to a certain extent, phonology. Each volume brings together a peer-reviewed selection of papers that were presented at one of the meetings, aiming to provide a representation of the spread of topics at that conference, and of the variety of research carried out nowadays on Romance languages within theoretical linguistics. For an overview of all books published in this series, please see http://benjamins.com/catalog/rllt

Editor Frank Drijkoningen Utrecht University

Volume 4 Romance Languages and Linguistic Theory 2010. Selected papers from ‘Going Romance’ Leiden 2010 Edited by Irene Franco, Sara Lusini and Andrés Saab

Romance Languages and Linguistic Theory 2010 Selected papers from ‘Going Romance’ Leiden 2010 Edited by

Irene Franco Sara Lusini Leiden University

Andrés Saab Leiden University / National University of Comahue

John Benjamins Publishing Company Amsterdam / Philadelphia

8

TM

The paper used in this publication meets the minimum requirements of the American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data Going Romance (Conference) (2010 : Leiden, Netherlands) Romance Languages and Linguistic Theory 2010 : Selected papers from “Going Romance” Leiden 2010 / Edited by Irene Franco, Sara Lusini, Andrés Saab, Leiden University. p. cm. (Romance Languages and Linguistic Theory, issn 1574-552X ; v. 4) Includes bibliographical references and index. 1. Romance languages--Congresses. I. Franco, Irene, 1980- editor of compilation. II. Title. PC11.G65   2010 440--dc23 2012038857 isbn 978 90 272 0384 7 (Hb ; alk. paper) isbn 978 90 272 7247 8 (Eb)

© 2012 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

Table of contents Foreword From Romance clitics to case: Split accusativity and the Person Case Constraint Maria Rita Manzini Contextual conditions on stem alternations: Illustrations from the Spanish conjugation David Embick

vii 1

21

State nouns are Kimian states Antonio Fábregas & Rafael Marín

41

‘I know the answer’: A Perfect State in Capeverdean Fernanda Pratas

65

Stressed vowel duration and stress placement in Italian: What paroxytones and proparoxytones have in common Stefano Canalis & Luigia Garrapa

87

Serial prosodification and voiced stop geminates in Catalan Francesc Torres-Tamarit & Clàudia Pons-Moll

115

Interfacing information and prosody: French wh-in-situ questions Viviane Déprez, Kristen Syrett & Shigeto Kawahara

135

VP Ellipsis: New evidence from Capeverdean João Costa, Ana Maria Martins & Fernanda Pratas

155

Anti-repair effects under ellipsis: Diagnosing (post-)syntactic clitics in Spanish Andrés Saab & Pablo Zdrojewski

177

On the argument structure of the causative construction: Evidence from scope interactions Francesco Costantini

203

Index

221

Foreword The annual conference series Going Romance is an international initiative of the universities of the Netherlands that engages in linguistic research on Romance languages. Since its inception in the eighties of the past century, the conference has ­developed into a major European discussion forum where ideas about language and linguistics and about Romance languages are put in an interactive perspective, giving space to both universality and Romance-internal variation. Since just before the new millennium, the organization publishes a ­proceedings-like volume entitled Romance Languages and Linguistic Theory. An invitation to address a key-note lecture includes the possibility to publish the corresponding paper in the volume. For publications by selected speakers a separate review procedure has been agreed upon. The current volume, Romance Languages and Linguistic Theory 2010, contains a selection of the papers that have been presented at the twenty-fourth Going Romance conference, which was held at Leiden University from December 9–10. The conference was preceded, on December 8, by a workshop on MorphosyntaxPhonology interface theories. We would like to thank all those who contributed to the success of the XXIVth edition of Going Romance. First of all, our thanks go to the invited speakers, the selected speakers, the presidents of sessions, the participants and the discussants, for creating that lively atmosphere during the couple of days that we were together. Our thanks also go to a rather large set of colleagues for scoring the abstracts, helping with the organization, and for – last but not least – formulating comments and reviews on the papers that were submitted. We feel that the scientific quality of the current volume is largely dependent on their positively critical attitude. We are especially grateful to Roberta D’Alessandro and Johan Rooryck for the support with several organizational matters. Finally, our thanks go to the institutions that supported us financially and organizationally: The Royal Netherlands Academy of Arts and Sciences (KNAW) and the local research institute (Leiden University Centre for Linguistics LUCL).

 Foreword

The editors feel that the present volume reflects very well the actual content of the XXIVth edition. We thank Anke de Looper and Susan Hendriks for their work in editing and preparing the final manuscript. Leiden, August 2012 Irene Franco Sara Lusini Andrés Saab

From Romance clitics to case Split accusativity and the Person Case Constraint Maria Rita Manzini We reconstruct the notion of dative in terms of a Q(⊆) category, denoting an ‘inclusion’ relation – to be understood roughly as ‘possession’. In the light of this, we reconsider interactions between dative and 1st/2nd person in Romance. In Italian 3rd person clitics have two separate lexicalizations for accusative and dative, 1st/2nd person clitics have a single dative-like lexicalization. We construe this phenomenon in terms of split accusativity (DOM), i.e. 1st/2nd person object clitics are embedded as datives, as opposed to 3rd person ones that alternate between dative and accusative. We also suggest a reworking of the Person Case Constraint (PCC) as a constraint on the interpretation of Q(⊆) (i.e. valuing of a Q(⊆) probe). Under it, a 1st/2nd person object clitic must be interpreted as the argument of Q(⊆) (i.e. value it) if present, leaving a dative clitic (i.e. bearing Q(⊆) morphology) without an interpretation. Keywords:  case; dative; person; Differential Object Marking; Person Case Constraint

1.  Case: The oblique / dative In the minimalist approach of Chomsky (1995), properties such as gender (nominal class), number and person, which are intrinsically associated with nominal constituents, are lexical features. However, relations, such as theta-roles, are not features, but correspond simply to syntactic configurations. From this perspective, it is potentially problematic to find that Case is treated as a feature, i.e. as nominal class or number rather than as theta-roles. The fact that case is the only feature in Chomsky (1995) which is radically uninterpretable (i.e. which does not have an interpretable counterpart) is a reflex of the deeper difficulty in reconciling its traditionally relational core with its feature status. The solution at which Chomsky (2008) arrives is that the real underlying relation between case assigner and case assignee is an agreement relation, involving phi-features; case is but a reflex of this relation which appears on nominal constituents. It seems to us however that if case is reduced to agreement, the difference between



Maria Rita Manzini

a language which has just agreement (say, Italian) and a language like Latin, which has the ‘case’ reflex of agreement, remains unclear. Note also that within Distributed Morphology, the existence of syncretic ‘exponents’ associated with several descriptive case contexts often leads to the conclusion that the relevant entries are associated only with features of nominal class, number, etc. Hence the question as to what the difference may be between a language with just agreement and a language with the case reflex of agreement does not seem to become any clearer at PF. We will take as our starting point the only modern Romance language which has so-called case on nouns, i.e. Romanian. In (1a) we exemplify the dative plural masculine and feminine with an embedding under ‘give’. In (1b) we show that the forms of the dative completely overlap with those of the genitive, illustrated by a nominal embedding. The syncretism between genitive and dative characterizes all nominal forms in Romanian, so that we may equally well speak of an oblique case. The oblique forms in (1) have three separate layers of inflection. The leftmost layer is the nominal class morphology -i for the masculine plural and -e for the feminine plural. The second layer is an -l definiteness specification; though Romanian is often described as a language with post-nominal articles, here we assume that the definiteness morphology is generated as an inflection within the noun (cf. Dimitrova-Vulchanova & Giusti 1998). Finally the -or ending is associated with case. (1) a. (I)-l am dat băieţ-i-l-or / fet-e-l-or him.it I.have given boy-mpl-def-obl / girl-fpl-def-obl ‘I gave it to two boys / girls.’ b. pahar-ul băieţ-i-l-or / fet-e-l-or glass-msg.def boy-mpl-def-obl / girl-fpl-def-obl ‘the glass of the boys/ girls’

In minimalist approaches -or would be the lexicalization of an uninterpretable feature, which is either directly checked against a similar uninterpretable feature of the head of the construct (verb or noun), as in Chomsky (1995) or is checked as part of an independently defined Agree process. Given the doubts we expressed above (cf. also Baker & Vinokurova 2010), we are interested in exploring alternative analyses. A fairly obvious intuition about case, originally formalized by Fillmore (1968), is that -or in a language like Romanian is the inflectional equivalent of a preposition like to or of in English. If a preposition is a predicate introducing a relation between the argument it selects and another argument, so is case, or at



From Romance clitics to case

least oblique case to which we limit our discussion here.1 Thus in (1b) -or says that the noun that it attaches to (i.e. ‘the boys’ or ‘the girls’) bears the ‘genitive’ (roughly the ‘possession’) relation to the head noun. An idea put forth in very similar terms by various strands of literature is that ‘possession’ is in fact a surface manifestation of the more elementary part-whole relation. Thus Manzini and Savoia (2005, 2007) propose that the Romance clitic ne (syncretic in some varieties between genitive and dative) introduces a pronominal set which is a ‘superset-of ’ some other argument of the sentence (the theme, cf. Burzio 1986). Belvin and den Dikken (1997: 170) define the relation introduced by ‘have’ as ‘zonal inclusion’ in the following terms: “the ‘meaning’ of have … denotes a special kind of inclusion relation … dubbed ‘zonal inclusion’… Entities have various zones associated with them, such that an object or eventuality may be included in a zone associated with an entity without being physically contained in that entity… The type of zones which may be associated with an entity will vary with the entity.” Formally, we notate the relevant relation with ‘⊆’, though as indicated by the discussion that precedes, the ‘inclusion’ relation is to be construed not mathematically but as a looser ‘zonal inclusion’ one. Among DP-internal categories, we suggest that Q, given its relational core, is closest to case morphology instantiating this relation, which we notate Q(⊆), as in (2). (2) √ pahar

D ul

Q(⊆) or D l



√ băiet fet

N i e

The structure in (2) is interpretable as is. The Q(⊆) element takes as its complement its sister DP băieţil-/ fetel- and as its external argument the sister DP to its projection, i.e. paharul. One could nevertheless want to say that Q(⊆) is merged as a functional head as in (3). We could then speak of Q(⊆) as a probe and of its 1.  This interpretive property does not prevent case from being a nominal projection in (2) or (4), hence in particular transparent to binding. An anonymous reviewer reminds us of the transparency of prepositions like French à, English to etc. to binding, leading theorists to conclude that they too represent nominal projections (‘cases’) of sorts. We are non-committal in this latter point.





Maria Rita Manzini

arguments as valuing the probe, where by valuation we mean saturation of the arguments of Q(⊆).2

(3) Q(⊆) DP paharul DP băietl fetel

Q(⊆) or

Consider then dative, as exemplified in (1a). The morphological structure remains unchanged with respect to the so-called genitive in (2)–(3), and so does the content of the morphological categories involved. In particular, the second internal argument of ditransitives has been argued to be connected to possession at least since Kayne (1984). All that changes is the syntactic embedding. In (1a) Q(⊆) takes as its internal argument its sister DP băieţil-/fetel- and as its external argument the sister to its projection, i.e. the accusative clitic l, reconstructed in (4) in thematic position for simplicity. Correspondingly, the second internal argument of ‘give’, i.e. the traditional dative, participates in fixing the reference of the first internal argument, i.e. the accusative, by denoting a superset including it. In this instance, an abstract Q(⊆) functional head built into the structure would have the same position as the low Appl head of Pylkkänen (2002), Cuervo (2003) with which it could be taken to identify (this area of the structure, not necessarily assumed here, is shaded in (3), cf. Footnote 2).

2.  We have long taken the position that there are no uninterpretable properties, on grounds of restrictiveness and others (Manzini & Savoia 2008, cf. Brody 2003). Furthermore, if overtly merged material (for instance the case inflection in (2)) is sufficient to build the LF interface, we take it that recourse to functional categories (for instance the q(⊆) head in (3)) is unwarranted. On the difficulties that this latter conclusion appears to create, cf. Manzini and Roussou (2011). An anonymous reviewer is concerned about how direct cases fit into the picture. In a number of recent works (Manzini & Savoia 2011a, 2011b, 2012) we take cases in general to be elementary predicates/ operators. Nominative corresponds to the EPP operator (on its operator status, cf. Manzini & Roussou 2011); accusative is the Elsewhere case, hence simply the λ (argument-of) operator.



From Romance clitics to case

(4) Q(⊆) V dat DP l





DP băietl / fetel

Q(⊆) or

We will henceforth concentrate on datives, disregarding genitive structures and syncretism. The reason why we began our discussion with a language like Romanian, which has the syncretism between dative and genitive, is that the view of dative we wanted to put forth is perhaps most intuitively apprehended starting from syncretic oblique realizations (including genitive) – and it is at least indirectly supported by an account of the latter. Of course this syncretism is widely attested, characterizing, for instance, Albanian (Manzini & Savoia 2011a, 2011b) as well as the Romance varieties in which genitive and dative 3rd person clitics are both lexicalized by ne, as mentioned above. At the same time languages where no syncretism obtains are simply languages that lexically disambiguate between the two structures or readings in (2)–(3) and (4). 1.1  Split accusativity As is well-known, in many languages case assignment depends on the r­ eferential content of the argument DPs. This is often described in terms of an animacy hierarchy. Here for instance is the classical discussion by Dixon (1979: 85–86) based on the ‘potentiality of agency’ scale, i.e 1st person < 2nd person < 3rd person < Proper name < Human < Animate < Inanimate: “It is plainly most natural and economical to ‘mark’ a participant when it is in an unaccustomed role … A number of languages have split case-marking systems exactly on this principle: an ‘ergative’ case is used with NP’s from the right-hand end, up to some point in the middle of the hierarchy, and an ‘accusative’ case from that point on, over to the extreme left of the hierarchy. Though the phenomenon is often referred to under the heading of split ergativity, it is evident that in the typological continuum it touches what we may call split accusativity”. Similarly, using a different terminology, Aissen (2003: 473) states that “the ­factors that favor differential subject marking will be the mirror image of those that favor D[ifferential]O[bject]M[arking]”. The issue of interaction between case and person/ animacy will be considered here from the perspective afforded by the Romance languages and hence by split





Maria Rita Manzini

accusativity rather than ergativity. Split accusativity phenomena in Romance most often involve ‘accusative’ vs. ‘dative’. Consider for instance the fact that generally, Romance varieties present the same clitic form for the accusative and the oblique in the 1st/2nd person, as in Italian (5a), while in the 3rd person clitics distinguish accusative and dative, as in (5b) vs. (5c). (5) a. Mi ha aiutato / telefonato (to.)me he.has helped / telephoned ‘He helped/ tephoned me.’ b. Lo ha aiutato him he.has helped ‘He helped him.’ c. Gli ha telefonato to.him he.has telephoned ‘He telephoned him.’

The classical approach to asymmetries like these is to postulate a single underlying case system – and to assume that morphological mechanisms (perhaps Impoverishment and underspecification, in the way of Distributed Morphology) are responsible for the surface syncretism. Our objection to this type of account is that it assumes the existence of a component, the morphology, essentially devoted to opacizing the syntax. This is a distinctly non minimalist assumption; technically, in particular, Distributed Morphology violates Chomsky’s (1995) Inclusiveness, because of Late Insertion.3 On this basis, we will tentatively maintain that the descriptive syncretism reflects a genuine syntactic generalization. In (5a), the mi, ti 1st/2nd person forms have the same -i inflection as the 3rd person dative gli in (5c).4 This inflection clearly contrasts with that of the accusative in (5b), corresponding to nominal class morphology (-o for the masculine singular). The overt dative morphology of mi/ ti suggests that these forms are not directly embedded as the internal argument of the event. Rather, their embedding requires the presence of Q(⊆). In other words, the structure of embedding of mi/ ti in (5a) remains constant, as in (6), despite the fact that two different structures

3.  This point needs to be qualified. One could take the position that Inclusiveness is r­ estricted to syntax, leaving out Morphological Structure. Though such an assumption is possible, we question its necessity. If Manzini and Savoia (2005, 2007, 2011a, 2012) are correct, all Romance and Albanian morphology (including agreement, Case, voice, etc.) can be accounted for without any recourse to Late Insertion; this suggests that the latter is not empirically forced. 4.  This corresponds to the Latin -i inflection of the dative singular (in all classes excepting the 2nd), also syncretic with the genitive (in the 1st class); Romanian has an oblique singular -i inflection as well (-lu-i for the masculine and -e-i for the feminine).



From Romance clitics to case

of embedding are implied by the predicates ‘help’ and ‘telephone’ in (5b) and (5c) respectively with 3rd person clitics. (6)

Q(⊆) V telefonato aiutato √ m

Q(⊆) i

Our analysis yields the correct results at the morpholexical interface. But what about the interpretive interface? We have seen that with ditransitive verbs Q(⊆) establishes a relation between the argument to which it attaches and another argument present within the predicate (the VP). The question is what Q(⊆) does in (5a). We propose that in this instance the two arguments of Q(⊆) are the 1st/2nd person clitic and – we assume – the event itself. Intuitively, transitive predicates can be paraphrased by an elementary predicate associated with an eventive name. Thus aiutare ‘to help’ alternates with dare aiuto a ‘give help to’. Hale and Keyser (1993), Chomsky (1995) formalize this intuition about the complex nature of transitive predicates by assuming that they result from the incorporation of an elementary state/ event into a transitivizing (typically causative) predicate. Within such a conceptual framework it becomes clearer what we mean when we say that Q(⊆) takes as its arguments the (elementary) state/ event and the pronoun. Thus (5a) can be informally rendered as ‘He caused me to have help/ a telephone call’ or more directly ‘He caused me help/a telephone call’, as schematized in (7) (where EA is the External Argument and CAUSE informally notates the transitivizing predicate5). (7) a. EA CAUSE [Q(⊆) (telephone call, me)] b. EA CAUSE [Q(⊆) (help, me)]

We claim that the only way the 1st/2nd person pronoun in (7) can be introduced as an object is by making it into a ‘possessor’ i.e. an element that takes in its ‘zonal inclusion’ domain the elementary event – here ‘help, ‘telephone’. By contrast, 3rd person complements of ‘help’ (or rather ‘cause help’) are embedded in a canonical transitive (causative) structure comprising a nominative agent and an accusative

5.  In minimalist syntax the transitivizing predicate is standardly built into the structure in the form of a v functional head.





Maria Rita Manzini

theme, as in (5b) – though 3rd person arguments of ‘telephone’ are not, as in (5c) (a lexically governed alternation). This proposal applies to other languages where 1st/2nd person objects are associated with a different case system than 3rd person ones. The clitic system of standard Italian is used here not as the strongest example, but simply as an easily presented one. Manzini and Savoia (2010) review and analyze several Romance full pronouns systems with the relevant properties. Another illustration comes from Albanian, for instance the Arbëresh (Italo-Albanian) variety in (8) (from Manzini and Savoia to appear a). The single 1st person form mua in (8b) and (8c) contrasts with the two different accusative (atɒ) and dative (atiçǝ) forms for the 3rd person pronoun in (8a) and (8c), as for other DPs in general. (8) a. pan atɒ they.saw him ‘They saw him.’

Vena di Maida (Arbëresh)

b. mǝ pan mua me they.saw to.me ‘They saw me.’ c. j-a / m-ε jε atiçǝ / mua to.him-it / to.me-it he.gives to.him / to.me ‘He gives it to him/ to me.’

There are many languages where the pronoun system displays the ordinary accusative vs. dative paradigm for 1st/2nd person as for 3rd (Greek clitics, as briefly reviewed in (20), are an example). None of what precedes applies to them. From the perspective of UG, a grammar characterized by a uniform accusative-dative distribution across persons and a grammar characterized by systematic association of 1st/2nd person objects with oblique are equally possible. The parametric value holding in a specific language is learned, and there is no reason to believe that it depends on some constraint blocking the other value (unless we want to say that what is learned is the constraint itself, cf. Calabrese (2008)). The traditional view is (as already mentioned) that underlying cases are always identical for 1st/2nd and 3rd person, while the syncretism between dative and accusative in the 1st/2nd person is purely a morphological matter. This explains certain facts, which appear to be problematic for the present approach. Specifically, the 1st/2nd person argument of ‘help’ raises to the nominative position in the passive – while that of ‘telephone’ does not, as shown in (9). (9) Sono stato aiutato / *telefonato I.am been helped /    telephoned ‘I was helped / called on the telephone.’



From Romance clitics to case

It is intuitively clear what the relevant difference between the two predicates in (9) is. With telefonato ‘telephoned’, which requires all objects to be embedded as datives, Q(⊆) embedding is selected by the predicate (see the discussion of (7a) above). With aiutato ‘helped’ the present hypothesis (cf. (7b)) is that Q(⊆) is introduced only in order to provide an embedding for the 1st/2nd person as an object, while it is not selected by the predicate. Suppose that a dative argument introducing Q(⊆) is merged, for instance mi in (10). This argument satisfies the selectional requirement of ‘telephone’, but precisely because of its case properties, it is not available for satisfaction of the EPP. The passive cannot then be derived. (10) *[telefonato]/*[aiutato]

[m [Q(⊆) i]]

The other possibility is that a nominative pro is inserted, as in (11). Since pro is available for satisfaction of the EPP, the passive can be derived, but only with ‘help’ and not with ‘telephone’, as desired. Indeed pro does not introduce Q(⊆) and the selectional properties of ‘telephone’ are violated.6 (11) [aiutato]/*[telefonato] pro

6.  One of our anonymous reviewers suggests another difference between the argument of aiutato ‘helped’ and telefonato ‘telephoned’, namely past participle agreement. In Italian it is possible to have 1st/2nd singular agreement with participles like aiutato – but not with telefonato, as in (i). Like the passive in the text this pattern would seem to favor the more traditional analysis in terms of a deep abstract case independent of surface realizations. However agreement with aiutato is strictly optional in (i), though it would be obligatory for a 3rd person accusative clitic. In fact, in many Romance varieties the non-agreeing pattern in (i) is the only attested one (Manzini & Savoia 2005: § 5.1.2). (i) Vi ha aiutati / aiutato / telefonato / *telefonati You(pl) he.has helped-pl / helped-sg / telephoned-sg /    telephoned-pl ‘He has helped/ telephoned you.’ In the present perspective (i) should fall out of the same minimal difference between the two predicates ‘help’ and ‘telephone’ discussed in connection with passive in the text. We assume that since telefonato ‘telephoned’ selects Q(⊆), its agreement morphology should be interpreted as the internal argument of Q(⊆) – which is impossible in Romance. On the ­contrary we assume that the agreement morphology on aiutato ‘helped’ is interpreted as its theme. It must then be the case that this agreement morphology can be linked to the 1st/2nd person argument by a relation closer to chain formation than to feature checking (cf. Manzini & Savoia 2005, 2007 for a model of ‘agreement’ as chain formation). Under alternative theories the fact that in (i) agreement is at best optional (and often unattested) must give rise to some readjustment anyway.



 Maria Rita Manzini

2.  The person case constraint The split accusativity phenomena considered in Section 1.1 involve the embedding of one argument, notably the 1st/2nd person, leaving other aspects of sentential organization untouched. More complex patterns arise, where the embedding structure required by a given argument, e.g. again 1st/2nd person pronouns, depends on the overall referential properties and embedding structures of the other arguments. The Person Case Constraint (PCC) is an instance of this. There are several versions of the PCC but they all involve mutual exclusion between dative and 1st/2nd person (or possibly 1st person) objects, as in (12). (12) *Mi gli prendono come segretaria     me to.him they.take as secretary    ‘They take me as his secretary.’

Much recent literature on the PCC (among others Anagnostopoulou 2005; Bianchi 2005; Rezac 2008) shares the idea that the PCC depends on Agree/Move and on a Minimality constraint on it. However it also shares the assumption that the PCC depends entirely on person features – in other words the notion of case essentially disappears from the formal reconstruction of the phenomenon. For Rezac (2008: 68–69), what he calls the Case/Agree account of the Person Case Constraint “aims to explain the Person Case Constraint as a consequence of relativized minimality, whereby the intervening X in [(13)] blocks person Agree between Y and … H”. Specifically “dative X prevents H-Y person Agree”. Hence Y is disallowed in a dative configuration if it is 1st/2nd person (“person Agree”). (13) [H [X-DAT [Y

The problem is “the nature of the intervention of the dative for person Agree … Much work seeks to solve the riddle of this quirky partial intervention of the dative” (Rezac ibid.). For instance Anagnostopoulou (2008: 18) assumes that “1st, 2nd and reflexive pronouns are [+person] pronouns … while the person specification of 3rd person pronouns depends on the type of Case they have. Accusative-nominative/ direct object 3rd person pronouns lack person features altogether. On the other hand, 3rd person dative/indirect object arguments are understood as animate / affected, they encode point of view, properties encoded through person features”.7

7.  Anagnostopoulou explicitly refers to Adger and Harbour’s (2007) tripartite feature ­ontology under which (in her construal) 1st/2nd person are [+person], 3rd person datives are [–person], 3rd person accusatives lack a [person] feature. This feature ontology seems to us questionable since it is difficult to see what the difference may be between having the [non person] property and not having the [person] property.



From Romance clitics to case

Because of this, the dative checks the Person (P) feature of v in (14), which prevents checking by the 1st/2nd person, and ultimately yields illformedness. (14) [v[P]

[Dat [V 1st/2ndP

From an empirical point of view, there are reasons to doubt the connection between dative and animacy. Thus in Italian, datives clearly occur with inanimates, as in (15), and – we may add – locatives with animates, as in (16). (15) (Al vestito) gli/ci ho rifatto l’orlo (to.the dress) to.it/there I.have re-made the hem ‘I sew the dress’s hem on again.’ (16) (A mio padre) ci/gli somiglio (to my father) there/to.him I.resemble ‘I resemble my father.’

But suppose we grant to Anagnostopoulou (2008) her feature ontology (though cf. Footnote 7 on its richness). Another problem is that the relevance of the notion of dative, overtly denied, is surreptitiously reintroduced through the [person] feature. In other words, in the absence of any additional constraints, we expect features and functional specifications to recombine freely: how is then the link between dative case and [person] predicted? If it is just stated, then [person] becomes a diacritic (acceptable within an Agree account) for the very notion of dative that (for no reason except its theoretically unclear status) one wants to avoid. In fact the PCC is not so much a constraint as a family of constraints. The description that we have provided so far for the PCC implies that all combinations of a dative with a 1st/2nd person object are excluded. This corresponds to the socalled strong PCC, attested for instance in Greek (cf. (20) below). Yet Italian has at best what is generally called the weak PCC, i.e. a mutual exclusion between 3rd person datives and 1st/2nd person objects, allowing for the combination of two 1st/2nd person clitics, as in (17), where both assignment of argument roles (‘me to you’ or ‘you to me’) are generally deemed possible (Bianchi 2005; Cardinaletti 2007).8

8.  Things are even more complex than this since some data of Italian points to the so-called Me-first PCC, i.e. what really matters is that a 1st person should not cooccur with a (3rd person) dative. Thus Cardinaletti (2007) quotes (i) from early Italian. (i) e dì come gli ti se’ tutta data(Dante, Fiore; 173,2) and say how to.him yourself are all given      ‘and say how you gave yourself all to him’



 Maria Rita Manzini

(17) ?Mi ti prendono come segretaria    Me you they.take as secretary ‘They take me as your secretary/ you as my secretary.’

For Anagnostopoulou (2005) while the strong PCC depends on Agree, along the lines of (14), in weak PCC grammars it is Multiple Agree that applies, subject to a constraint that blocks dishomogenous features in the search space of the v probe. This filters out combinations of 3rd person with 1st and 2nd, as opposed to combinations of 1st with 2nd. Nevins (2007) in turn generalizes Multiple Agree; variation (including the so-called Me-first PCC and other patterns, cf. Footnote 8) is defined entirely by the constraints placed on the search space of the v probe (dishomogenous features, or other). Therefore “the strong and the weak PCC both arise in configurations in which the two objects enter Agree with a single Probe. However, they should be seen as separate constraints” (Anagnostopoulou 2008: 18), since the mode and/or the content of the probing differ. 2.1  A Case-based account In short, syntactic accounts of the PCC are generally based on [person] (probes). One of our desiderata is to dispense with the rich ontology of person reference implied by these accounts. First, we want to avoid such empirically dubious notions as the identification of dative with animacy – as well as three-valued feature systems like Anagnostopoulou’s (2008), Adger and Harbour’s (2007) (cf. Footnote 7). Second, we expect the weak and the strong PCC to be essentially the same constraint. Third, existing accounts of the PCC are oblivious to the potential connection between the PCC and split accusativity (DOM) phenomena, which also involve 1st/2nd person pronouns and the dative (here the argument of Q(⊆)). We want to capture this connection. Consider then the Italian examples, beginning with (12) or schematically (18), where the shaded area in the structure corresponds to the abstract functional head necessary if the analysis is to be couched in probe-goal terms. Informally, what goes wrong in (18) is that if mi is interpreted as the ‘possessor’ in the ditransitive structure, then gli cannot be so interpreted – which means that it cannot have an interpretation. In other words, if mi values Q(⊆) as in (18), then gli cannot value it, leading a failure in the interpretation of its Q(⊆) features. (18) [Q(⊆)(mi) [ m [Q(⊆)i]] [gl[Q(⊆)i]]

The reason why it is mi in (18) that must be interpreted as the argument of Q(⊆) (i.e. value it) is the core of the PCC, namely that a 1st/2nd person clitic must be interpreted as the internal argument of Q(⊆) (i.e. value it) when both are instantiated in the same local domain.



From Romance clitics to case 

Consider (17), which is generally deemed acceptable. Whether it is mi ‘me’ or ti ‘you’ that is interpreted as the possessor in the ditransitive structure, the PCC as just defined is satisfied, because a P(erson/Participant) clitic is interpreted as the argument of Q(⊆) (i.e. values it). Furthermore we can assume that the other P clitic can receive an alternative interpretation – namely by the discussion in Section 1.1 as the ‘possessor’ of the elementary event. In probe terminology, we may assume that whether it is mi or ti that values the Q(⊆) probe selected by the predicate, the other can introduce its own Q(⊆) probe, as in (19). (19) a. [Q(⊆)(ti) [Q(⊆)(mi) [m [Q(⊆)i]] [t[Q(⊆)i]] b. [Q(⊆)(mi) [Q(⊆)(ti) [m [Q(⊆)i]] [t[Q(⊆)i]]

Consider Greek next. The first difficulty is that in Greek the oblique (traditionally ‘genitive’) 1st/2nd person forms mu/ su ‘to.me/ to.you’ are clearly distinct from the accusative forms me/se ‘me/ you’. The -u nominal inflection in Greek can be taken to be a realization of Q(⊆), but the -e morphology is not. Therefore the PCC violations in (20) do not depend on the cooccurrence of two dative/ Q(⊆) elements. Furthermore, Greek, unlike Italian, has the strong PCC, excluding cooccurrences of a 1st/2nd person accusative not only with a 3rd person dative, as in (20a), but also with a 1st/2nd person dative, as in (20b) (from Anagnostopoulou (2008)). (20) a. *Tha tu me stilune    future him.obl me.acc send.3pl     ‘They will send me to him.’ b. *Tha su me sistisune    future you.obl me.acc introduce.3pl    ‘They will introduce me to you.’

Let us begin with (20a), as schematized in (21). In present terms, in (20a), though Q(⊆) is introduced by the 3rd person dative tu, the 1st person me must nevertheless be interpreted as its argument. In other words, the PCC works as in Italian, i.e. Q(⊆) must take as its internal argument (i.e. as the possessor) a 1st/2nd person argument where present. In probing terms the Q(⊆) probe must be valued by the 1st person clitic – but me cannot value it because it is not associated with Q(⊆) morphology. (21) [Q(⊆)(me) *[t[Q(⊆)u]]

me

Let us then consider (20b). If me is interpreted as the argument of Q(⊆) (i.e. values it), as in (22a) then the result is illformed more or less for the same reasons as in (21). In (22b), by contrast, the morphosyntactic properties of su successfully support its interpretation as the ‘possessor’, i.e. its valuation of Q(⊆). The reason

 Maria Rita Manzini

why the result is illformed is the core of the strong vs. weak PCC parameter. The weak PCC requires that Q(⊆) be checked by some P element in its local domain, if present; the strong PCC requires that Q(⊆) be checked by all P elements – with the result that it cannot be satisfied in a double P object configuration. (22) a. [Q(⊆)(*me) *[s[Q(⊆)u]] me b. [Q(⊆)(*su)    [s[Q(⊆)u]] me

Now, the account of the PCC that we have been proposing is distinctly shallower than those reviewed in Section 2, since it simply states the association of Q(⊆) with 1st/2nd person. In the spirit of the accounts in Section 2, one may try to derive this association though Relativized Minimality. However we are not interested in pursuing this line of research for two reasons. First we are not convinced that it is necessitated by empirical evidence, as we will see in a moment in ­connection with Rezac (2008); second, it creates certain very general, but in our view very real problems when it comes to ‘repairs’ (Section 2.2). Rezac (2008) argues that evidence from Basque strongly supports a Relativized Minimality account for the PCC. In Basque what are constrained by the PCC are agreement clusters, as in (23)–(24). Rezac’s observation is that the very same cluster of 1st absolutive and 3rd dative is allowed in (23) in a configuration comparable to ‘I drew closer to Peru’ and not in (24), in a configuration comparable to ‘I am to Miren’s liking’ (glosses are simplified here). (23) Nii Peru-rij hurbildu ni-a-tzai-oj. I.abs Peru-dat approached 1-aux-3 ‘I approached Peru.’ (24) */??Nii Miren-ij gustatzen ni-a-tzai-oj.      I.abs Miren-dat liking 1-aux-3      ‘Miren likes me.’

Rezac further notes that the behavior in (23)–(24) correlates with other properties. With locative predicates in (23), the absolutive can bind a reflexive dative – while the reverse holds with experiencer predicates in (24), i.e. the dative can bind a reflexive absolutive. Therefore in his analysis, (24) has a di-transitive structure, with the absolutive argument (the theme) generated lowest in the vP and the dative intervening between it and its v probe. This derives the PCC violation under the Relativized Minimality schema in (13). By contrast, Rezac proposes that in (23), the absolutive enters the derivation above vP, yielding an ABS-DAT structure and crucially no PCC violation under his assumptions. Rezac states that “details of the structure of ABS-DAT verbs are not relevant” (i.e. of the structure proposed for (23)). Unfortunately, the absolutive appears to be a theme in (23) exactly as in (24) so that by the configurational theory of



From Romance clitics to case 

theta-roles of Chomsky (1995) it should have the same position in both sentences. But whether this is a real problem or not, note that in order to mimic Rezac’s account here, we just need to assume that the absolutive argument in (23) is inserted higher than the VP domain (or the domain of Q(⊆)) where the PCC is computed in (18). It is irrelevant whether the derivation in (18) is blocked by an intervention constraint like Relativized Minimality or by some other constraint, as it is here (more on Relativized Minimality in Section 2.2).9 Summing up, we propose a syntactic account of the PCC based on dative, i.e. Q(⊆) (probes), rather than on person (probes). As desired, this allows us to dispense with the rich ontologies of person reference implied by alternative accounts. ­Second, the weak and strong PCC are effectively the same constraint, requiring association of Q(⊆) with a 1st/2nd person clitic, where present. If some P clitic (where present) must be associated with Q(⊆), the weak PCC is derived; if all P clitics must be associated with Q(⊆), we derive the strong PCC. Finally, and importantly here, the PCC is conceptually (though not causally) connected to split accusativity (DOM) phenomena which also involve the association of 1st/2nd person pronouns with the ‘possessor’ (i.e. argument of Q(⊆)) role. In these respects, the relatively shallow formulation of the PCC advocated here contains elements of unification (explanation) that more abstract formulations seem to lack.

9.  In a way reminiscent of (23), Romance clitic clusters of the type in (12), though excluded in ditransitive contexts, have been argued to be possible in other contexts, for instance when one of them is an ‘ethical dative’, as in Catalan (i), quoted by Adger and Harbour (2007). (i) No te li faran res. not to.you to.him they.will.do anything ‘They won’t do anything to him (on you).’ In recent literature the issue of the varying types of datives has been taken up in the Applicative literature, which proposes the existence of two Appl heads (Pylkkänen 2002; Cuervo 2003). The low Appl head corresponds to the Q(⊆) element considered so far. The ethical dative of traditional literature corresponds to a high Appl head. In present terms we can translate the insights of the Applicative literature by assuming that Q(⊆) can structure not only the low predicative area of the sentence, but can also introduce an operator/ predicate attached much higher in the sentence. An ethical dative then is an argument x which takes an entire proposition/ situation in its domain of inclusion – hence the entire situation is construed as being to the benefit or to the detriment of x. Now, we excluded (18) in the text on the basis of the idea that the 1st/2nd person clitic must be interpreted as the argument of Q(⊆) (i.e. must value it) leaving the Q(⊆) features of the 3rd person clitic without an interpretation. We propose that this needs not happen in (i), to the extent that two different Q(⊆) elements are introduced in two different domains. The clitic te introducing/ valuing the higher Q(⊆) needs not interfere with the lower Q(⊆) introduced/ valued by li.

 Maria Rita Manzini

2.2  A grammar without repairs We now get to the important question of repairs. For instance, a PCC violation can be circumvented by inserting the locative in place of the 3rd person dative in some varieties of French (Rezac 2006), as in (25). Rezac (2006) explicitly excludes that the locative may be treated as a ‘paraphrase’ for the 3rd person dative. The reason is that the supposed paraphrase is restricted to a particular context. Therefore Rezac (2006) advocates a model under which the violation of the PCC is overcome in the derivation by the merger of an ‘additional probe’ checking the locative (on ­additional probes cf. Bejar & Rezac 2009). (25) Philippe vous y présentera Philip you there will.introduce ‘Philip will introduce you to him.’

However adopting the view that a violation is first introduced and then repaired is quite expensive. At worst, it involves backtracking, since the derivation survives point of crashing (the Relativized Minimality violation yielding the PCC) to achieve well-formedness. At best, it involves Late Insertion (hence a violation of Inclusiveness) since it is the derivation that decides what lexicalization the argument will have, and therefore lexicalization is forced to take place at the end of it. Because of these reasons, we effectively embrace the view that Rezac rejects, namely that examples like (25) involve no violation (of Relativized Minimality or other) and no repair – but only alternative means of lexicalization (‘paraphrase’). There is no contradiction between the notion of alternative lexicalization and that of contextual restriction. In (25) the form of the problem is that y lexicalizes the second internal argument of présentera ‘will introduce’ as a locative, but only in the context defined by the 1st/2nd person clitic. By applying the conclusions of Section 2.1 to French, in (25) the Person clitic is interpreted as the internal argument of Q(⊆) (i.e. values it), along the lines of (26). We can then describe the relevant ‘repair’ simply by saying that the second argument of a ditransitive can be lexicalized by a locative (directional) in the domain of an independently satisfied (valued) Q(⊆). (26) [Q(⊆)(vous) [ vous [ y

In other words, what we propose is that the context created by the Person clitic in (26) should not only be viewed as creating a potential violation but also as enabling



From Romance clitics to case 

lexical choices impossible in other contexts.10 In (26) the Person clitic introduces a Q(⊆) possessor structure (or values it). The Q(⊆) context so created allows for the lexicalization of the second argument of the ditransitive as a locative directional (i.e. as a goal) at the same time as it blocks its lexicalization as a possessor (i.e. as a dative). In the absence of a Person clitic, a dative clitic is required to introduce/ value Q(⊆) and a locative (directional) clitic does not suffice. In this sense there is no violation (of Relativized Minimality or other) and no repair – but a set of (language specific) lexicalization choices, encompassing both descriptive violation and descriptive repair. A potential PCC violation can be rescued in other ways, in particular by introducing the 3rd person oblique as a non-clitic pronoun, for instance as a possessive pronoun in (27). (27) Mi prendono come suo assistente me they.take as his assistant ‘They take me as his assistant.’

This leads us to a question implicit throughout, namely why the PCC would hold of clitics, but not of DPs. In the Agree accounts of the PCC the implicit assumption seems to be that clitics are agreement morphology. From the present point of view, the most natural move is to argue that the interpretive (or probe-goal) relations that we have played out in the VP domain in (18)–(19) or (21)–(22) actually characterize a dedicated clitic domain, of the type postulated by Sportiche (1996) or by Manzini and Savoia (2005, 2007). In conclusion, our goal has been to reconstruct the notion of oblique case in terms of a Q(⊆) category and on the basis of this, to explain both split accusativity (DOM) phenomena and Person Case interactions in the PCC. These two phenomena are conceptually (not causally) linked in the present account, though not in alternative analyses, as far as we can see. We argued in favour of a shallower ­formulation of the PCC and more generally in favour of projection of syntactic structure from the lexicon, in keeping with minimalist postulates, and hence against theoretical devices that undermine it, including repair and late lexicalization.

10.  A relevant issue in this connection is how the mutual exclusion and repairs seen in the PCC compare to other descriptively similar phenomena in the clitic domain (e.g. the ‘double-l ’ constraint and the Spurious se of Spanish) or beyond (e.g. the double-n constraint of English vs. Romance Negative Concord). We refer the reader to Manzini and Savoia (2005, 2007, 2011a) for accounts of these phenomena along similar lines to those adopted here.

 Maria Rita Manzini

References Adger, David & Harbour, Daniel. 2007. “The Syntax and Syncretisms of the Person Case Constraint.” Syntax 10: 2–37. Aissen, Judith. 2003. “Differential object marking: iconicity vs. economy.” Natural Language & Linguistic Theory 21: 435–483. Anagnostopoulou, Elena. 2005. “Strong and weak person restrictions: A feature checking analysis.” In Clitics and Affixation, Lorie Heggie & Fernando Ordoñez (eds), 199–235. Amsterdam: John Benjamins. Anagnostopoulou, Elena. 2008. “Notes on the Person Case Constraint in Germanic (with special reference to German).” In Agreement Restrictions, Roberta D'Alessandro, Susann Fischer & Gunnar Hrafn Hrafnbjargarson (eds), 15–47. Berlin: Mouton de Gruyter. Baker, Mark & Vinokurova, Nadezhda. 2010. “Two modalities of case assignment in Sakha.” Natural Language and Linguistic Theory 28: 593–642. Bejar, Susana & Rezac, Milan. 2009. “Cyclic Agree.” Linguistic Inquiry 40: 35–73. Belvin, Robert & Den Dikken, Marcel. 1997. “There, happens, to, be, have.” Lingua 101: 151–183. Bianchi, Valentina. 2005. “On the syntax of personal arguments.” Lingua 116: 2023–2067. Brody, Michael. 2003. Towards an elegant syntax. London: Routledge. Burzio, Luigi. 1986. Italian syntax. Dordrecht: Kluwer. Calabrese, Andrea. 2008. “On Absolute and Contextual Syncretism.” In The bases of inflectional identity, Andrew Nevins & A. Bachrach (eds), 156–205. Oxford: Oxford University Press. Cardinaletti, Anna. 2007. “On different types of clitic clusters.” University of Venice Working Papers in Linguistics 17: 27–76. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: The MIT Press. Chomsky, Noam. 2008. “On Phases.” In Foundational Issues in Linguistic Theory. Essays in Honor of Jean-Roger Vergnaud, Robert Freidin, Carlos P. Otero & Maria Luisa Zubizarreta (eds), 133–166. Cambridge, Mass.: The MIT Press. Cuervo, Maria Cristina. 2003. Datives at large. Ph.D. dissertation, MIT Cambridge, Mass. Dixon, Robert M.W. 1979. “Ergativity.” Language 55: 59–138. Dimitrova-Vulchanova, Mila & Giusti,Giuliana. 1998. “Fragments of Balkan Nominal Structure.” In Possessors, predicates and movement in the determiner phrase, Artemis Alexiadou & Chris Wilder (eds), 333–360. Amsterdam: John Benjamins. Fillmore, Charles. 1968. “The case for case.” In Universals in linguistic theory, Emmon Bach & Robert Harms (eds), 1–88. London: Holt, Rinehart and Winston. Hale, Kenand & Keyser, Samuel J. 1993. “On argument structure and the lexical expression of syntactic relations.” In The view from Building 20, Ken Hale & Samuel J. Keyser (eds). Cambridge, Mass: The MIT Press. Kayne, Richard. 1984. Connectedness and binary branching. Dordrecht: Foris. Manzini, M. Rita & Roussou, Anna. 2011. “Movement: A view from the left.” Parallel Domains, USC, Los Angeles, 6–8 May. Manzini, M. Rita & Savoia, Leonardo M. 2005. I dialetti italiani e romanci. Morfosintassi generativa. Alessandria: Edizioni dell’Orso. Manzini, M. Rita & Savoia, Leonardo M. 2007. A unification of morphology and syntax. Studies in Romance and Albanian varieties. London: Routledge. Manzini, M. Rita & Savoia, Leonardo M. 2008. “Uninterpretable features are incompatible in morphology with other minimalist postulates.” InFoundational Issues in Linguistic Theory.



From Romance clitics to case 

Essays in Honor of Jean-Roger Vergnaud, Robert Freidin, Carlos P. Otero & Maria Luisa Zubizarreta (eds), 5–15. Cambridge, Mass.: The MIT Press. Manzini, M. Rita & Savoia, Leonardo M. 2010. “Case as denotation: variation in Romance.” Studi Italiani di Linguistica Teorica e Applicata (SILTA) 39: 409–438. Manzini, M. Rita & Savoia, Leonardo M. 2011a. Grammatical Categories. Cambridge: Cambridge University Press. Manzini, M. Rita & Savoia, Leonardo M. 2011b. “Reducing ‘case’ to denotation: Nominal inflections in Albanian.” Linguistic Variation 11: 76–120. Manzini, M. Rita & Savoia, Leonardo M. 2012. “Case’ categories in the Geg Albanian variety of Shkodër.” Res Albanicae 1: 23–42. Manzini, M. Rita and Savoia, Leonardo M. To appear a. “The interaction of DOM and ­discourse-linking in arbëresh pronouns: case as nominal property.” In Issues in Albanian Syntax, Giuseppina Turano, R. Memushaj & Flora Koleci (eds). Muenchen: Lincom. Nevins, Andrew. 2007. “The representation of third-person and its consequences for personcase effects.” Natural Language and Linguistic Theory 25: 273–313. Pylkkänen, Liina. 2002. Introducing Arguments. Ph.D. dissertation, MIT, Cambridge, Mass. Rezac, Milan. 2006. “Escaping the Person Case Constraint.” Linguistic Variation Yearbook 6: 97–138. Rezac, Milan. 2008. “The syntax of eccentric agreement: The Person Case Constraint and absolutive displacement in Basque.” Natural Language and Linguistic Theory 26: 61–106. Sportiche, Dominique. 1996. “Clitic constructions.” In Phrase structure and the lexicon, Johan Rooryck & Laurie Zaring (eds), 213–276. Dordrecht: Kluwer.

Contextual conditions on stem alternations Illustrations from the Spanish conjugation* David Embick

University of Pennsylvania This paper looks at the analysis of stem alternations: a type of non-affixal morphological change. Alternations of this type are controversial because they look in principle like they can be analyzed either with distinct stem forms in memory, or with (morpho)phonological rules that derive alternants from a single underlying form. I argue that the locality conditions on contextual allomorphy provide an answer to part of this controversy. It is shown that certain stem alternations in Spanish verbs–diphthongization, as in e.g. pensar/pienso (‘think’); raising, as in e.g. pedir/pido (‘ask’) – cannot be treated with stored stems, because the alternations do not occur under the locality conditions that apply to contextual allomorphy. These alternations must be treated (morpho) phonologically. The implications of this view are explored; this includes a conjecture that reclassifies different types of “morpheme specific” alternations in the grammar. Keywords:  allomorphy; alternations; morphology; Spanish; stems

1.  Introduction This paper examines the phenomenon of stem alternation, also called stem allomorphy. To a first approximation, this is a type of allomorphic alternation that is characterized by a non-affixal change. Examples of stem allomorphy are common. For instance, certain verbs in English undergo changes in the context of the

*  Thanks for comments from audiences at Georgetown University, the University of ­California, Santa Cruz, the University of Delaware, Syracuse University, McGill University, Going R ­ omance 24 in Leiden, the University of Toronto, the University of Connecticut, and Cornell University; also for comments/discussion in Gene Buckley’s fall 2009 phonology seminar at the University of Pennsylvania, as well as Penn’s F-MART reading group in Fall of 2010. For comments on specific points or on earlier written versions, thank you as well to Gene Buckley, Morris Halle, Marielle Lerner, Laurel Mackenzie, Rolf Noyer, and Andrés Saab.

 David Embick

past tense morpheme, to yield alternations like sing/sang, break/broke, tell/told, and so on. Or, in the Spanish conjugation, there are verb-specific alternations in stem-vowels, so that e.g. the verb pensar ‘to think’ has the first person present indicative form pienso, with a diphthong, whereas first person plural pensamos has a monophthong. In each of these examples, the alternation is morphological in the sense that it is triggered by a particular morpheme, or applies to certain morphemes, and not others. Moreover, which morphemes do and do not participate in the alternation is not predictable as far as the synchronic grammar is concerned. The analysis of stem alternations is controversial because most theories make available two distinct ways in which they could be analyzed: one in which the alternants exist as separate objects in memory (this is called Stem Storage below), and another in which the alternants are derived by rule from a single underlying form (I will call this a Morphophonological analysis). The first type of analysis makes stem alternation a kind of suppletive contextual allomorphy; the second makes it part of the phonology, in the broad sense. Both contextual allomorphy and phonological changes have independent motivation in the grammar. The pressing theoretical question is what evidence there is for treating stem alternation with one or the other mechanism. The main line of argument of this paper is that the general theory of locality in contextual allomorphy provides a decisive answer to part of this controversy. Stem storage theories treat different stems as (suppletive) contextual allomorphs. However, there is a certain type of stem alternation that is conditioned by contextual factors in a way that is impossible for contextual allomorphy. Alternations with this property cannot be treated with Stem Storage; instead, they must be treated Morphophonologically. 1.1  Stem alternations in context As noted above, the challenges presented by stem alternations derive from the fact that there appear to be two possible means of treating them. This point is best illustrated by considering the two phenomena that exemplify the two potential analytical options. The first phenomenon, illustrated in (1a), shows two realizations of the second person singular agreement morpheme in Latin: -istī in the Perfect indicative tense of the verb, and -s in other tenses. The second phenomenon, seen in (1b), involves the English plural morpheme, which surfaces as /s/, /z/, or /әz/, in a way that is predictable from the phonology of its host: (1) a. Latin agr[2sg] laudāv-istī ‘You (have) praised’ (perfect) laudā-s ‘You praise’ (present)



b.

Contextual conditions on stem alternations 

English [pl] cat, cat-s (/s/) dog, dog-s (/z/) church, church-es (/әz/)

In each of (1a,b), there is an alternation in the general sense: an object that is “the same” at some level of description (Latin AGR[2 sg]; English [pl]) is expressed by distinct phonological realizations (Latin -istī and -s; English /-s/, /-z/, and /-әz/). Despite this superficial similarity, the patterns in (1a) and (1b) are analyzed differently in most theories of grammar. The class of phenomena represented by Latin AGR[2sg] involves two phonological realizations that are by hypothesis not relatable by the phonology; rather, the realizations are suppletive contextual allomorphs of AGR[2sg]. Unlike Latin AGR[2sg], the English (regular) plural realizations can be related to one another by the phonology, such that the surface realizations /s/, /z/ and /әz/ are derived phonologically from a single exponent that has the underlying form /-z/. The difference between Latin (1a) and English (1b) is thus as follows. In the Latin example, the morphology deals with two distinct objects, -istī and -s, each of which exists in memory as part of a distinct Vocabulary Item. For the English plural, on the other hand, there is one morphological object (i.e. one Vocabulary Item) at play, and its exponent has the underlying phonological representation /-z/; the distinct surface realizations of /z/ are the result of the phonology. Suppletive allomorphy like Latin (1a) and (normal) phonological processes like English (1b) provide two clear endpoints for the study of alternations. The difficult cases are those that do not fit neatly into either of these two extremes. The English sing~sang and Spanish pensar~pienso examples are of this type. They are not part of the “normal” phonology, because the relevant processes apply only to certain morphemes (or are triggered morphologically). At the same time, though, there are reasons for being cautious about treating such alternations as suppletive allomorphy: sing/sang and pens/piens share most of their segmental material, and thus do not look like suppletive allomorphs in any obvious sense. 1.2  The empirical question The preceding subsection examines two different types of alternations: one phonological, and one morphological. In principle, stem allomorphy could be handled in either of these two ways; i.e. either with (morpho)phonological rules operating on a single morpheme, or with distinct morphemes in memory. These two types of theories are defined as follows, employing sing~sang for illustration:1 1.  The Stem Storage type of theory is most familiar in the recent theoretical context within certain Lexicalist theories (see Carstairs-McCarthy 1992 for discussion), as well as in “dual

 David Embick

(MP)  Morphophonological Theory: There is a single underlying form; surface differences are the result of (morpho)phonological rules.[i.e. SING is part of both sing and sang; a rule triggered by T[+past] derives the latter.] (SS)  Stem Storage Theory: There is suppletive contextual allomorphy; the different irregular alternants are stored in memory. [i.e. sing and sang exist as stored alternants, and are inserted in appropriate contexts.] In this paper I put to the side conceptual arguments that have been made in favor of MP or SS theories (see Footnote 2 below). Instead, I develop a line of argument that asks under what locality conditions stem alternations take place (cf. also Kiparsky (1996)). The reasoning is as follows: if stem alternations are treated with SS, they are instances contextual allomorphy and must be subject to the locality conditions that characterize this type of alternation. Thus, a stem alternation that is triggered in a way that is impossible for contextual allomorphy cannot involve stored stems; rather, it must be treated morphophonologically. Along these lines, Section 3 shows that there are stem alternations in S­ panish that are (i) not part of the “normal” phonology, because they are restricted to apply to certain morphemes, but which (ii) do not obey the locality conditions on contextual allomorphy. The argument (which is summarized in Section 4) is that alternations of this type must be treated with MP, not SS. The argument just outlined is partial, in the sense that it can be made for certain stem alternations, but not for others; for the latter type, either the MP or SS analyses could in principle work. This important point is taken up in Section 4. Section 5 offers some general conjectures about the manner in which morphological and phonological factors relate to the locality conditions under which an alternation may take place, and Section 6 offers concluding remarks.

2.  Conditions on contextual allomorphy Contextual allomorphy is found when a single morpheme like T[+past] – a functional head – takes different forms depending on what is in its local environment.

route” models of morphology like Pinker and Prince (1988) and Pinker and Ullman (2002). Morphophonological theories are found (with internal differences) in early generative works like Chomsky (1957), Halle (1959), and Chomsky and Halle (1968), as well as in later works like Halle and Marantz (1993) and Embick and Halle (2005). In fact, these are only recent manifestations of the two positions; the debate between versions of these positions has been going on for a long time. See Kilbury (1976), Dressler (1985), the papers collected in Singh (1996), and Scheer (2010) for overviews and perspectives.



Contextual conditions on stem alternations 

In English, for example, T[+past] is realized as -Ø in the context of verbs like hit and sing (past tenses hit-Ø and sang-Ø); as -t in the context of e.g. bend and leave (past tenses ben-t and lef-t); and as the “default” -d elsewhere (as in play/play-ed, kiss/kiss-ed, and so on). In the version of Distributed Morphology (Halle & Marantz 1993) that is assumed here (Embick & Marantz 2008; Embick 2010) contextual allomorphy is analyzed with the operation of Vocabulary Insertion (VI), which applies to morphemes (terminal nodes). Competition between Vocabulary Items – typically, with ordering by specificity determining the winner – yields one winner and “blocking” of less-specified Vocabulary Items. So, for example, -t is inserted at T[+past] in the context of e.g. LEAVE ; as a result, the default -ed is not inserted: (2) Structure T υ √Root

T[+past] υ

(3) Vocabulary Items for Tense T[+past] ↔ -t/___ { LEAVE , BEND , …} T[+past] ↔ -Ø/___ { HIT , SING , …} T[+past] ↔ -d

Contextual allomorphy can occur only under certain locality conditions. According to the theory developed in Embick (2010), allomorphic interactions are constrained by the manner in which Vocabulary Insertion operates, and by the interaction of linear and cyclic locality conditions. Three different conditions are at the center of this theory. The first of these conditions enforces “inside out” cyclicity (e.g. Halle & Marantz 1993; Bobaljik 2000): (A1) Insertion proceeds from the inside-out. The ordering on insertion imposed by (A1) has consequences for the types of information (morphological or phonological) that may be referred to in Vocabulary Insertion; see below. A second condition on allomorphy advanced in Embick (2010) (see also references cited there) specifies a linear condition on contextual allomorphy: (A2) Contextual allomorphy requires concatenation (linear adjacency). Concatenation is represented with        ⁀ , such that X      ⁀ Y is read as “the terminal X is immediately left-adjacent to the terminal Y”; in these terms, (A2) holds that X may show contextual allomorphy determined by Y only when X      ⁀ Y (or Y      ⁀ X).

 David Embick

As a linear relation, concatenation (and therefore contextual allomorphy) can ignore intervening syntactic brackets. Certain nodes are invisible for the concatenation process. For example, in English past tense verb (2–3), the (phonologically unrealized) v head does not intervene between the Root and T[+past] (Embick 2003, 2010). This allows T[+past] to be conditioned contextually by certain Roots (and vice versa; see Section 5). Beyond (A1) and (A2), it appears that cyclic domains (phases) also impose constraints on when nodes may interact for allomorphic purposes: (A3) Two nodes can see each other for allomorphic purposes only when they are both active in the same cycle. For some views on how phase boundaries (Chomsky 2000, 2001) are relevant to morphology see Embick and Marantz (2008), Marantz (2007) and the implementation in Embick (2010). The main arguments of this paper are framed with respect to (A1,A2), (A1) in particular; the main results thus follow in any theory that incorporates this position. I include (A3) in this initial overview for completeness, and because ultimately the study of stem alternations must take into account cyclic domains as well. Taken together, (A1,A2) constrain possible allomorphic interactions in a way that can be illustrated in (4), which shows a complex head (4a) and its linearization as a Root with suffixes (4b):

(4) a. Complex head Z Z

Y X

√Root b. Linearization:

Y X

ROOT -X-Y-Z (= ROOT       ⁀ X, X      ⁀ Y , Y      ⁀ Z)

By (A1), VI occurs first at X, then at Y , then at Z. Thus, VI at X could be sensitive to either morphological or phonological features of the Root, but only to morphosyntactic features of Y; similarly, VI at Y could in principle see either phonological or morphosyntactic features of X but can look “outwards” only to morphosyntactic features of Z; and so on. In short, a node may show inward sensitivity to either morphosyntactic or phonological features, but it may show outward sensitivity only to morphosyntactic features, because outer nodes do not (by (A1)) have phonological content at that stage. An additional point is that by (A2) insertion at e.g. X could only be affected by ROOT or Y . The reason for this is that only the Root and Y are concatenated with X.



Contextual conditions on stem alternations 

2.1  Extension to stem alternation The MP and SS theories can be compared as follows. According to SS, stem alternation is analyzed as contextual allomorphy. In terms of the preceding subsection, this means that the distinct stems would have to be treated as separate Vocabulary Items competing for insertion; in essence, a ROOT would have different stems Stem1, Stem2, ..., as shown in (5):2 (5) ROOT ↔ STEM1/ 〈environment 1〉 ROOT ↔ STEM2/ 〈environment 2〉 

The important question with reference to hypothetical VIs like those in (5) is under what conditions the distinct stem alternants appear. By hypothesis, contextual allomorphy is subject to (A1,A2). Thus, if an alternation is conditioned by (i) a non-adjacent element; (ii) an “outer” node’s phonological properties; or (iii) a phonological property of the “word”, then it cannot be suppletive (i.e. it cannot be contextual allomorphy); rather, it has to be some sort of (morpho)phonological change. Before looking at the specifics, a more general note is in order. This paper assumes that there are at least some constraints on when stem allomorphy may be triggered. The alternative to this, which I refer to as “Anything Goes”, holds that stem alternations could be triggered by any feature – or any bundle of features – anywhere in the context of the stem, in a way that does not respect any type of locality. “Anything Goes” is clearly a worst-case scenario for this part of the interface. There is no reason to assume it is correct, and I will assume below that approaches that allow reference to arbitrary bundles of features à la “Anything Goes” should be rejected. 3.  Two alternations in Spanish verbs The alternations from Spanish examined in this section are restricted to a certain class of Roots, and, as such, are not part of the normal phonology in any obvious way. This is the type of phenomenon that looks in principle like it could be

2.  In most of the cases that are examined below, this means allomorphy for Roots. It therefore has to be assumed that Roots are subject to Vocabulary Insertion; see Embick (2000) for relevant discussion. One way to do this is by saying that a Root like e.g. SING could be realized by distinct phonological forms like sing, sang, and sung; see 3.1 below.   A conceptual argument against (5) is that it makes the relationship between stem alternants suppletive, so that sing/sang is represented in the same way as go/went; see e.g. Embick and Halle (2005). As mentioned earlier, I will put this type of objection to the side, and concentrate on the empirical predictions of theories that implement (5).

 David Embick

treated either with MP or with SS. The argument is that an SS treatment of these alternations does not comply with (A1,A2), because of the way in which they are conditioned by outer phonology. From this it follows that – in spite of the restriction to certain Roots or morphemes – stem storage must be rejected, in favor of a morphophonological approach. 3.1  Diphthongization The alternation between simple vowels and diphthongs in Spanish – referred to as diphthongization – is item-specific, in the sense that certain verbs with /o/ and /e/ stem vowels alternate (6a), while other verbs with the same vowels do not undergo the alternation (6b). The present indicative forms of two verbs are shown in (6c):3 (6) Diphthongization and listedness a. Diphthongization: pensar ‘think’, poder ‘be able to’, tender ‘hang’, sentar ‘sit’ b. No Diphthongization: tensar ‘tauten’, poner ‘put’, podar,‘prune’ rentar ‘yield, rent’ c. Present Indicative forms for pensar and tensar p/n

pensar

tensar

1s

pienso

tenso

2s

piensas

tensas

3s

piensa

tensa

1p

pensamos

tensamos

2p

pensáis

tensáis

3p

piensan

tensan

The fact that diphthongization is not found in all verbs with /e/ and /o/ vowels in the Root is responsible for the tension between MP and SS analyses. Harris (1969), for instance, develops an analysis of the former type, whereas e.g. Hooper (1976) (probably) argues for the latter. In the framework of Section 2, treating diphthongization with stem storage SS requires an analysis with stem allomorphs pens and piens of the Root PENS . A provisional analysis with competing stems is shown in (7), where ENV1 and ENV2 are abbreviations for the hypothetical contextual specifications for these two stem allomorphs: (7) PENS ↔ pens/—ENV1 PENS ↔ piens/—ENV2 3.  The alternation typically involves /e~ie/ and /o~ue/. According to the standard description, there are a few verbs with underlying /i/ that alternate, such as adquirir ‘acquire’, and maybe one verb with stem /u/ that diphthongizes (jugar ‘to play’).



Contextual conditions on stem alternations 

As noted above, Vocabulary Items like those in (7) are available in any theory that allows late insertion for (at least the phonology of) Roots. The crucial aspect of (7) is what determines whether one or the other VI is used; i.e. what must be specified in the ENVs in order for the correct distribution of stems to be derived. The standard view is that the alternation is conditioned by stress. As can be seen in (8), the diphthong occurs when the stem vowel is stressed, and otherwise the simple vowel is found ((8) departs from orthographic practice by marking the stress in all forms):4

(8) Forms of pensar ‘to think’ 1s

2s

3s

1p

2p

3p

pr. ind.

piénso

piénsas

piénsa

pensámos

pensáis

piénsan

pr. subj.

piénse

piénses

piénse

pensémos

penséis

piénsen

pret.

pensé

pensáste

pensó

pensámos

pensastéis

pensáron

impf.

pensába

pensábas

pensába

pensábamos

pensábais

pensában

Although only four tenses are shown in (8), the pattern according to which the diphthong occurs under stress is exceptionless in the verbal system. The fact that stress determines the distribution of alternating diphthongs in this way has direct consequences for the comparison of SS and MP theories. This is because sensitivity to stress along the lines seen in (8) requires information about stress placement that is not available when insertion at the Root node would have to take place in the SS theory. Concretely, the verb forms shown in (8) are realizations of the complex head structure (9), which consists of a v head, a TH(eme) node, a Tense node, and an AGR(eement) node (Oltra-Massuet 1999; Oltra-Massuet & Arregi 2005):

(9) Verbal structure T T υ υ

√Root

AGR T

TH υ

4.  For stress in the Spanish verb see Oltra-Massuet and Arregi (2005) and references cited there.

 David Embick

The choice between pens- versus piens- at the Root node requires reference to the position of stress in the entire word. The realization of stress in turn requires insertion at the outer nodes (v, TH, T, AGR). This scenario is ruled out by (A1), which holds that insertion at the Root node must precede insertion at outer nodes.5 This argument against SS relies on the idea that diphthongization is phonologically determined. For this reason, it is worth considering an alternative that employs the VIs in (7), but with contextual conditioning by morphosyntactic features, not phonology. In principle, nothing in (A1,A2) prevents outwardlooking contextual allomorphy, as long as it is conditioned by morphosyntactic (and not phonological) features on local nodes (see Embick 2010 for examples). And, if diphthongization could be treated morphosyntactically, it would not provide an argument against stem storage. In the case at hand, however, there is little motivation for a morphological treatment. Given only the present tense verb forms in (8), the non-diphthongized stem form could be restricted to first and second person plural environments (there are various ways in which this could be done). But, such an analysis fails to account for the broader generalization that alternating diphthongs occur under stress elsewhere in the language (in nouns, adjectives, etc.): (10) viéjo ‘old’, vejéz ‘age’ niéve ‘snow’, nevádo ‘snowy’ miél ‘honey’, melóso ‘like honey’ Venezuéla ‘Venezuela’, Venezoláno ‘Venezuelan’

The fact that the same phonological factor regulates the alternation in verbs, nouns, and adjectives points to the same conclusion: this alternation is phonologically determined.6 Different types of (morpho)phonological analyses of diphthongization could be given in the framework developed here. One factor that complicates the analysis 5.  In addition to this, there is no sense in which the choice would be determined by the properties of a morpheme concatenated with the Root, as required by (A2). 6.  There is a set of prima facie exceptions in which an alternating diphthong is found without being stressed on the surface. This is found with evaluative morphology like diminutives; e.g. viéjo ‘old (person)’, cp. vejéz ‘age’; but diminutive viej-ít-o ‘old (person)-DIM’ (see discussion and references in Halle et al. (1991)). It appears, however, that this exceptionality is part of a larger generalization about the status of (certain types of) diminutives; see Bachrach and Wagner (2006) for a morphophonological treatment of some related phenomena in Brazilian Portuguese, and for additional discussion of the syntax of such morphemes Wiltschko and Steriopolo (2007) and de Belder et al. (2009).   In addition, the behavior of diphthongization with different derivational morphemes is an important topic, but goes beyond the scope of this paper.



Contextual conditions on stem alternations 

of diphthongization is that, in addition to there being non-alternating simple vowels (recall that e.g. tensar does not diphthongize, while pensar does), there are also non-alternating diphthongs in the language: e.g. frecuénto ‘I frequent’, frecuentó ‘s/he frequented’; Viéna ‘Vienna’, vienés ‘Viennese’ (Harris 1985: 32). Thus, the Roots and morphemes that have alternating vowels have to be distinguished from the Roots and morphemes that do not. In principle, the relevant distinction could be made either phonologically (by positing distinct underlying representations for alternating and non-alternating segments) or morphologically (i.e. diacritically). For example, Harris (1985) represents alternating diphthongs as phono-­ logically special, with two timing slots, only the first of which is linked to a vowel. In this analysis, the empty position is associated with a vowel when it is in the rime of a stressed syllable, yielding a diphthong; if this association does not occur, a simple vowel surfaces (see also Inkelas et al. 1997). On the other hand, theories in which phonological rules can make reference to the identity of particular morphemes – as I assume to be possible here – make the other option available. The alternating morphemes can be diacritically specified to undergo diphthongization (or monophthongization, if it is assumed that the diphthong is underlying). After stress is calculated in the word, certain morphemes (like PENS ) are subject to diphthongization if stressed, with the rule (or rules) making reference to PENS or a diacritic it bears.7 The difference between the phonologically-special and morphological diacritic approaches connects with a larger debate between theories that appeal to phonological exceptionality (or prespecification) on the one hand, versus theories employing morpheme-specific phonology (morphological or lexical diacritics on rules) on the other. This question – and some related questions about the structure of the phonology – go beyond the scope of the present discussion.8

7.  With respect to the “morphological” analysis, it is important to note that the theory that I assume here still allows Roots like PENS to be visible as Roots when the stress of the whole word is calculated. In terms of the fleshed out version of (A3) of Section 2 (see Embick 2010), the verbs that have been examined to this point, which have the structure in (9), are contained within one cyclic domain. In other words, there is no “Bracket Erasure” (or equivalent) within (9); as a result, the Root still exists as a morphological object, and can be referred to as such, when the morphophonology reaches the outermost morpheme in (9). When the stress in the entire word is calculated, it is known whether e.g. PENS or TENS is present, and whether or not there is stress on the potentially alternating vowel. A diphthongization rule that has morphological conditioning can apply at that stage to produce the correct results. 8.  The phonological and morphological solutions in the text share certain properties, and could incorporate some common assumptions. For instance, as discussed by Halle et al. (1991) and others, in terms of a theory with cyclic versus non-cyclic phonological rules,

 David Embick

The important conclusion from this initial argument is that the distribution of stem alternants provides an argument against the SS theory. 3.2  “Raising” Spanish verbs of Conjugation III (the -ir conjugation) show an alternation that is often referred to as raising; this is because in diachronic terms it involves the raising of mid vowels. As will be shown below, the “raising” alternation is better viewed as the result of a lowering or dissimilation rule in the synchronic grammar, as originally proposed by Harris (1969); for consistency of reference, however, I retain the term raising verbs for this class. The raising phenomenon is seen in verbs like pedir ‘to ask’, which has e.g. 1s present pid-o, with stem /i/, but e.g. 1pl present pedimos, with stem /e/. Almost all of the verbs of Conjugation III that show an /e/ vowel in the infinitive alternate with /i/ in this way; (11) illustrates with further forms of pedir:9 (11) Forms of pedir 1s

2s

3s

1p

2p

3p

pr. ind.

pido

pides

pide

pedimos

pedís

piden

pr. subj.

pida

pidas

pida

pidamos

pidáis

pidan

pret.

pedí

pediste

pidió

pedimos

pedisteis

pidieron

impf.

pedía

pedías

pedía

pedíamos

pedíais

pedían

impf. subj.

pidiera

pidieras

pidiera

pidiéramos

pidierais

pidieran

fut

pediré

pedirás

pedirá

pediremos

pediréis

pedirán

cond

pediría

pedirías

pediría

pediríamos

pediríais

pedirían

­ iphthongization is part of the non-cyclic phonology (Harris 1989 argues for this point against d Halle and Vergnaud’s (1987) cyclic analysis of the rule). In terms of the model assumed here, one way to implement this is by saying that the rule(s) that result in diphthongization apply when the boundary of the entire word is reached; that is, at an M-Word boundary, in the sense of Embick and Noyer (2001).   These assumptions are important when further cases beyond the verbs are taken into consideration (diminutives, category-changing derivations, compounds), but I will not examine the matter further here. 9.  There are a few exceptions, i.e. verbs with stem /e/ in the infinitive but no /i/ forms: e.g. agredir ‘attack’, transgredir ‘transgress’, sumergir ‘submerge’ are listed in Malkiel 1966: 472; Harris 1969: 115 lists divergir ‘diverge’ and concernir ‘concern’ as well.   In addition to the /e/~/i/ alternation, there are a few verbs in which /o/ alternates with /u/ in the same way.



Contextual conditions on stem alternations 

There is no rule of the normal phonology that would raise /e/ to /i/.10 The apparent irregularity of the phonological process relating /e/-stems and /i/-stems (along with the fact that the alternation is restricted to verbs of Conjugation III) makes the alternation a prime candidate for stem storage.11 As can be seen in (11), the distribution of alternants is complicated. In terms of morphosyntactic features, the environments are as follows: (12) a.  ped: 1 and 2 pl present indicatives; non-third person preterites; all imperfects, futures, and conditionals. b.  pid: 1, 2, 3 sg, and 3 pl present indicatives; all the present subjunctives; all the imperfect subjunctives; 3s and 3p preterites.

All of the verbs in the raising class alternate in exactly the same way as pedir, with one further complication. A subset of the raising verbs (e.g. mentir ‘to lie’) also show diphthongization. With verbs of this latter type, diphthongs appear in exactly the expected forms (i.e. those where the stem vowel is stressed).12 The factors that determine the distribution of ped and pid in (11) do not appear to be morphosyntactic: there is no coherent set of tense, mood, or person/ number features that could be referred to in conditioning one of the alternants. Thus, if the distribution of stem alternants had to be stated in a way that did not refer to the phonology, the only conceivable treatment would be one in which the environments taking each stem form are simply enumerated; reference to bundles of features in this way amounts to “Anything Goes” (see Section 2). An analysis that makes reference to morphosyntactic features thus looks very unpromising. Moreover, it is unmotivated: just as with Diphthongization, there is a phonological generalization about the /e/~/i/ alternation. Building on ­Harris (1969), it can be treated as a Dissimilation process, in which underlying /i/ is lowered when the following syllable contains /i/. This is stated as Dissimilation in (13):13 (13) Dissimilation: i →e/—(C)i

10.  For some relevant historical discussion of raising processes in Spanish see Malkiel 1966. 11.  Harris (1969: 115) treats the alternation with a “minor rule” that is lexically restricted. Linares et al. (2006) use the exceptions of the kind noted in the text as evidence for the irregular nature of the alternation, in spite of the fact that there are very few verbs with unchanging stem /e/ in Conjugation III. For a developmental angle on these verbs see Mayol (2007). 12.  Thus mentir ‘to lie’ has three different surface alternants, as seen in e.g. 1s indicative miento, 1p indicative mentimos, 1p subjunctive mintamos. 13.  The /i/ that triggers Dissimilation has to be a nucleus; in e.g. 3s past pidió it is a glide.

 David Embick

A simple way of encoding which Roots are undergoers is by marking the Roots subject to Dissimilation rule diacritically (or by restricting the rule to Conjugation III verbs, and marking the non-undergoers as exceptions). In the same way that the analysis with underlying /e/ has some exceptions (see Footnote 9), the analysis with underlying /i/ and Dissimilation must make use of Root-specific information. In fact, the analysis with underlying /i/ has more Root-specific exceptions to Dissimilation than there would be to a Raising rule; e.g. vivir, vivo, vivimos; escribir, escribo, escribimos; etc. Putting these points together, a (morpho)phonological analysis of the Raising verbs is straight-forward, as long as it is acknowledged that the phonological process can be Root-specific. And, if the alternation is treated with a phonological trigger in this way, then it cannot be stem storage: by (A1) reference to outer phonological material is not possible in Vocabulary Insertion. In sum, the “Raising” verbs look like a typical candidate for stem storage (morpheme-specific alternation), but do not show the locality conditions that apply to suppletive contextual allomorphy. This alternation must be treated morphophonologically, not with stem storage. 4.  Interim summary The last section looks at two stem alternations in Spanish verbs. These are both prime candidates for stem storage because they implicate specific morphemes, or, more neutrally, something beyond the normal phonology. The argument is that these alternations cannot be treated with stem storage, because of the locality relations between the elements undergoing and triggering the change: that is, in each case, the stem alternation is triggered in a way that cannot be contextual allomorphy in a theory with (A1–3). As noted above, the argument makes particular use of (A1), because it involves sensitivity to outer phonological properties; (A2) may be implicated as well. Another way of stating the general conclusion is this: morphological or lexical conditioning does not require storage of alternants. Rather, at least some alternations that are morpheme-specific in the relevant way must be treated as part of the (morpho)phonology. This conclusion, though important, is partial; this is because other stem alternations in the informal sense are not subject to this line of argument. For example, a stem alternation that looks outwards to non-phonological information could still be treated with Stem Storage, as long as it respects (A2–3). Consider e.g. English long/length, strong/streng-th. Here, the stem change is (presumably) triggered by an outer morpheme, not its phonology. Locality-wise, i.e. by (A2–3), this could be treated as a case of contextual allomorphy, with e.g. long and



Contextual conditions on stem alternations 

leng as suppletive allomorphs; or, it could be treated morphophonologically. The locality-based argument does not apply. On the general point that morpheme-specificity does not necessarily require storage, the results of this paper agree strongly with Kiparsky (1996). Kiparsky rejects the idea that morphological conditioning is necessary or sufficient for determining the nature of an alternation, and argues that “There are both purely phonologically conditioned morpholexical alternations [=phonologically conditioned allomorphy; DE], and conversely, morphologically conditioned phonological rules” (1996: 16). The perspective that Kiparsky advances is quite similar to the one outlined above: in particular, the fact that an alternation is morphologically conditioned does not necessarily make it a case of stored allomorphs. Rather, the defining criteria that distinguish alternations with storage from those that are morphophonological have to do with “...the nature of the alternation, the locality relation between the focus and the triggering context, and the relationship of the process to other rules of the system” (1996: 17; my emphasis). A number of details about the analysis do not look the same in Kiparsky’s lexicalist model as they do in the syntactic approach assumed here, and there are other differences (with respect to e.g. productivity) as well; but the overall emphasis of Kiparsky’s argument, which concentrates on locality of conditioning, is exactly on target. 5.  A Question and a conjecture This paper argues against the idea that morphological conditioning forces an analysis with storage of alternants. However, in this section I will suggest further that while morphological conditioning does not determine the status of an alternation in terms of SS or MP, it is perhaps relevant to the locality conditions that apply to an alternation. After framing the question of whether phonological versus morphological conditioning involves distinct locality domains, I present a set of predictions that can be deployed to investigate this aspect of phonological form. A key factor in the arguments from Spanish verbs in Section 3 is that the stem alternations are triggered by outer phonological properties. This, for the reasons described in Section 2, is not possible for suppletive contextual allomorphy in a theory in which (A1) applies to Vocabulary Insertion. At the same time, there are other types of stem allomorphy (in the descriptive sense) which do not have a phonological trigger or target in this sense. The alternations found in the English past tense are of this type. Particular Roots like SING or BREAK undergo changes in the context of particular morphemes like T[+past], to yield alternations like sing/sang and break/broke. In my view, the observation to be emphasized is that

 David Embick

this type of alternation requires reference to two morphemes as morphemes, i.e. not as phonological objects. The effects of locality in this kind of alternation can be approached in a few steps; to begin with, I assume that the structure underlying the past tense forms is as follows: (14) Structure of past tense T

υ √Root

T[+past] υ

As discussed in Section 2, the v head does not intervene linearly between the Root and T[+past] when v is not realized phonologically.14 When Vocabulary Insertion at T[+past] occurs, this node is concatenated with the Root: ROOT       ⁀ T[+past]. This allows T[+past] to see the Root for its allomorphy (recall (3) above). Importantly, this means that all of the stem changes in English past tense (and participial) forms occur under concatenation. So, in the same way that T[+past] can see the concatenated Root for the purposes of its allomorphy, the Root is also concatenated with T[+past]. The derivation of sang is thus as follows:15 (15) derivation of sang a. Structure: [[ SING v] T[+past] ] b. Linearization/Pruning: SING       ⁀ T[+past] c. Vocabulary Insertion: SING    ⁀   T[+past,-Ø] d. Stem change: sing → sang/___    ⁀   T[+past,-Ø]

(The last line is shorthand for the process(es) changing the rime of SING in the context of T[+past]). With this view of English stem changing at hand, the question that can be posed is whether there is a general difference in kind between the following two types of alternations: (16) Types of Alternation a.  Type 1: Alternations that involve a combination of morphological and phonological information; i.e. in which either the target or the trigger is identified in only phonological terms.

14.  Embick (2010) discusses a Pruning operation that deletes null nodes in the relevant way. There are some other options for analyzing this “transparency effect” that appeal to more general properties of Root/category relations, but I will not dwell on them here. 15.  For concreteness (15) assumes that the stem changing rule occurs after VI at T[+past]. It can be assumed for convenience that Pruning of v occurs after VI inserts -Ø at that node.



Contextual conditions on stem alternations 

b.  Type 2: Alternations that refer only to morphemes qua morphemes; i.e. where both the trigger and target are specific morphemes (or Roots).

The reasoning is as follows. The concatenation condition (A2) on contextual allomorphy says that morphemes must be concatenated in order to be visible to each other. In the light of what is observed with the changes to stems of the sing/sang type, it is possible that (A2) is a specific manifestation of a more general principle that subsumes both contextual allomorphy and the Type 2 alternations in (16). This more general principle is stated as the conjecture in (17): (17)  Morpheme Interaction Conjecture (MIC): PF Interactions in which two morphemes are referred to as morphemes occur only under linear ­adjacency (concatenation).

The intuition behind the MIC is that the information type that is referred to in the structural description of an alternation determines the locality conditions under which the alternation takes place. When morphemes have to interact as morphemes, they must be concatenated in order to see each other. On the other hand, when phonological representations are referred to in an alternation, the locality conditions that apply are phonological in nature – i.e. need not respect concatenation of morphemes – but rather respect phonological locality. Investigating the MIC is at the heart of the research program that is advanced in this paper. If the MIC turns out to be correct, then there are two different kinds of rules responsible for stem alternations in the informal sense (i.e. the descriptive term stem alternation covers more than one grammatical phenomenon). One type (the one seen in the Spanish case studies) is truly morphophonological in nature, and a second type (illustrated with the English past tense) requires reference to two morphemes. The two types of alternation are defined in (18), which replaces (16): (18) Revised Rule Typology a.  Morphophonological Rules: Phonological rules in which either the trigger, or target, but not both, is morphological. Expected to occur under locality conditions characteristic of phonological representations; NOT (A2). b.  Morpheme/Morpheme Readjustments: Rules that change the form of

one morpheme when it is in the context of another morpheme, in which both the trigger and the target are referred to as particular morphemes. Expected to occur under locality conditions characteristic of morphological representations; i.e. under concatenation (A2).

If (18) is on the right track, many further questions can be posed concerning the ordering and interaction of different rule types (as envisioned by Kiparsky 1996).

 David Embick

Crucially, though, it remains to be seen first whether the MIC holds when additional case studies are taken into account.16 6.  Conclusions The main claims of this paper are centered on the tension between MP and SS analyses of stem alternation, and on what empirical (not conceptual) arguments can be advanced in favor of one view versus the other. A primary claim is that this tension cannot be resolved in the absence of a general theory of the locality conditions under which allomorphic changes take place. With this particular emphasis in mind, the analysis of stem alternations must be situated against the theory of locality that applies to the Vocabulary Insertion operation, which I assume to be based on (A1–3). The reason for this is that an SS theory with stored stems treats these stems as suppletive allomorphs; as such, the locality conditions on stem allomorphy are expected to be identical to those found for the contextual allomorphy found in Vocabulary Insertion more generally. The case studies from the Spanish conjugation that are examined in Section 3 illustrate a type of alternation that is a typical candidate for stem storage (because of Root- or morpheme-specificity). But the locality conditions under which these alternations take place are not compatible with (A1–3). Because these alternations do not obey the same locality conditions as (suppletive) contextual allomorphy, it is concluded that at least this type of stem alternation cannot be treated with stem storage. Instead, an analysis that makes use of (morpho)phonological rules operating on a single underlying form is required. The particular arguments of this paper are centered on (A1), and the idea that outer phonology cannot be referred to for contextual allomorphy at an inner node. The (A1)-based argument against SS can be formed for some instances of stem allomorphy – like those from S­ panish – but does not apply to stem allomorphy, for reasons outlined in Section 4. As a point for ongoing research, it is suggested in Section 5 that stem alternations in the descriptive sense might actually be two distinct phenomena in the grammar: a Morphophonological type, which operates in terms of phonologically-defined representations, and a Morpheme/Morpheme type, which operates in terms of the concatenation of morphemes. The idea that concatenation is required for all

16.  A further consequence of (18) is that it might force a reexamination of phonologically conditioned suppletive allomorphy (see e.g. Paster (2006)), where Vocabulary Insertion makes reference to phonological information.



Contextual conditions on stem alternations 

i­ nteractions in which two morphemes must see each other as morphemes is stated as the Morpheme Interaction Conjecture (MIC). As pointed out in the introduction, the analysis of stem alternations has played an important role in many theoretical models since the early twentieth century, and is also central to the more recent flurry of activity that has come to be known as the “past tense debate” (e.g. Marslen-Wilson & Tyler (1998)). Whatever is made of the full range of conclusions discussed in these domains, it is a striking fact that prior work on this topic devotes very little attention to the locality conditions that restrict allomorphic interactions. This paper shows why locality considerations must come to the front of this discussion, and provides a preliminary framework for further investigation of this part of the interface between structure and sound.

References Bachrach, Asaf & Michael Wagner. 2006. “Syntactically driven cyclicity vs. output-output correspondence: The case of adjunction in diminutive morphology,” paper presented at the Penn Linguistics Colloquium. Carstairs-McCarthy, Andrew. 1992. Current Morphology, Routledge, London. Chomsky, Noam. 1957. Syntactic Structures, Mouton, The Hague. Chomsky, Noam. 2000. “Minimalist Inquiries: The Framework”, in Roger Martin, David Michaels & Juan Uriagereka, eds, Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, MIT Press, 89–156. Chomsky, Noam. 2001. “Derivation by Phase,” in Michael Kenstowicz, ed., Ken Hale. A Life in Language, MIT Press, Cambridge, MA, 1–52. Chomsky, Noam & Morris Halle. 1968. The Sound Pattern of English, Harper and Row, New York. de Belder, Marijke, Noam Faust & Nicola Lampitelli. 2009. “On inflectional and derivational diminutives,” paper presented at NELS 40, Cambridge MA. Dressler, Wolfgang U. 1985. Morphonology: The dynamics of derivation, Karoma Publishers, Ann Arbor. Embick, David. 2000. “Features, Syntax and Categories in the Latin Perfect,” Linguistic Inquiry 31: 2, 185–230. Embick, David & Rolf Noyer. 2001. Movement operations after syntax. Linguistic Inquiry 32: 4, 555–595. Embick, David. 2003. “Locality, Listedness, and Morphological Identity,” Studia Linguistica 57: 3, 143–169. Embick, David. 2010. Localism versus Globalism in Morphology and Phonology, MIT Press, Cambridge, MA. Embick, David & Morris Halle. 2005. “On the status of stems in morphological theory,” in T. Geerts & H. Jacobs, eds, Proceedings of Going Romance 2003, John Benjamins, Amsterdam/Philadelphia, 59–88. Embick, David & Alec Marantz. 2008. “Architecture and Blocking,” Linguistic Inquiry 39: 1, 1–53. Halle, Morris. 1959. The Sound Pattern of Russian, Mouton, The Hague.

 David Embick Halle, Morris, James Harris & Jean-Roger Vergnaud. 1991. “A reexamination of the Stress ­Erasure Convention and Spanish Stress,” Linguistic Inquiry 22: 1, 141–159. Halle, Morris & Alec Marantz. 1993. “Distributed Morphology and the Pieces of Inflection,” in Kenneth Hale & Samuel Jay Keyser, eds, The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, MIT Press, Cambridge, MA, 111–176. Harris, James W. 1969. Spanish Phonology, MIT Press, Cambridge, MA. Harris, James W. 1985. “Spanish Diphthongization and Stress: A Paradox Resolved”. Phonology Yearbook 2, 31–45. Hooper, Joan. 1976. An Introduction to Natural Generative Phonology, Garland Academic Publishers. Inkelas, Sharon, C. Orhan Orgun & Cheryl Zoll. 1997. “The implications of lexical ­exceptions for the nature of grammar,” in Iggy Roca, ed., Constraints and derivations in phonology, Oxford University Press. Kilbury, James. 1976. The Development of Morphophonemic Th ­ eory, volume 10 of Amsterdam Studies in the Theory and History of Linguistic Science, John Benjamins. Kiparsky, Paul. 1996. “Allomorphy or Morphophonology?” in Rajendra Singh and Richard Desrochers, eds., Trubetzkoy’s Orphan, John Benjamins, Amsterdam/Philadelphia, 13–31. Malkiel, Yakov. 1966. “Diphthongization, Monophthongization, Metaphony: Studies in Their Interaction in the Paradigm of Old Spanish -IR Verbs,” Language 42: 2, 430–472. Marantz, Alec. 2007. “Phases and words,” in S. H. Choe et al. ed., Phases in the theory of grammar, Dong In Publisher, Seoul. Marslen-Wilson, William & Lorraine K. Tyler. 1998. “Rules, representations, and the English past tense,” Trends in Cognitive Sciences 2: 11, 428–435. Mayol, Laia. 2007. “Acquisition of Irregular Patterns in Spanish Morphology,” in Ville Nurmi & Dmitry Sustretov, eds, Proceedings of the Twelfth ESSLLI Student Session. Oltra-­Massuet, Isabel & Karlos Arregi. 2005. “Stress by Structure in Spanish,” Linguistic Inquiry 36: 1, 43–84. Oltra-Massuet, Maria Isabel. 1999. On the Notion of Theme Vowel: A New Approach to Catalan Verbal Morphology, Master’s Thesis, MIT. Paster, Mary. 2006. Phonological Conditions on Affixation, Doctoral dissertation, University of California at Berkeley. Pinker, Steven & Alan Prince. 1988. “On language and connectionism: analysis of a parallel distributed processing model of language acquisition,” Cognition 28, 73–193. Pinker, Steven & Michael Ullman. 2002. “The past and future of the past tense,” Trends in Cognitive Sciences 6: 11, 456–463. Scheer, Tobias. 2010. A guide to morphosyntax-phonology interface theories: How extra-­ phonological information is treated in phonology since Trubetzkoy’s Grenzsignale, Mouton de Gruyter, Berlin. Singh, Rajendra, ed. 1996. Trubetzkoy’s Orphan, John Benjamins, Amsterdam/Philadelphia. Wiltschko, Martina & Olga Steriopolo. 2007. “Parameters of Variation in the Syntax of Diminutives,” in Proceedings of the 2007 Annual Conference of the Canadian Linguistics Association.

State nouns are Kimian states* Antonio Fábregas & Rafael Marín

Universitetet i Tromsø / Université Lille 3 – CNRS (UMR 8163) The study of states in the verbal domain has recently been enriched with the distinction between K-states and D-states (Maienborn 2005; Rothmayr 2009). This new line of research has not been extended to state denoting nouns, which have been in general much less studied than those nouns denoting objects or events (Grimshaw 1990). This paper takes this task and shows that in Spanish noun-denoting states systematically behave like K-states, even when they are derived from D-state verbs. We further argue that only verbs which contain a structure with a K-state meaning can have corresponding state denoting nouns. The result is that, far from being idiosyncratic, it is possible to predict – given the aspectual properties of a verbal predicate – whether a state can be expressed in the nominal domain. Keywords:  states; K-state; D-state; nominalizations; deverbal nouns; aspect

1.  Preliminaries and background concepts: States and nouns The study of nouns derived from or related to verbs has mainly focused on nominal constructions denoting events (Chomsky 1970; Grimshaw 1990; Alexiadou 2001), and rarely on those that denote states (with exceptions; cf., for instance, Pesetsky 1995). This article tries to contribute to the understanding of these nominalizations by addressing three interrelated questions about their grammatical behaviour: a. What kind of states are those expressed by state nominalizations? Are state nominalizations Kimian states, Davidsonian states or both? *  Antonio Fábregas’ research is supported by grants Aurora 199749/V11 and DAAD 199852

from the Norwegian Research Council. Rafael Marín’s research is supported by grants FFI2009-07456 and FFI2010-1506 from the Spanish Ministry of Science and Education. We are grateful to the audience of Going Romance, three anonymous reviewers and the editors of the volume for invaluable comments to previous versions of this article. The usual disclaimers apply.

 Antonio Fábregas & Rafael Marín

b. Are the kinds of state denoted by a nominalization dependent on the properties of the base or do they follow from independent principles? c. What determines whether a state verb has a related state noun, derived or not? Using European Spanish data, we will argue that (a) all state nouns behave like Kimian states, (b) even when the verbal base is a Davidsonian state. We will further show that (c) not all Davidsonian-state-denoting verbs have associated state nouns: only those that contain a Kimian state as a part of their internal morphosyntactic structure do. Before we proceed, an important caveat is in order. The goal of this paper is to explore one of the possible distinctions that languages make inside the domain of stativity, but we are not assuming, or want the reader to assume, that the distinction between D-states and K-states exhaust the classes of statives available in grammar. There are other potentially relevant distinctions. The individual- /stage-level contrast (Arche 2006) would be one. Another one would be whether stativity is obtained through external aspect (like the progressive, Parsons 1990) or is defined at the aktionsart level, and, finally, those that emerge from different argument structures possibly associated to different kinds of states (see, for psychological predicates, Meinschäfer 2003) should also be considered in an exhaustive analysis that goes beyond the boundaries of this contribution. This is the reason why we will adopt (§ 1.1) a neutral definition of state that refers to one of their most salient semantic properties. The goal in doing so is to remain neutral and not to arbitrarily leave some stative predicates out. Further research on the matter is required to see how the K- vs. D-state division is integrated with these other classifications, and whether the one discussed in this paper should be viewed as a subclass of one of the other divisions or a superclass that includes them as subclasses. Keeping this in mind, let us begin with our discussion. 1.1  Preliminaries: A working definition of state As a working definition of state, we use a conceptual criterion, in line with Dowty (1979), Krifka (1989), Maienborn (2003, 2005) and Rothmayr (2009): states are homogeneous predicates which fulfill the subinterval property, namely that for any subinterval t’ – no matter how small- included in the temporal interval t during which a predicate holds it is also true that the predicate holds. Assuming this definition, among the nominalizations which unambiguously denote states in Spanish we find the following (1a). They are all related to verbs which have a reading which also satisfies the subinterval property, making them also stative verbs following the criterion adopted in this paper (1b). (1) a.  posesión ‘possession’, preocupación ‘preoccupation’, preferencia ­‘preference’, diversión ‘amusement’, entretenimiento ‘entertainment’, aburrimiento ‘boredom’, existencia ‘existence’



State nouns are Kimian states 

b.  poseer ‘possess’ preocupar ‘worry’, preferir ‘prefer’, divertir ‘amuse’, ­entretener ‘entertain’, aburrir ‘bore’, existir ‘exist’

Choosing this particular conceptual criterion to define states admittedly has the disadvantage that when a verb has several possible readings compatible with different truth conditions, the judgments of speakers might not be identical with respect to whether they fulfill the subinterval property, and therefore can be either considered as states, or not. Consider, for instance, the verb llover, ‘to rain’. Does it denote an activity, as suggested by an interpretation in which the verb denotes ‘water falls from the sky and hits the ground’? In that case, it does not fulfill the subinterval property, because at some t’ there would not be a drop of water hitting the ground. Or does it denote a state and the verb is interpreted as ‘there is water falling in the sky’? In that case, at all t’ there would be some drop of water on its way down to the ground. This problem must be acknowledged, but in this paper, we have refrained from using examples that, like this one, are not clearly associated to one conceptual representation in the intuitions of speakers. Finally, we need to make explicit some tests that determine that a noun is stative. States can have a temporal extension but are non dynamic and do not have natural boundaries. If we take a noun like preocupación, (2), which is unambiguously stative, and we compare it with a noun like operación, (3), which is unambiguously eventive, we see how these properties get reflected in the behaviour of stative nouns: these nouns accept PP modifiers unambiguously denoting temporal extension (2a) – just like event nouns (3a), but unlike these nouns, cannot be the subject of the predicate tener lugar ‘take place’, which requires dynamic entities (compare 2b and 3b). Finally, these nouns reject the plural, unless the noun stops denoting a state and is interpreted as an object (2c), while at least some event nouns allow this number form (3c).1 (2) a. una preocupación de varios meses a preoccupation of several months ‘a several-month-long preoccupation’

1.  Of course, state-denoting nouns are not the only ones that reject plurals. Famously, deverbal event nouns coming from activities also tend to reject it (i). Even if plural marking is not enough to diagnose that a noun denotes a state, combining this test with the others presented is enough to differentiate state nouns from deverbal activity nouns. See ­Alexiadou, Iordachioaia and Soare (2011) for the complex issue of how plural is interpreted in ­nominalizations. (i) {el aprendizaje / *los aprendizajes} de lenguas de Borges     the learning.sg     the.pl learning.pl of languages of Borges ‘Borges’ {learning / *learnings} of languages’

 Antonio Fábregas & Rafael Marín

b. *Una preocupación tuvo lugar entonces.  a preoccupation took place then *‘A preoccupation took place then.’ c. #preocupaciones  worries ‘things that worry someone’ (3) a. una operación de varias horas an operation of several hours ‘a several-hour-long operation’ b. Una operación tuvo lugar ayer. an operation took place yesterday ‘An operation took place yesterday.’ c. operaciones operations ‘actions of operating on someone’

In this paper we consider state denoting nouns those nouns that have the properties presented in (2).2 1.2  Kimian states and Davidsonian states The recent literature has refined the notion of state, proposing a crucial division between two classes of non-dynamic eventualities (Maienborn 2003, 2005; ­Rothmayr 2009). The first class, Davidsonian states (henceforth D-states), are pseudo-stative verbs which relate to event verbs in allowing place and manner modification; this class includes verbs like lie, sit, wait, glow, gleam, bubble or sleep. The second class (Kimian states; K-states for short) are those state verbs that reject such modifiers, like own, resemble, weigh, cost or copular constructions; they denote the instantiation of a set of properties in a given individual during a time span. In other words, they do not denote an event of any kind, and therefore they reject all kinds of event modifiers. Empirical differences follow from this. D-states, having an event variable, allow (a) a locative modifier to locate this variable; (b) a manner modifier to qualify this variable and (c) an underspecified quantifier to measure the temporal extension of this variable. Let us show each one of these differences in turn. The first empirical difference between D-states and K-states is that the latter never allows a place modifier to locate the eventuality.

2.  Cf. Fábregas et al. (2012) for a more detailed discussion on this issue.



State nouns are Kimian states 

(4) a. Luis yacía junto a la ventana.(Spanish) Luis lied next to the window ‘Luis was lying next to the window.’ 

[see also Maienborn 2005: (27b)]

b. *Luis pesa 50 kilos junto a la ventana.(Spanish)  Luis weighs 50 kilos next to the window. *‘Luis weighed 50 kilos next to the window.’  [see also Maienborn 2005: (26a)]

When a place modifier is combined with a K-state, it can be interpreted as a frame adverbial which restricts the domain where a particular proposition is true, as in (5). (5) Juan pesa tres kilos en Plutón.(Spanish) Juan weighs three kilos in Pluto ‘Juan weighs three kilos in Pluto.’

(5) means that in Pluto, it is true that Juan weighs three kilos, while in the earth this is probably false. Secondly, K-states never allow manner adverbials, but D-states do. (6) a. Luis yacía tranquilamente en la cama. Luis lied calmly on the bed ‘Luis was calmly lying in bed.’  [see also Maienborn 2005: (31b)] b. *Juan posee ahorradoramente mucho dinero.  Juan owns thriftily much money *‘Juan thriftily owns much money.’  [see also Maienborn 2005: (30b)]

Notice, however, that there is the question of whether some of these manner modifiers are out because they presuppose that there must be an agent able to control that manner. It is crucial, therefore, to use manner adverbs referred to speed or other notions that can modify an eventuality without making reference to the role of a possible agent. (8) shows that even with this restriction, Maienborn’s characterization is right. (8) Juan poseía (*lentamente) mucho dinero. Juan owned (slowly) much money Intended ‘Juan, little by little, owned a lot of money.’

Finally, K-states never allow the temporal reading of adverbial expressions ambiguous between a time and a degree reading, like German ein bisschen (‘a little’) or its Spanish equivalent un poco. This is admitedly the most problematic of the three criteria. One important property of this test is that the quantifier

 Antonio Fábregas & Rafael Marín

itself cannot lexically denote a temporal extension, which is fulfilled by un poco but not by for three days.3 (9) a. Juan yació en la cama un poco. Juan lied in the bed a bit ‘Juan lied in bed for a short while.’  [see also Maienborn 2003: (37b)] b. #Carol se parecía un poco a su abuela.  Carol SE resembled a bit to her grandmother ‘Carol slightly resembled her grandmother.’  [see also Maienborn 2005: (39)]

With this background in mind, we now turn to state denoting nouns in order to see whether the distinction between D-states and K-states is also relevant in the nominal domain or not. 2.  Nouns coming from K-state verbs Let us start with state nouns derived from or related to K-state verbs, as those in (10): (10) a.  posesión ‘possession’, coste ‘cost’, peso ‘weight’, pertenencia ‘belonging’, existencia ‘existence’ b.  admiración ‘admiration’, preocupación ‘preoccupation’, preferencia ‘preference’, suposición ‘assumption’, aburrimiento ‘boredom’, satisfacción ‘satisfaction’

3.  Note that when modifiers that lexically encode a temporal extension (minute, year, month…) are used, the difference between K-states and D-states disappears: (i)

a. John owned this house only for three years. b. Mary resembled her grandmother only during the first years of her ­childhood.

This is perhaps not so surprising if we assume that in (i) the temporal meaning is obtained conceptually and is not dependent on the presence of a temporal variable inside the structure of the predicate. To the extent that these entities denote states, and to the extent that states are different from qualities in allowing for a temporal extension, the concepts they express should be conceptualisable as having a temporal extension. Notice that nouns that do not express events might also allow for these temporal modifiers when the temporal meaning is encoded lexically (un novio de tres años ‘a boyfriend of three years, a boyfriend who has lasted three years’). A different question is whether a temporal interpretation of the modifier can emerge in the absence of lexical cues that encode temporality. Here, and only here, is where the previous test identifies a distinction between two classes of states.



State nouns are Kimian states 

(10a) includes the Spanish counterparts of German and English K-states discussed in Maienborn (2005). In (10b) we find nouns related to some psychological verbs that behave as K-states when built as statives (Rothmayr 2009).4 We will show that, as expected, all these nouns behave like K-states. They act as K-states with respect to place and manner modification, and the temporal extension reading of underspecified quantifiers. Remember that passing these tests is taken as a sign that there is an event variable inside the modified element (cf. § 1.2). 2.1  Incompatibility with place modifiers First of all, state nouns always reject place modifiers that locate the eventuality. (11) a. su posesión / peso (*bajo la ventana) his possession / weight (*under the window) ‘his possession / weight *(under the window)’ b. su preocupación / preferencia (*bajo la cama) his preoccupation / preference (*under the bed) ‘his preoccupation / preference (*under the bed)’

This is not due to some incompatibility between locative modifiers and nominalizations. Note that when the nominalization denotes an event, the PP modifiers are generally licensed, to the extent that the original verb allows them (12). (12) a. su encuentro (junto al puente) their encounter (next to-the bridge) ‘their encounter by the bridge’ b. su canto (bajo la ventana) his singing (under the window) ‘his singing below the window’ c. su operación (en el quirófano dos) his operation (in the operating-room #2) ‘his operation in operating room 2’ 4.  Two anonymous reviewers direct our attention to an important (and complex) issue: psychological verbs do not form one single natural class with respect to their aspect ­ (cf. ­Meinschäfer 2003; Marín & McNally 2011). Here we concentrate only on their stative reading, and among those that allow for such reading, those that have been previously classified as K-states. This, however, should not prevent us from noticing and mentioning some complications that still are unresolved in the study of psych-verbs. The first one is whether it is possible to find eventive nominalizations from psych-verbs at all; this has been argued, for instance, by Meinschäfer (2003), and, if true, it would mean that the internal structure of some of these predicates is more complex than has been assumed. Secondly, if the event nominalization is available for at least some of them, the problem of why psych-nominalizations tend to denote states becomes more acute (see Fábregas & Marín 2012 for a suggestion).

 Antonio Fábregas & Rafael Marín

Apparent counterexamples to this generalization can be shown not to be place modifiers locating the eventuality. Some of them, like the one in (13), are frame adverbials that restrict the truth of the proposition. (13) is interpreted as ‘In this galaxy, given its biological rules, the presence of water is associated to the presence of life’. (13) la asociación entre la vida y la presencia de agua the association between the life and the presence of water en esta galaxia in this galaxy ‘the association between life and the presence of water in this galaxy’

In other cases, what looks like a place modifier is actually a modifier that presents a particular situation. This is what happens with the cases in (14), where the PPs do not express a place, but actually conditional situations that trigger different perspectives of what nature is: in one case, as a scientific matter and on the other, as an object of entertainment. This is so because schools and parks are interpreted not only as places, but also as domains where people act in a particular way. (14) a. el aburrimiento de Juan con la naturaleza en la escuela the boredom of Juan with the nature in the school ‘the boredom of Juan with nature at school’ b. la satisfacción de Juan con la naturaleza en el parque the satisfaction of Juan with the nature in the park ‘the satisfaction of Juan with nature in the park’

When we use PP modifiers that unambiguously introduce pure locations, not situations, the noun phrases become ungrammatical (15). (15) a. el aburrimiento de Juan con la naturaleza (*en una silla) the boredom of Juan with the nature    on a chair *‘the boredom of Juan with nature on a chair’ b. la satisfacción de Juan con la naturaleza (*bajo un árbol) the satisfaction of Juan with the nature     under a tree *‘the satisfaction of Juan with nature under a tree’

The final set of apparent counterexamples refers to phrases where the locative is actually placing the object, not the eventuality (16). (16) una breve interrupción de internet en este barrio a short interruption of internet in this neighbourhood ‘a short interruption of internet in this neighbourhood’



State nouns are Kimian states 

It is the internet which is being located in this reading; this can be shown because (16) means the same as (17), where the locative is unambiguously introduced as a modifier of the subject noun. (17) El internet de este barrio está interrumpido. the internet of this neighbourhood is interrupted ‘Internet is interrupted in this neighbourhood.’

2.2  Incompatibility with manner denoting adjectives Secondly, state nouns reject adjectives that express manners. Given that in general states lack agents, it is not surprising that adjectives such as those in (18a) are rejected by them, because they express behaviours that require an agent. However, in principle, nothing prevents a state from combining with adjectives like those in (18b), which do not require conscious control from the subject, or those in (18c), which denote properties of the internal organization of the objects in a situation. (18) a. generous, careful, furious b. fast, slow, quick, accidental, good, silent c. elegant, scruffy, orderly, disorderly, upright, rigid

Even when we restrict ourselves to the adjectives in (18b) and (18c), state denoting nouns reject them: (19) a. su (*silenciosa) posesión / pertenencia her    silent possession / belonging *‘her silent {possession / belonging}’ b. su (*elegante) aburrimiento / preocupación his (elegant) boredom / preoccupation *‘his elegant {boredom / preoccupation}’

Again, this is not a general property of nouns. When the noun denotes an event, the same adjectives are allowed in the same pre-nominal position (20). We assume that this reading is possible because the derived still allows the event variable of its base available for modifiers. (20) a. su elegante interpretación her elegant performance ‘her elegant performance’ b. su buena aplicación de la ley his good application of the law ‘her good application of law’

 Antonio Fábregas & Rafael Marín

c. su silencioso llanto his silent crying ‘his silent weeping’

There are some apparent counterexamples, like those in (21), but it can be shown that, in those, the adjective denotes properties of the holder of the state, not of the way in which the eventuality is performed. (21) a. su elegante presencia his elegant presence ‘he was present and he was silent’ b. su inquieto aburrimiento his restless boredom ‘he was bored and he was restless’

In contrast with (20a), where what is elegant is the performance itself – while perhaps the actors themselves were acting in a vulgar way, according to the script –, (21a) requires compulsorily that the person present is elegant, and is thus incompatible with a scenario where the person is behaving in a vulgar way. 2.3  Unavailability of temporal readings with ambiguous adjectives Finally, when combined with an adjective which is ambiguous between a time and a degree reading, state nouns systematically reject the temporal interpretation. Here we test it with the adjective reducido ‘reduced’, which in general allows the two interpretations – or even favours a temporal one – when combined with event denoting nouns (22). (22) a. una representación reducida de Hamlet a performance reduced of Hamet ‘a short performance of Hamlet’ b. un ataque reducido contra los enemigos an attack reduced against the enemies ‘a short attack against the enemies’

When combined with state nouns, however, only a degree reading emerges: (23) a. un coste / peso reducido a cost / weight reduced ‘a small cost / weight’ b. una preocupación reducida a preoccupation reduced ‘a slight preoccupation’

Therefore, we conclude that nouns derived from K-state verbs behave, as expected, as K-states. In the next section, we are going to show that nouns derived from D-states, quite surprisingly, also behave as K-states.



State nouns are Kimian states

3.  From D-state verbs to nouns 3.1  From D-state verbs to K-state nouns Among Spanish D-state verbs that have a corresponding state denoting noun we find those included in (24a), with their corresponding nouns in (24b). The list is short, as these cases are not frequent. (24) a.  brillar ‘shine’, esperar ‘wait’, silbar ‘whistle’, lucir ‘shine’, relucir ‘shine’, resplandecer ‘glow’, chirriar ‘creak’ b.  brillo ‘shining’, espera ‘waiting’, silbido ‘whistling’, luz5 ‘light’, glow’, ­resplandor ‘glowing’, chirrido ‘creaking’

As we are going to show in this section, brillo-type nouns (24b) behave as K-states, although related to D-state verbs. 3.1.1  Asymmetries with place modifiers The verb brillar ‘to shine’ is a D-state, which allows for locative modifiers that place the eventuality; however, its related noun rejects this modifier (25).6 (25) a. La lámpara brilló un momento frente a la casa. the lamp shined a moment in.front of the house ‘The lamp shined for one moment in fron of the house.’ b. su brillo de un momento (*frente a la casa) its shine of a moment (in.front of the house) ‘its momentary shine in front of the house’

5.  This noun is ambiguous between a state reading and a physical object reading. In the state reading, it rejects the plural and accepts temporal modifiers (i). In some languages the two readings are differentiated by gender. In Catalan, llum ‘light’ is masculine in the object reading and feminine in the reading that rejects plural and combines with temporal modifiers. (i) a. una reducida luz de pocos segundos a reduced light of few seconds ‘a brief shine of a few seconds’ b. *unas reducidas luces de pocos segundos  some reduced lights of few seconds *‘some brief shinings of a few seconds’ 6.  A phrase like (i) is grammatical, but notice that the place modifier locates the object, not the eventuality: the person has to be in the park for the phrase to express a true situation. (i) la espera de Juan en el parque the wait of Juan in the park ‘Juan’s waiting in the park’



 Antonio Fábregas & Rafael Marín

3.1.2  Asymmetries with manner modifiers The following contrast shows that even when the state verb allows for manner modifiers, its related noun rejects the equivalent manner adjective. The contrast is illustrated in (26) with the verb asociar in its stative reading ‘to be associated’ and in (27) with the verb brillar, ‘to shine’. .

(26) a. Estas dos ideas se vinculan elegantemente en su teoría. these two ideas SE associate elegantly in her theory ‘These two ideas elegantly associate in her theory.’ b. su (*elegante) vinculación en su teoría their (elegant) association in her theory *‘their elegant association in her theory’ (27) a. La lámpara brilló accidentalmente. the lamp shined accidentally ‘The lamp accidentally shined.’ b. su (*accidental) brillo its     accidental shine *‘its accidental shine’

3.1.3  Asymmetries with temporal readings of modifiers Finally, state verbs that allow the temporal reading of un poco ‘a little’ disallow the same interpretation of the adjective reducido ‘reduced’. (28) a. La lámpara brilla un poco. the lamp shines a bit Two readings: The lamp shines for a short time. The lamp shines with a bit of intensity. b. su reducido brillo its reduced shine One reading: The lamp shines with little intensity. (29) a. La tetera silbó un poco. the teapot whistled a bit Two readings: It whistled for a short time. It whistled with a bit of intensity. b. el reducido silbido the reduced whistling One reading: It whistled with a bit of intensity.

Therefore, we conclude that state nouns behave as K-states even when the corresponding verbs are D-states.



State nouns are Kimian states 

3.2  Two classes of D-state verbs In addition to what we have seen in § 3.1, there are other D-state verbs – actually, most of them – that lack a state nominalization. Either they do not have any corresponding noun (and have to use the infinitive in nominal contexts; cf. 31 below) or the noun denotes a physical object. We call these verbs without a corresponding state noun Stubborn D-states, because they refuse to become K-states. They typically belong to one of the classes represented in (30): (30) a. Predicates of position and posture  yacer ‘lie’, rodear ‘suround’, estar sentado ‘be sitting’, estar erguido ‘be standing’, estar tumbado ‘be lying’ b. Predicates of emission of liquids manar ‘flow’, fluir ‘flow’ c. ‘Manner’ of staying predicates colgar ‘hang’, flotar ‘float’, pender ‘hang’

These verbs lack any kind of nominalization to express the corresponding state. If they have to be expressed in a nominal context, an infinitive – which can syntactically act as a noun – has to be used (31).7 (31) el suave fluir del río the soft flow.inf of-the river ‘the quiet flow of the river’

In contrast, as we have already seen, there are other D-state verbs that have a corresponding noun denoting a similar eventuality; we call them Flexible D-states because they can become K-states in the nominal domain. Leaving aside esperar ‘wait’, they are verbs of light and sound emission: (32)  brillar ‘shine’, chirriar ‘creak’, pitar ‘whistle’, resplandecer ‘shine’, silbar ‘whistle’, lucir ‘glow’, relucir ‘glow’

7.  There are several reasons to doubt that infinitives are real nominalizations. As happens also with English verbal gerunds (Chomsky 1970), their behaviour shows that they keep big parts of their internal verbal structure. They almost never get demotivated meanings, and frequently allow internal arguments in accusative case. Moreover, to the best of our knowledge, there are no lexical exceptions, that is, we do not know of any verbs that cannot produce an infinitive used in a nominal context. All this suggests that, even if they had to be considered nouns at a lexical level, they are built over external functional layers where there is no access to the Aktionsart or argument structure of the verb. As this kind of  ‘nominal’ constituent does not interact with aspect, we will leave it outside our discussion.

 Antonio Fábregas & Rafael Marín

These verbs have related nouns also denoting a state (remember the tests in 2): un pitido de varios segundos ‘a whistling of several seconds’. Thus, we conclude that D-state verbs can be divided in two classes: one which is always a D-state, and therefore can never appear as a noun, and another one which allows for more flexibility and can appear as a K-state in the nominal domain. In the next section we will propose that the difference comes from the fact that these two classes of state verbs are different in their internal structure.

4.  Analysis: Some D-states contain a K-state Our analysis can be summarized like this. Given what we saw in § 2 and § 3, nouns denote K-states (as claimed independently by Maienborn 2003). Thus, for a verb to be able to have a related noun denoting a state, it needs to be able to express a K-state. This is not possible for all verbs: it crucially depends on the condition that the internal structure of the verb must contain a projection whose denotation is a K-state. This is the projection that is used to build a state denoting noun in the grammar; when it is absent, there is no state noun. Stubborn D-states are verbs which lack this projection in their internal structure, while Flexible D-states are verbs which contain that structure. 4.1  Flexible D-states contain a K-state: Semantic evidence The first thing that we need to show is that Flexible D-states contain a K-state. ­Following Maienborn, a K-state is the instantiation of a property – in the restricted sense – in an individual during some time-span. Thus, we expect that Stubborn D-states will show that they do not contain this kind of component, while ­Flexible K-states will show evidence of it. Here we will follow Kennedy and McNally (2005). These authors analyze gradable properties (P) as relations between individuals (x) and degrees (d) (cf. 33). (33) λdλxλP[P(d)(x)]

Degrees are associated to properties in this restricted sense, but this does not mean that all properties will have a degree variable (some adjectives, like perfect or complete, do not have it in the same sense as fat or tall do, because they are not scalar). What we expect is that Stubborn D-states will not show a degree variable, because they do not contain a K-state, while some Flexible D-states, those coming from gradable properties, will contain it. To be clear: our analysis predicts that no Stubborn D-state will have a degree variable, but that some F ­ lexible D-states will. With this in mind, consider the reading of quantifiers with our



State nouns are Kimian states 

verbs. Bosque and Masullo (1998) notice that the meaning of adverbial quantifiers depends on the ingredients denoted by the verb they modify. Importantly for our purposes, they differentiate between an I-reading (intensity reading), in which the adverb modifies the degree of the property denoted by the verb (34a) and a variety of E-readings, which do not require a degree variable inside the denotation of the verb, and among other interpretations can result in a temporal extension reading (34b). (34) a. Juan trabaja mucho. Juan works a lot ‘Juan puts a lot of effort in his work.’ b. Juan trabaja mucho. Juan works a lot ‘Juan works for long periods of time.’

Under the light of this contrast, consider the class of Stubborn D-states (35) vs. the class of Flexible D-states (36). Systematically, the examples in (35) – which never have a related state denoting noun – reject the I-reading of the adverb (and in some cases, such as 35c and 35d, even reject the adverbial modification), while those in (36) allow this degree reading. (35) a. Juan yació mucho. Juan lied a lot ‘Juan lied for a long time.’ b. Juan estuvo tumbado mucho. Juan was lying a lot ‘Juan was lying for a long time.’ c. ??El río fluyó mucho.     the river flowed a lot ??‘The river flowed a lot.’ d. ??La fuente manó mucho.    the fountain flowed a lot ??‘The fountain flowed a lot.’ e. %El cuadro colgó mucho de la pared.     the picture hanged a lot from the wall ‘The picture hanged on the wall a long time.’ (36) a. La lámpara brilló mucho. the lamp shined a lot ‘The lamp shined with much intensity.’ b. El fuego resplandeció mucho. the fire glowed a lot ‘The fire glowed with much intensity.’

 Antonio Fábregas & Rafael Marín

c. La tetera silbó mucho. the tea-pot whistled a lot ‘The tea-pot whistled with much intensity.’ d. Esta máquina pitó mucho. this machine whistled a lot ‘This machine whistled with much intensity.’

Evidence that the reading in (35) is temporal while the same adverb allows a degree reading in (36) comes from contrasts like (37). As with Stubborn D-states the degree reading is not available, adding a modifier that states that the length of the state was short triggers a contradiction with these verbs (37a); the contradiction never arises with Flexible D-states, because they allow the degree reading of the adverb (37b). (37) a. #Juan estuvo sentado mucho, pero por poco tiempo.  Juan was sitting a lot, but for short time ‘Juan was sitting a lot, but for a short while.’ b. La lámpara brilló mucho, pero por poco tiempo. the lamp shined a lot, but for a short time ‘The lamp shined a lot, but for a short while.’

We propose that this distinction – the presence vs. the absence of a gradable property equivalent to a K-state – is what determines whether the same root will be able to materialize as a noun. Additional evidence for this is that, with other aspectual classes of verbs, those that have a state noun generally allow for I-­readings of quantity adverbials, showing that they contain a K-state with a degree variable in their structure. (38a) and (38b) illustrate this for verbs which, to begin with, are K-states according to the tests; (38c) and (38d) illustrate this for verbs that denote changes of state. The corresponding state nouns, derived or not, are presented in (39). (38) a. Juan quiere mucho a María. Juan loves much acc Mary ‘Juan loves Mary a lot.’ b. Juan se preocupa mucho. Juan se worries much ‘Juan is much worried.’ c. El accidente interrumpió mucho el servicio. the accident interrupted a lot the service ‘The accident greatly interrupted the service.’ d. La tubería se rompió mucho. the pipe se broke a lot ‘The pipe got severely broken.’



State nouns are Kimian states 

(39) a. la querencia de Juan por María the liking of Juan for María ‘Juan’s liking María’ b. la preocupación de Juan por la economía the preoccupation of Juan for the economy ‘Juan’s preoccupation about the economy’ c. la interrupción del servicio durante varias horas the interruption of-the service for several hours ‘the service’s being interrupted for several hours’ d. la rotura de la tubería durante varias horas the breaking of the pipe for several hours ‘the pipe’s being broken for several hours’

The different denotation of a Stubborn D-state versus a Flexible D-state explains the different behaviour of the adverbs presented in the previous section. (40) is the denotation of a D-state according to Maienborn (2005). More precisely, given the distinction we have introduced, we claim that this is the denotation of a Stubborn D-state, that is, a D-state which does not contain a K-state. No property and thus no degree variable is present. (40) λPλeλx[P(e) ∧ HOLDS(e, x)]

Thus, the denotation of yacer ‘to lie’ would be the one in (41), after existential ­closure of the event variable. (41) ∃eλx[lie(e) ∧ HOLDS(e, x)]

We predict that the I-reading will never emerge. In contrast, (42) is the denotation of a K-state according to Maienborn (2005): the instantiation z, bound to times and worlds, of a property P applied to an individual x. If that property is understood as a relation between degrees and individuals (following Kennedy & McNally’s proposal), we obtain the formula in (43). (42) λxλPλz[z ≈ [P(x)]] (43) λdλxλPλz[z ≈ [P(x)(d)]]

As not all properties are gradable, not all K-states will contain this degree variable, but this element can be present only when there is a property.8 For those whose property is conceived as gradable, the degree variable will remain. The denotation 8.  One clear example of a non-gradable K-state is own. In such case, the degree variable would simply not be present in the denotation of the property. It might not be accidental that own is defined by a relation between a possessor and a possessee, which is not gradable either when instantiated in other grammatical categories.

 Antonio Fábregas & Rafael Marín

of a Flexible D-state like brillar ‘to glow’ will contain this kind of structure inside its denotation. The formula in (44) is obtained over the formula in (40), adding to it the denotation of a K-state. A Flexible D-state, therefore, denotes the event of an individual x instantiating a property in a degree (d). (44) λeλxλd∃z[glow(e) ∧ HOLDS(e, x) ∧ z ≈ [glow(d)(x)]]

This D-state contains a degree variable, which comes with the property denoted by the K-state, and as such it can be bound by a quantifier, as shown in (45). This explains the availability of the I-reading with Flexible D-states. (45) λPλx∃d [much (d) ∧ P(d)(x)] (λeλxλd∃z[glow(e) ∧ HOLDS(e, x) ∧ z ≈ [glow(d)(x)]])

In short, our proposal is that Stubborn D-states denote states which cannot be further decomposed in semantic terms; in contrast, the state denoted by a ­Flexible D-state should be decomposed, crucially singling out a K-state that carries a degree variable. Before moving to the next section we would like to address directly a complication in the present analysis, which has also been pointed out to us by an anonymous reviewer. As not all properties are gradable, the test that we have developed here only goes one way: when a predicate is gradable, it contains a property; when it is not, we cannot know if it contains a property or not, and there are obvious cases of K-states that do not allow gradability (like own). This is a serious issue, which connects with the independent problem of what gradability is and how is it codified in the grammar. We are not in a position to give a convincing answer at this point, but at least the direction that the analysis has to take to address these problems is relatively clear. The ultimate issue that ­underlies this problem seems to us to be what determines that a property is gradable, and that can be reduced (as proposed in Kennedy & McNally 2005) to independent properties of scalar structure (or lack thereof). One suggestive possibility is that non-gradable K-states are built over categories that do not provide the relevant primitives for scale structures, like some central-coincidence prepositions (Hale & Keyser 2002) expressing sharp properties (‘inside’ or ­‘not-inside’). These objects would not be gradable in the scalar way (much, very, a bit…), as no scale underlies them, but the property expressed by them might still be subject to some modification which might be understood as non-­scalar gradation, as with completely or partially (John {completely / partially} owns this land), which can refer to the extension of the overlap between two notions. In any case, this proposal must be developed and in the meanwhile such imperfections are a problem for the analysis.



State nouns are Kimian states 

4.2  Matching the semantics with the internal projections of the verb Our next step is to show that the previous semantic analysis can actually explain that Stubborn D-states do not have state nouns by pairing these semantic interpretations with projections inside the internal structure of the verb. The first useful piece of data is that the denotation of different classes of nouns corresponds to the notions that we have discussed in the previous section. Next to the well-known class of nouns that denote events (46a) – that is, dynamic eventualities that occupy temporal extensions –, we have the class of state nouns studied in this paper (46b). As we have shown, these state nouns denote K-states, that is the instantiation, during a time span, of a property in an individual. Finally, we also have another major kind of nouns, those that denote simple properties (46c); notice that these nouns can also be deverbal, even if they are more frequently deadjectival. (46) a. destruc-ción destroy-nom ‘destruction’ b. preocupa-ción worry-nom ‘preoccupation’ c. modera-ción moderate-nom ‘temperance’

There is a certain inclusion relationship between the denotation of these three nominalizations. The class in (46b) denotes the same as the class in (46c), with the addition of the temporal extension; the class in (46a) can denote the same as the class in (46b), only adding information about a dynamic activity. We propose, in fact, that this inclusion relation provides us with a syntactic hierarchy, where semantic ingredients are added by individual heads inside a functional sequence. We will argue that an event nominalization contains a verbal structure whose highest head is ProcP (defining dynamicity and being responsible for introducing the event variable which can be bound by time, place and manner modifiers in an event nominalization). A state nominalization crucially lacks this head, and the highest verbal projection present is StateP, which lacks the event variable. Finally, a quality nominalization does not have any verbal projection, which explains that it cannot be connected to temporal extension of any kind. We propose that the temporal extension is made possible by the presence of verbal projections. Thus, the difference between quality nominalizations such as those in (46c) and the other two classes is reflected syntactically in the presence

 Antonio Fábregas & Rafael Marín

vs. absence of heads belonging to the verbal domain. The distinction between the event noun in (46a) and the state noun in (46b) is due to the kind of verbal projections present: the dynamic part is introduced by a verbal head denoting a process or activity (ProcP) and the state part is introduced by a non dynamic head denoting precisely a state (StateP). The inclusion relation established between the three kinds of nominalization suggests that the projection that denotes a quality can be selected by the projection that denotes a state, adding to the property the temporal extension. The possibility that dynamic eventualities contain a state (as opposed to the impossibility that states contain a dynamic eventuality as a part of their denotation) suggests again that the state projection can be selected by the Process projection. The resulting hierarchy is given in (47), paired with its semantic interpretation in (48). (47) [ProcessP [StateP [PredicationP]]] (48) [P(e)(x) ∧ [z ≈ [P(d)(x)]]]

The hierarchy in (47) is reminiscent of Ramchand’s (2008) proposal that verbs can be decomposed in at least two projections: one dynamic, corresponding to ProcessP, and another one of stative denotation, which she labels ResultP when it is selected by ProcessP. We follow the spirit of her proposal, as in her account ResultP is a head that denotes a state. Our change is that, in her account but not in ours, the interface interprets the head-complement syntactic relation between Proc and Res as a logical relation of cause-effect, with the consequence that the state is always interpreted as a result state (whence the label). We want to leave this semantic relation underspecified at LF, so that the state can be interpreted as the result of the process (as in become quiet) or as a state coextensive with the process (as in remain quiet). This will be dependent on the kind of event denoted by ProcessP; if it is a D-state, there will not be any transition and the state will simply be interpreted as a state where the argument is kept while the process lasts. As for the projection that expresses gradable properties without any temporal variable, notice that it cannot be paired with an adjectival phrase. If it did so, we would predict, counterfactually, that all verbs containing a degree variable will be deadjectival. This does not seem to be true. For instance, the verb brillar ‘to shine’ is related to brillo ‘shine’, which is a noun, although a noun which expresses a quality rather than a physical individual. Its main characteristic is that it must denote a property of individuals; this makes it equivalent to the Predication Phrase (Bowers 1993), the projection that allows a potential predicate to become a predication by getting a subject, and therefore we label the projection in this way. This captures the intuition that brillo, even as a noun, is ‘relational’ in the sense that it does not denote a physical object, but a quality of physical objects.



State nouns are Kimian states 

This hierarchy can be paired with the three kinds of nominalizations presented in (46): ProcessP is used to build an event nominalization (like destruction); StateP is used to build a state noun (like preoccupation) and PredicationP is used to build a quality noun (like moderation; cf. Roy 2010 for independent evidence that quality nouns are built using this projection). (49) Event noun: [n[ProcP… [√]]]9 State noun: [n[StateP [PredP [√]]] Quality noun: [n[PredP [√]]]

Given the syntactic hierarchy presented in (47), with the denotations in (48), the reason why Stubborn D-states cannot have state nouns is clear. The state noun is built using StateP, but Stubborn D-states do not contain StateP. We know this because StateP builds its denotation over the Predication phrase, and Stubborn D-states do not have a degree variable. The internal structure of a Stubborn D-state is the one in (50): simply ProcessP, without StateP and without PredicationP. (50) [ProcP [√lie]]

In contrast, Flexible D-states contain a StateP and a PredicationP, because they contain a degree variable. Their internal structure is the one in (51), and the state noun is built removing ProcP and using the remaining constituent headed by State. (51) [ProcP [StateP [PredP [√shine]]]]

Verbs which are already K-states, like preocupar ‘to worry’, are projections of StateP without ProcP; this is consistent with Ramchand’s (2008) analysis of psychological verbs as projections of a stative head; psychological verbs are analyzed as consistently K-states in Rothmayr (2009). (52) [StateP [PredP [√worry]]]

Finally, dynamic verbs that have a state noun, like romper ‘to break’ are projections of a ProcP that contains a StateP. This is consistent with the analysis of these verbs as changes of state, which is standard (cf. Pustejovsky 1991, among many others). In our account, these verbs are different from Flexible D-states in the semantics associated to ProcP, which in this cases will be a transition, and not the holding of a property. (53) [ProcP [StateP [PredP [√break]]]]

9.  We remain neutral as to whether InitP or any equivalent projection introducing agents is present or not in event nominalizations, as this is orthogonal to our purposes.

 Antonio Fábregas & Rafael Marín

Our account shows, therefore, that the classes of nouns that a verb can denote are dependent on how their internal structure is articulated.10 5.  Conclusions and extensions This paper has explored the grammatical behaviour of state denoting nouns, trying to determine what kind of states they denote, and the properties that their corresponding verb must have in order to produce a state noun. We have shown that state nouns always denote K-states. This is a confirmation of Maienborn’s (2003) main empirical contribution to the analysis of states across categories: nouns (and adjectives) are consistently K-states. As a consequence, when a verb denotes a D-state and lacks the projections that are required to denote a K-state, it is not able to have a corresponding noun. The reason is that its semantic type cannot be manifested as a noun, because that requires that the verb to be able to denote a K-state. Additionally, we have shown that the kind of noun that a verb can have associated to it (event, state, quality…) depends on the semantic and syntactic ingredients that compose the verb’s internal structure. We have argued that a decomposition that takes seriously the semantics denoted by the verb can provide us with an explanation of why some verbs lack certain types of nominalization. Specifically, we have argued that only those verbs that contain a degree variable can have a state noun, something which follows on the assumption that K-states are built over properties that relate degrees and individuals.

10.  A fair question, posed to us by one anonymous reviewer, is how this structure maps with the morphophonological exponents. Here we are assuming a system with cumulative exponence – where single exponents can spell out more than one head- and where syncretism is expressed by relations of containment between sequences of heads (cf. Caha 2009, the ­Superset Principle). The idea is that the exponent that we call ‘the verbal stem’ covers a number of heads inside the verbal structure (as Ramchand 2008 also assumes), but, crucially, in some cases it starts lexicalising in PredP. In other words, in the event reading of the verb preocupar ‘to worry’, the stem preocupa- lexicalises [ProcP[StateP [PredP]]]; in the state reading of the same verb, where the tree corresponds to [StateP [PredP]] dominating the same root, the same exponent is used because Spanish happens to lack a more specific lexical item used to spell out preocupar in the absence of ProcP. The stem modera- has even more flexibility. It can lexicalise a root under [InitP[ProcP[StateP[PredP]]]] in the causative reading of moderar ‘moderate’, but also the same root when in the syntax it just projects as [PredP]. Which exponents are used for each structure are, as usual, idiosyncratic of each language, so we expect to find languages where the quality noun of a verb is expressed with a different stem.



State nouns are Kimian states 

There is one point in our analysis that we have not addressed: what prevents a Stubborn D-state to have an ‘event’ nominalization? As these verbs contain a ProcP, which is the base to build an event nominalization, we would expect them to have related nouns. Although the problem deserves further study, we would like to suggest here that having ProcP is a necessary but not sufficient condition to have a related event noun. Presumably the denotation of ProcP, and more specifically whether it denotes a truly dynamic eventuality or not, will also play a role. If event nouns must denote a dynamic eventuality, then Stubborn D-states would be excluded from the nominal domain, as their ProcP does not denote this notion and they also lack StateP and PredP. However, exploring in depth this aspect is left for further research.

References Alexiadou, Artemis. 2001. Functional Structure in Nominals. Nominalization and ergativity. Amsterdam & Philadelphia: John Benjamins. Alexiadou, Artemis, Giannina Iordachioaia & Elena Soare. 2011. “Plural marking in argumentsupporting nominalizations”. Layers of aspect ed. by Patricia Cabredo Hofherr & Brenda Laca, 1–22. Stanford: CSLI. Arche, María. 2006. Individuals in Time. Tense, Aspect and the individual/stage distinction. Amsterdam & Philadelphia: John Benjamins. Bosque, Ignacio & Pascual J. Masullo. 1998. “On verbal quantification in Spanish”. Studies on the Syntax of Central Romance Languages ed. by Olga Fullana & Francesc Roca, 9–63. Girona: Universitat de Girona. Bowers, John. 1993. “The syntax of predication”. Linguistic Inquiry 24, 591–656. Caha, Pavel. 2009. The nanosyntax of case. Ph.D. dissertation, University of Tromsø. Chomsky, Noam. 1970. “Remarks on nominalization”. Readings in English Transformational Grammar ed. by R.A. Jacobs & P.S. Rosenbaum, 232–286. Waltham: Ginn and Co. Dowty, David R. 1979. Word Meaning and Montague Grammar. Dordrecht: Reider. Fábregas, Antonio & Rafael Marín. 2012. “The role of Aktionsart in deverbal nouns: State nominalizations across languages”. Journal of Linguistics 48, 35–70. Fábregas, Antonio, Rafael Marín & Louise McNally. 2012. “From Psych Verbs to Nouns”. ­Telicity and change of state in natural languages: implications for event structure ed. by Violeta Demonte & Louise McNally, 162–185. Oxford: Oxford University Press. Grimshaw, Jane. 1990. Argument Structure. Cambridge (Mass.): MIT Press. Hale, Kenneth & Samuel J. Keyser. 2002. Prolegomenon to a Theory of Argument Structure. ­Cambridge (Mass.): MIT Press. Kennedy, Chris & Louise McNally. 2005. “Scale structure, degree modification and the s­ emantics of gradable predicates”. Language 81, 345–381. Krifka, Manfred. 1989. Nominalreferenz und Zeitkonstitution: Zur Semantik von Massentermen, Pluraltermen und Aktionsarten. München: Wilhelm Fink. Maienborn, Claudia. 2003. Die logische Form von Kopula-Sätzen. Berlin: Akademie-Verlag. Maienborn, Claudia. 2005. “On the limits of the Davidsonian approach: The case of copula ­sentences”. Theoretical linguistics 31, 275–316.

 Antonio Fábregas & Rafael Marín Marín & McNally. 2011. “Inchoativity, change of state, and telicity: Evidence from Spanish reflexive psychological verbs”. Natural Language and Linguistic Theory 29, 467–502. Meinschäfer, Judith. 2003. “Nominalizations of French psychological verbs”. Selected Papers from Going Romance ed. by J. Quer, J. Schroten, M. Scorretti, P. Sleeman & E. Verheugd, 231–246. Amsterdam & Philadelphia: John Benjamins. Parsons, Terence. 1990. Events in the Semantics of English. A Study in Subatomic Semantics. Cambridge (Mass.): MIT Press. Pesetsky, David. 1995. Zero syntax: Experiencers and Cascades. Cambridge (Mass.): MIT Press. Pustejovsky, James. 1991. “The syntax of event structure”. Cognition 41, 47–81. Ramchand, Gillian. 2008. First Phase Syntax. Cambridge: Cambridge University Press. Rothmayr, Antonia. 2009. The structure of stative verbs. Amsterdam & Philadelphia: John Benjamins. Roy, Isabelle. 2010. “Deadjectival nominalizations and the structure of the adjective”. The Syntax of Nominalizations across Languages and Frameworks, ed. by A. Alexiadou & M. Rathert, 129–159. The Hague: De Gruyter Mouton.

‘I know the answer’ A Perfect State in Capeverdean* Fernanda Pratas In Capeverdean, there is a puzzling temporal reading for some occurrences of sabe ‘know’, as opposed to all eventive and some stative sentences: N sabe risposta has a present reading, ‘I know the answer’, whereas N kume pexe and N kridita na Nhor Des have past readings, ‘I ate the fish’ and ‘I believed in God’, respectively. The proposal in Pratas (2010) accounts for this puzzle in the following way: (i) all these predicates denote past eventualities, as an effect of a zero operator that adds a termination to atelic and a completion to telic situations; (ii) the particular property of N sabe risposta ‘I know the answer’ lies in its complex structure: it includes a past culmination, ‘I got to know the answer’, but its temporal reading is anchored on a consequent state (Moens & Steedman 1988), ‘[now] I know’. That previous proposal, however, does not provide an explanation for the nonexistence of analogous temporal readings for other situations. The present paper tackles this problem, putting forward an analysis based on Perfect State theories: (i) the zero morpheme is a null Perfect marker; (ii) only for predicates like sabe risposta ‘know the answer’ may a Perfect State be the direct result of the past eventuality, as defined in Smith (1991); I argue that this result state is part of the event structure; (iii) for other predicates, the Perfect State is merely an abstract state of the event’s having occurred (Parsons 1990; ter Meulen 1995); therefore, it does not participate in the event structure and, as such, it is not part of the situation temporal schema. Keywords:  Capeverdean; statives; tense; event structure; null Perfect

*  I am very grateful to my Capeverdean consultants from Santiago Island, especially Ana Josefa Cardoso and José António Brito. I also thank the audience of Going Romance 2010 and three anonymous reviewers, for their truly enriching comments and suggestions, and Alan Munn and Cristina Schmitt, for their precious help with the English editing. All the errors and problems are, however, my own responsibility. Research for this paper has been funded by FCT, through the project Events and Subevents in Capeverdean (PTDC/ CLE-LIN/103334/2008).

 Fernanda Pratas

1.  Introduction In Capeverdean, a Portuguese-based Creole language1, the lexical property of stativity does not account for the different temporal readings of predicates. This fact constitutes one further challenge to the Language Bioprogram Hypothesis (LPH, Bickerton 1981, 1984). According to that hypothesis, Creole languages have a relative tense system, in which stativity plays a crucial role. The relevant relation for the current purposes is the following: unmarked stative verbs have a non-past reading, unmarked eventive verbs have a past reading.2 Interestingly, if we compare the temporal readings of (1a) and (1b) we might conclude that Bickerton (1981, 1984) is right: the first is a bare state and has a present reading, whereas the second is a bare eventive and has a past reading. (1) a. N sabe risposta. 1sg know answer3 ‘I know the answer.’ b. N kume pexe. 1sg eat fish ‘I ate the fish.’

The problem with this generalization, however, is that other stative p ­ redicates consistently pattern with eventives regarding the interactions with ­temporal morphology and interpretation: their bare forms cannot have a present reading. For reasons of space, this paper focuses on kridita ‘believe’ and lenbra ­‘remember’, as examples of verbs that may occur in stative situations and whose bare forms can never be interpreted as present. A list of such verbs is presented in (2), but others, which have not yet been studied, will possibly be included in future descriptions:

1.  Cape Verde is a former Portuguese colony off the West African coast. 2.  Bickerton’s generalizations regarding Creole languages have been more popular among scholars outside linguistics than among linguists. In Veenstra (2008), several linguists’ ­objections to Bickerton’s views are presented. Furthermore, one recent work challenging the default values of Creoles’ TMA system is van de Vate (2011), on Saamáka. 3.  List of abbreviations used in this paper: 1sg/1pl – 1st person singular/plural; acc – ­accusative; comp – complementizer; cond – conditional; fut – future; hab – habitual; neg – negation; pft – perfect; poss – possessive; prep – preposition; pres – present; prog – progressive; pst – past; quant – quantifier; tma – temporal morpheme (this is used for preverbal ta, which has a complex modal function); top – topic.



‘I know the answer’ 

(2)  ama ‘love’, atxa ‘find’ (have an opinion), divinha ‘estimate’ / ‘conjecture’, gosta di ‘like’, konsigi ‘be able to’, kridita ‘believe’, lenbra ‘remember’, ntende ‘understand’, obi ‘hear’, odia ‘hate’, odja ‘see’, pensa ma ‘think that’, spera ‘hope’ / ‘wait’, txera ‘smell’.

For all of them, a present interpretation requires the preverbal morpheme ta.4 This is illustrated in (3), for ‘believe in God’. (3) a. N kridita na Nhor Des (duranti 5 anu). 1sg believe prep sir god    for 5 year ‘I believed in God (for 5 years).’ [I was a believer for 5 years] *‘I believe in God.’

[past] [present]

b. N ta kridita na Nhor Des. 1sg tma believe prep sir god ‘I believe in God.’ [I am a believer]

[present]

Pratas (2010) proposed that these predicates exhibit an eventive-like behavior in this respect. This means that, just like the eventives, as illustrated in (1b), their allegedly bare forms are marked by a zero operator available in the language that adds a termination to atelic situations and a completion to telic situations. Processes are atelic situations; they are inherently unbounded. Culminated processes and culminations are telic situations; they are intrinsically bounded (they have a natural endpoint).5 The past reading in (1b) and in (3a) is an effect of this zero operator. As for the present reading in (1a), N sabe risposta ‘I know the answer’, the explanation is that it has a complex event structure: a subevent of the type BECOME (Dowty 1979) plus its consequent state (Moens & Steedman 1988). The culmination, marked by the same zero operator, has the meaning ‘[now] I got to know’ (where the reference time coincides with the speech time). For the consequent state, assuming that states are true of instants of time (Taylor 1977), we have the meaning ‘[now] I know.’ Note that there is an important distinction between this proposal and the one in Gehrke and Grillo (2009) for the English ‘know’. These authors have argued that a subevent of the type BECOME is added to a stative situation. The proposal for sabe risposta ‘know the answer’ in Pratas (2010) focuses first on the c­ ulmination.

4.  No specific gloss is provided for ta here; the complexity of this morpheme is described in subsection 2.1.1. 5.  The aspectual classes terminology used in the present paper is the one in Moens (1987) and not the one in Vendler (1957). Thus, we have processes, culminated processes and culminations, rather than activities, accomplishments and achievements.

 Fernanda Pratas

When there is some overt information pointing to a past reading, such as the relevant temporal expressions (onti ‘yesterday’, simana pasadu ‘last week’, etc.), the temporal anchor on the consequent state is canceled and we get a past eventuality (4a). When this past reading is not assured by such information, the temporal interpretation is associated with the consequent state (4b). (4) a. Onti, N sabe risposta. yesterday 1sg know answer ‘Yesterday, I got to know the answer.’ b. N sabe risposta. 1sg know answer ‘I know the answer (now).’

This proposal in Pratas (2010), however, leaves a crucial question unresolved: why is this type of complex event structure not available for other predicates? More specifically, why is it that other past culminated processes, such as kume pexe ‘eat the fish’, or culminations, such as txiga sedu ‘arrive early’, do not have consequent states of the same type? In fact, the likely consequent states in these cases are, respectively, ‘[now] I have the fish eaten’ and ‘[now] I am here early’. The goal of the present paper is threefold. First, it presents the Capeverdean data that challenge Bickerton’s (1981, 1984) generalization (Section 2). Briefly, it will be shown that in this Creole language there are stative predicates whose present interpretation requires the preverbal morpheme ta; this means that nonpast cannot be obtained through their unmarked forms. Although the verbs under analysis, just like any other from the list in (2), may also enter eventive constructions, this study focuses on their stative occurrences, and the stative properties of these occurrences will also be shown. Secondly, this paper argues that the complex structure of N sabe risposta ‘I know the answer’ is better accounted for if we assume that the zero operator (Pratas 2010) is a null Perfect (Section 3).6 This proposal can be summarized as follows: all allegedly bare forms of lexical verbs7 in root clauses are marked by a null Perfect morpheme. Perfect sentences denote a state located at reference time, which is due to the prior occurrence of a closed situation (Smith 1991: 147). I argue that only for sabe risposta ‘know the answer’ may a Perfect State be the

6.  A null Perfect analysis has been proposed in van de Vate (2011) to explain the difference in temporal interpretation between stative (present) and non-stative (past) bare verbs in Saamáka. In Section 2.3, some differences between the two proposals are described. 7.  Copula verbs and modals have a different behaviour. This will briefly be shown in the present paper for the copulas e be.individual-level and sta be.stage-level.



‘I know the answer’ 

direct result of the past eventuality (Smith 1991). This explains why this Perfect State is part of the event structure. When the Perfect State is merely an abstract state resultant of the event’s occurrence (Parsons 1990; ter Meulen 1995), this type of structure does not take place. Finally, the present paper contends that the nonexistence of such complex event structures for predicates containing other verbs is not problematic at all, since some idiosyncrasies of ‘know’ have been attested for other languages; for instance, semantic restrictions involving ‘know’ and the Perfect have also been found in Korean (Choi 2010) (Section 4). In Section 5, some final remarks will be presented. 2.  Some Capeverdean statives need ta for a non-past reading This section aims at presenting the Capeverdean predicates that challenge Bickerton’s (1981, 1984) generalization (2.1) and the arguments in favor of the stative nature of these predicates (2.2). 2.1  The data that resist the stativity explanation The Capeverdean data presented in this subsection challenge Bickerton’s (1981, 1984) generalizations on the temporal systems across Creole languages (2.1.2). Before this final purpose, however, Subsection 2.1.1 briefly describes the contributions of the overt temporal morphemes in this Portuguese-based Creole. 2.1.1  Overt temporal morphemes Capeverdean postverbal morpheme -ba marks past. This is illustrated in the following contrast. (5) a. N sata kume pexe. 1sg prog eat fish ‘I am eating (the) fish.’ [answer to: what are you doing now?] b. N sata kumeba pexe. 1sg prog eat:pst fish ‘I was eating (the) fish.’ [answer to: what were you doing at 1pm?]

The temporal interpretation in (5b) can only be accounted for if -ba, which does not occur in (5a), brings a past meaning.8

8.  For a more detailed argumentation in favor of the Past nature of -ba, see Pratas (2010). There are two other postverbal morphemes, -du and -da, that occur in passives. Since passive constructions are beyond the scope of the present paper, I will not discuss these

 Fernanda Pratas

The other temporal morphemes are sata, which marks progressive,9 also illustrated in (5), and ta, which has a complex modal/quantificational function. Both of them are preverbal. Several important observations regarding ta must be listed here. First, ta may enter future and conditional constructions with different situation types. This is illustrated in (6), with a process predicate, and in (7), with a stage-level state. (6) a. N ta nada dumingu ki ta ben. 1sg tma swim Sunday that tma come ‘I will swim next Sunday.’ b. Si N podeba, N ta nadaba dumingu. if 1sg can:pst 1sg tma swim.pst Sunday ‘If I could, I would go swim on Sunday.’ (7) a. N ta sta na kaza des ora. 1sg tma be.stage-level prep house ten hour ‘I will be at home by ten o’clock.’ b. Si N podeba, N ta staba na kaza des ora. if 1sg can:pst 1sg tma be:pst prep house ten hour ‘If I could, I would be at home by ten o’clock.’ Secondly, it may express habituality, a reading here illustrated with an eventive: (8) a. N ta kume pexe. 1sg tma eat fish ‘I eat fish.’

[everyday, sometimes, habitually]

b. N ta kumeba pexe. 1sg tma eat.pst fish ‘I used to eat fish.’

[everyday, sometimes, habitually]

morphemes here. For a description of Capeverdean passives, see Pratas (2007) and Rendall (in preparation). 9.  In Praia, the capital city of the country, there is another form for progressives: auxiliary sta + ta, like in (i): (i) a. Djon sta ta papia ku bo. Djon be tma talk with 2sg ‘Djon is talking to you.’ b. Djon staba ta papia ku bo. Djon be:pst tma talk with 2sg ‘Djon was talking to you.’ The differences between the two constructions have been extensively described in Baptista (2002) and Pratas (2007).



‘I know the answer’ 

Both interpretations in (8) denote a type of generalization that acquires properties typical of individual-level states, in the present (8a) and in the past (8b). Generics and habituals have been identified as ‘general statives’ (Smith 2003). Therefore, they have an unbounded interpretation and, assuming that states are true of instants of time (Taylor 1977), the combination with a point-like time reference naturally emerges. We can now assert that the eventuality kume pexe ‘eat fish’ can only be located in the present if: (i) it is expressed as ongoing (such as in (5a): N sata kume pexe ‘I am eating (the) fish’), or (ii) it is modified by ta (8a). In both cases, we have derived states: I am assuming here that (i) “the progressive operator turns sentences into stative sentences” (Vlach 1981: 284), and that (ii) habituals are general statives (Smith 1991, 2003). Finally, as we have seen in (3b) and as will be discussed in the next subsection, with lexical verbs in basic-level stative situations, ta may mark present (nonhabitual/generic) or future. As is certainly expected at this point, a key observation regarding ta is that it is incompatible with the progressive.10 The next subsection describes some Capeverdean data that defy Bickerton’s (1981, 1984) generalizations on the temporal system of Creoles; more specifically, it will be shown that some bare statives cannot have a present interpretation. 2.1.2  Why stativity is not enough Bickerton’s (1981, 1984) generalizations concerning the temporal system in Creole languages may be summarized as follows: (9) a. Non-overtly marked stative forms have a present reading b. Non-overtly marked non-stative forms have a past reading

The following Capeverdean examples seem indeed to support this proposal. First, the bare copula verb in (10) denotes a stative situation and is present. 10.  For more details on the meanings of ta, and also on the several Capeverdean environments that denote some sort of stative situations (thus, they are licit in the Present), see Pratas (2010). Also, in Pratas (2011a,b), it has been proposed that ta seems to involve quantification over possible worlds: – present habitual: if nothing prevents it, ‘I eat fish (habitually)’ (universal quantification) – future: if nothing prevents it, ‘I will eat fish (tomorrow)’ (existential quantification) This modal contribution of ta, however, is still under study; more specifically, all the p ­ roblems raised by a parallel between genericity/habituality and universal quantification must be considered and resolved.

 Fernanda Pratas

(10) Djon e altu. Djon be.individual-level tall ‘Djon is tall.’

Moreover, the examples in (1), here repeated in (11), show an important distinction: (11a) has a present reading, whereas (11b) has a past reading. Both are lexical verbs, but the first one participates in a stative situation and the second participates in an eventive situation. (11) a. N sabe risposta. 1sg know answer ‘I know the answer.’ b. N kume pexe. 1sg eat fish ‘I ate (the) fish.’

A distinction in the temporal readings of sabe ‘know’ and kume ‘eat’, illustrated in (11) for their bare forms, has counterparts in other simple declarative sentences, marked by the overt morphemes described in 2.1.1. Observe the examples with ta (12): (12) a. N ta sabe risposta. 1sg fut/ *pres know answer ‘I will know the answer.’ *‘I know the answer.’

[future] [present]

b. N ta kume pexe. 1sg fut / hab eat fish ‘I will eat (the) fish.’ ‘I eat fish.’ [everyday, sometimes, habitually]

[future] [present]

When marked by ta in a simple declarative sentence, sabe risposta ‘know the answer’ can only have a future interpretation. In contrast, kume pexe ‘eat (the) fish’ may be present or future. Consider the following context: Maria is leaving the room and utters any of these sentences; we understand that she is going to take an action leading to get to know the answer / eat the fish. Now imagine that we are just having a conversation with Maria and she declares any of these sentences: depending on the discourse context, (12b) may either have a present or a future reading, as indicated in the English translations; as for (12a), there is no discourse context allowing for a present reading. Now observe the examples with -ba (13). (13) a. N sabeba risposta. 1sg know:pst answer ‘I knew/used to know the answer.’



‘I know the answer’ 

b. N kumeba pexe. 1sg eat:pst fish *‘I ate/used to eat fish.’

The only possible meaning for (13b) is a past before past, ‘I had eaten (the) fish’. We will return to this context in Section 3, when the null Perfect proposal is explained. Cross-linguistically, this is an anaphoric reading and, for this reason, it is odd without a context: we need a reference time prior to the speech time, the situation being prior to that reference time. For reasons of space, I will not provide here all the contexts where each reading might occur. For now, the important revelation in (13) is that sabeba (‘know:pst’) may have a simple past reading (in the sense that the situation time is prior to speech time), whereas kumeba (‘eat:pst’) may not. As has been said before, the particular contrast between the bare forms in (10) and (11a), on the one hand, and in (11b), on the other hand, could be taken to support Bickerton’s (1981, 1984) generalization regarding the temporal systems across Creole languages. The former are stative predicates and have a present reading, the latter is an eventive and has a past reading. Problems arise when we observe other stative situations. As has been noted in various works on Capeverdean (Silva 1985; Suzuki 1994; Baptista 2002; Pratas 2007; Borik & Pratas 2008), the unmarked forms of these predicates cannot have a present reading: their temporal interpretation is necessarily past. (14) a. N lenbra tenpu di nha mosindadi. 1sg remember time prep poss.1sg youth ‘I remembered the time of my youth.’ [past] *‘I remember the time of my youth.’ [present] b. N kridita na Nhor Des. 1sg believe prep God. ‘I believed in God.’ *‘I believe in God.’

[past] [present]

The present tense readings are only obtained with the preverbal morpheme ta: (15) a. Dona Juana ten 80 anu, mas e ta lenbra Ms Juana have 80 year but 3sg tma remember tenpu di si mosindadi. time prep poss.3sg youth

‘Ms Juana is 80 years old, but she remembers the time of her youth.’ [her memory is in good health] [present]

b. N ta kridita na Nhor Des. 1sg tma believe prep sir god ‘I believe in God.’ [I am a believer]

[present]

 Fernanda Pratas

At this point, one could argue that these predicates are eventives, whose stative reading in the present is of the same nature as the generic/habitual N ta kume pexe ‘I eat fish.’ [everyday, habitually]. Strong evidence against this is presented in the next subsection. 2.2  Stative properties of these present situations This subsection deals with an almost immediate question when we look at the data in (15): are the predicates under analysis real statives? Even if they are, what prevents us from saying that their stative reading in the present is of the same nature as the generic/habitual N ta kume pexe ‘I eat fish.’ [everyday, habitually]. One of the reasons why they cannot be considered generics/habituals is related to their syntax: they are not compatible with agent-oriented adverbials, such as di abuzu ‘deliberately’. As Smith has claimed, “habitual sentences do not have the syntactic characteristics of basic-level statives. They allow the forms that are related to agency and control, unlike other statives.” (Smith 1991: 42) Therefore, I propose that E ta lenbra tenpu di si mosindadi ‘She remembers the time of her youth’ and N ta kridita na Nhor Des ‘I believe in God’ are basic-level statives and not generics or habituals of some sort. This holds for any stative situation with verbs from the list in (2). Two examples are presented in (16), with the statives gosta di ‘like’ and atxa ‘find’: (16) a. Maria ta gosta di xokolati *di abuzu. Maria tma like of chocolate    of abuse ‘Maria likes chocolate *on purpose / *deliberately.’ b. Djon ta atxa Maria bunita *di abuzu. Djon tma find Maria pretty    of abuse ‘Djon thinks (*on purpose / *deliberately) that Maria is pretty.’

Both verbs in (16) are marked by ta, which conveys a present reading, and both reject agent-oriented adverbials, which shows that, semantically, they are basiclevel statives. Another reason why we know that the predicates in (15) and (16) are basiclevel statives is related to their semantics: in the present tense, they need not be interpreted as generalizations or regularities (Smith 1991, 2003). Just like other non-derived states, they have no internal structure (there is no mapping of times to stages). As such, they are true of instants of time (Taylor 1977). One aspectual test to investigate the stative behavior of predicates concerns the interactions of these with the reference times given in the discourse context. Vlach (1981: 284) points out for English that “the progressive operator turns sentences into stative sentences and the defining characteristic of stative sentences is their way of interacting with point adverbials.” This is clearly observed in the following contrast:



‘I know the answer’ 

(17) English (Vlach 1981: 273–274): a. Max was here when I arrived. [state] b. Max was running when I arrived. [progressive] c. Max ran when I arrived. [non-stative]

Whenever the point/instant defined in the adverbial expression is able to fall within the time interval denoted by the main clause situation, this situation is of the stative type: either a basic-level stative or a derived-stative. This is shown, respectively, in (17a) and (17b): Max was already in the place/ already running before my arrival. In (17c), with a non-stative predicate (a process), Max necessarily started running at, or slightly after the moment of my arrival.11 For Capeverdean, where the progressive also has a stativizing function,12 these tests would result in aspectual operations that obscure the conclusions we are looking for. In (18), however, our goal of figuring out which predicates show a stative behavior is met when the intended instants of time are provided with recourse to a different strategy: the time reference given in a narrative.13 This narrative aligns some punctual occurrences in the present tense, making it possible to understand whether there is the above mentioned inclusion relation or a sequential interpretation of events. This method allows us to figure out which predicates reveal properties typical of stative situations. Stative situations are the ones that, coming in second place in each narrative line, hold for a period of time that includes the described punctual event, e.g. ‘a man enters the bar’. In other words, they are true of that instant of time. Since similar predicates may correspond to distinct situation types across languages, the relevant interpretation of the second predicate in these Capeverdean examples is provided.

11.  An anonymous reviewer pointed out that: “Smith (1997: 49) gives the following examples: John was dumbfounded when Harry threw the glass – s/he argues that this has a sequential reading parallel to Max ran when I arrived, in addition to its overlapping reading, and yet it is stative.” This is not relevant here, since the relation that is to be proven may be stated in these terms: whenever a sequential reading is imposed (in other words, whenever an overlapping reading is prohibited), this means that we have a dynamic /eventive situation. This does not make any predictions about the possible sequential readings of stative situations. 12.  Vlach states that this “is not just one fact about the progressive; this is what progressive is FOR” (Vlach 1981: 284). In other words, the “function of the progressive operator is to make stative sentences […]” (Vlach 1981: 274). This function of the progressive has been assumed for Capeverdean in Pratas (2010). 13.  For the relevant interactions between situation types in narratives, see Moens (1987) and Smith (2003).

 Fernanda Pratas

(18) a. Un omi ta entra na bar. E sta duenti.14 one man pres enter prep bar. 3sg be sick ‘A man enters the bar. He is sick.’ [basic-level state – overlapping] b. Un omi ta entra na bar. E sata kume banana. one man pres enter prep bar. 3sg prog eat banana ‘A man enters the bar. He is eating a banana.’ [progressive / derived state – overlapping] c. Un omi ta entra na bar. E ta kume banana. one man pres enter prep bar 3sg pres eat banana ‘A man enters the bar. He eats the banana.’ [culminated process – sequence] d. Un omi ta entra na bar. E sabe risposta. one man pres enter prep bar. 3sg know answer ‘A man enters the bar. He had (already) come to know the answer. / He knew the answer.’ Ambiguous: [culmination – reverse sequence] or [state – overlapping] e. Un omi ta entra na bar.  E ta kridita one man pres enter prep bar. 3sg pres believe na Nhor Des. prep God ‘A man enters the bar. He believes in God.’ [basic-level stative – overlapping] f. Un omi ta entra na bar. E ta lenbra one man pres enter prep bar. 3sg pres remember si mosindade. his youth ‘A man enters the bar. He remembers the time of his youth.’ Ambiguous: [culmination – sequence] or [state – overlapping]

Summarizing the relevant results of this test, we have: (i) kridita na Nhor Des ‘believe in God’ in (18e) clearly yields a stative interpretation; the man already believed in God when he entered the bar (in the same fashion that he also was sick (18a) / was eating a banana (18b)); (ii) with lenbra si mosindadi ‘remember 14.  In Capeverdean narratives, the present tense may have a non-generalizing interpretation (see Smith 2003 for a discussion on the various aspectual entities and tense in different ­discourse modes in English). This is also important for the present purposes, since the clauses with eventive predicates marked by ta do not have the generic or habitual reading that is expected for the same sentences outside a narrative sequence. This is the case of the first clause in each example, and also the second clause in (18c) and in (18f) (for the latter, this is the case when we focus on the culmination reading).



‘I know the answer’ 

his youth’, in (18f), a culmination reading is preferred (we have a sequence); (iii) crucially, for sabe risposta ‘know the answer’, in (18d), the culmination is preferred and we have a sequence in reverse; he got to know the answer before he enters the bar. This strikes a clear contrast with the other predicates and emphasizes the interpretative prominence of the prior culmination ‘get to know’. Note that the specific restrictions imposed by each predicate have been respected: for the necessary present interpretation with the verbs lenbra ‘remember’ and kridita ‘believe’, the obligatory morpheme ta has been used. If it were used with sabe ‘know’, the reading would be ‘he will know the answer’. This section has demonstrated that kridita na Nhor Des ‘believe in God’ and lenbra si mosindadi ‘remember his youth’ are basic-level satives: (i) they are not compatible with agent-oriented adverbials; (ii) in simple sentences in the present tense, they are not interpreted as regularities or generalizations and yet they are true of instants of time (Taylor 1977). These properties hold for any other verb from the list in (2). Therefore, we can now positively assert that not all Capeverdean states obey Bickerton’s (1981, 1984) generalization regarding the temporal interpretation of their bare forms. Even when they participate in basic-level stative situations, the unmarked forms of the lexical verbs listed in (2) cannot have a present interpretation. Again, it is important to stress that these verbs can also occur in non-stative predicates, but these eventive situations are not relevant for the current discussion. The temporal readings for the combinations of different Capeverdean predicate types with the various morphemes are summarized in Table 1. Table 1.  Temporal readings of different verb types combined with the different morphemes, in simple sentences and in out of the blue contexts Unmarked

-ba

ta

ta + -ba

sata

sata+-ba

sta duenti ‘be sick’ (stage-level)

pres

pst

?fut

?cond

–

–

e altu ‘be tall’ (individual-level)

pres

–

–

–

–

–

sabe risposta ‘know the answer’

pres

pst

fut

?cond

–

–

kridita na Nhor Des ‘believe in God’ all eventives

pst

? pst pft

pres

pst

–

–

pst

? pst pft

pres hab

pst hab

prog

pst prog

(i)  – means that these combinations are either impossible or odd in out of the blue contexts. (ii)   The combination of ta or -ba with e ‘be’ does not exist. For a past reading, there is the form era ‘was’, such as in Djon era altu ‘Djon was tall’ (iii) The question mark for the combination of sta duenti or sabe risposta with ta or with ta+-ba. means that, if these simple sentences occur in out of the blue contexts, which is not common, Future or Conditional, respectively, is their only possible interpretation The same applies to the question mark for the combination of ‘believe’, just as all the verbs listed in (2), or all eventives with -ba.

 Fernanda Pratas

The next section presents the analysis in Pratas (2010) and elaborates on the proposal that the zero operator is, in fact, a null Perfect morpheme.

3.  The relevant state is a Perfect state In Pratas (2010) it has been argued that sabe risposta ‘know the answer’ has a complex event structure: a culmination + a consequent state (Moens & ­Steedman 1988). The culmination ‘got to know / found out the answer’ functions here as a subevent of the type become (Dowty 1979), the temporal reading of the sentence being anchored on the consequent state. In this case, it coincides with the speech time, ‘[now] I know’. In (19), we have the same type of relation between an event and its consequent state, with the only difference that the temporal anchor precedes the speech time – this temporal anchor is visible here in the past morpheme -ba. (19) N sabeba risposta. 1sg know:pst answer ‘I knew the answer.’

The problem that is left unsolved by that proposal is the following: why is an equivalent consequent state not available for other telic situations? Compare the possible consequent states for the sentences in (20), the culminated process that we have been discussing, in (20a), and a culmination, in (20b): (20) a. N kume pexe. 1sg eat fish ‘I ate (the) fish’ b. N txiga sedu. 1sg arrive early ‘I arrived early.’

As has been mentioned earlier, there are, indeed, consequent states for these past events, namely ‘[now] the fish is eaten’ and ‘[now] I am here early’. These are not, however, of the same sort as the one for sabe risposta ‘know the answer’. What is the reason for this distinction? In the present paper I propose that the zero operator marking these so-called bare verb forms is, in fact, a null Perfect. Assuming a Perfect analysis for these contexts provides a straightforward explanation for the question just mentioned: only for sabe risposta ‘know the answer’ may a Perfect State be the direct result of the past eventuality.



‘I know the answer’ 

Before we proceed with the implications of this proposal, however, two clarifications are needed. First, the Perfect is not to be confused with Perfective, which refers to a closed aspectual viewpoint (Smith 1991: 147). The Perfect is a semantically complex category that involves certain temporal and aspectual characteristics; crucially, a perfective viewpoint is part of this semantic complexity, since Perfect constructions denote a state located at Reference Time, and this state is due to the prior occurrence of a closed situation. Portner (2003) similarly clarifies this distinction between the Perfect and perfective aspect, declaring that the “latter has to do with notions like the completion/non-completion of events, or whether they are viewed as an unanalyzed whole (e.g. Comrie 1976; Smith 1991; Kamp & Reyle 1993; Singh 1998).” (Portner 2003: 466) The author upholds that “the English perfect is perfective”, a characteristic that “it shares with the simple past”. Secondly, a null Perfect analysis has been proposed in van de Vate (2011) to explain the difference in temporal interpretation between stative (present) and non-stative (past) verbs in Saamáka. Although there are some significant similarities between the two Creole languages, there are also some important distinctions regarding their tense and aspectual systems. Therefore, the null Perfect analysis proposed here for Capeverdean is built on different grounds and has several distinct implications. For instance, van de Vate (2011) shows that Saamáka unmarked non-statives convey a perfect experiential reading and unmarked statives may have a perfect universal reading (van de Vate 2011: 48). The latter could not be the case in Capeverdean, since: (i) bare statives with copula verbs, like e altu ‘be tall’, can never have a perfect interpretation – they are straightforwardly present; (ii) the bare forms of lexical verbs that may enter stative constructions, such as the ones listed in (2), always have a past interpretation (thus, they do not have a universal perfect reading); (iii) cases like sabe risposta ‘know the answer’ or konxe Lisboa ‘know/be familiar with L ­ isbon’ could be the ones analyzed as having a universal perfect reading; ­however, I argue that the current proposal, based on a complex structure of these situations, seems to better account for the distinct temporal meaning of these Capeverdean predicates. Assuming for Capeverdean that the zero morpheme functions as a null Perfect marker has many interesting consequences. It accounts for the distinct temporal readings in the following way: (i) all bare forms of lexical verbs in root clauses are marked with a null Perfect, be they stative or non-stative (thus, it is not the case that these stative predicates show an eventive-like behavior, since stativity is not a key property here); (ii) the distinct temporal readings depend on the type of Perfect State located at Reference Time. I propose that for sabe risposta ‘know the answer’, the Perfect State is a type of consequent or result state

 Fernanda Pratas

(Moens & Steedman 1988; Smith 1991). Moreover, I argue that this ­consequent / result nature causes this Perfect State to be part of the complex structure of a ­situation. On the other hand, even for other eventive predicates that involve culminations, like kume pexe ‘eat the fish’ (culminated process) or txiga sedu ‘arrive early’ (­culmination), the Perfect State is a type of resultant state (Parsons 1990; ter Meulen 1995). As Portner (2011) puts it, the “resultant state is to be distinguished from a result state. A resultant state is not an ordinary state which has been caused by the past event described by the sentence, but rather a kind of abstract state of the event’s ‘having occurred’.” (Portner 2011: 1230) In the latter case, the Perfect State is not part of the event structure. Note that, for both instances, we obtain a relation between Event Time and Reference Time as defined in Reichenbach (1947): Present Perfect: e < r, s (Event Time < Reference Time, Speech Time) Past Perfect: e < r < s (Event Time < Reference Time < Speech Time)

Not surprisingly, the relevant temporal relation – Event Time precedes Reference Time – is apparently clearer for kume pexe ‘eat the fish’ than for sabe risposta ‘know the answer’.15 This is so because, as has just been argued, the latter situation has a more complex temporal schema, which includes the Perfect State. This temporal relation is illustrated in (21) for the Past Perfect: (21) a. Kantu mi era nobu, N ta pensaba When 1sg was young, 1sg tma think:pst ma N sabeba tudu risposta. that 1sg know:pst all answer

‘When I was young, I used to think I knew all the answers.’

15.  For the current purposes, a null Perfect analysis intends to account for the distinct temporal readings of bare verbs in Capeverdean. Cross-linguistically, the Perfect may have different readings (existential, universal, continuative, etc.) that are related, among other things, to the Akktionsart of predicates. For a recent overview on the different readings of the English Perfect, see Portner (2011). Although there is an ongoing study of the different Perfect readings in Capeverdean, it is beyond the scope of the present paper. However, I acknowledge that this ongoing research must deal with the fact that lenbra ‘remember’, kridita ‘believe’ or any other verb of the list in (2) may also occur in eventive situations, which will probably bring consequences to the different Perfect readings available for each of them. Also, there are several differences across languages regarding the interaction of the Perfect with temporal adverbials. This must also be studied for Capeverdean.



‘I know the answer’ 

b. Di noti N fika mariadu. N kumeba txeu pexe na djanta. at night 1sg get sick. 1sg eat:pst lot fish at dinner ‘In the evening I got sick. I had eaten a lot of fish for dinner.’

Finally, the perfective viewpoint involved in Perfect constructions (Smith 1991: 148–149) accounts for the previously mentioned effects of a zero operator. In Pratas (2010) it has been proposed that this zero morpheme adds a termination to atelic situations and a completion to telic situations. Under the null Perfect analysis, we have the notion of a prior occurrence of a closed situation – the specific perfective viewpoint aspect that is part of this semantically complex category.16 In this section, I have argued that the complex structure of predicates of the type N sabe risposta ‘I know the answer’ is better accounted for if we assume that the zero operator (Pratas 2010) is a null Perfect morpheme. This proposal also ­provides a uniform explanation for the past readings of bare eventives, be they telic or atelic, and of basic-level statives. As a final note to this proposal, it is important to show that, just like has been argued in Pratas (2010), this zero operator/null Perfect is in complementary distribution with the progressive marker sata. Observe the contrasts in (22) and (23): (22) a. N kume pexe. 1sg eat fish ‘I ate (the) fish.’ b. N sata kume pexe. 1sg prog eat fish ‘I am eating (the) fish.’ [answer to: what are you doing now?]

16.  Observe one bare eventive closed situation, with an atelic verb: (i) ??Djon nada i inda e sata nada.     Djon swim and still 3sg prog swim ‘Djon swam and he is still swimming.’ This example is odd, for a continuation to a closed situation is not expected. In other words, we know that a perfective aspect is at stake here, even though it is the perfective aspect that is part of the Perfect. As opposed to this, Djon nada parmanha interu i inda e sata nada ‘Djon swam all morning and he is still swimming’ is good. Here we have two distinct events: “Djon swam all morning” is one clearly closed/bounded event (the morning, an argument of this event, has ended) and “Djon is still swimming” is another. In other words, no continuation is at issue.

 Fernanda Pratas

(23) a. N kumeba pexe. 1sg eat:pst fish ‘I had eaten (the) fish’ b. N sata kumeba pexe. 1sg prog eat:pst fish ‘I was eating (the) fish.’ [answer to: what were you doing at 1pm?]

In (22), we have a temporal reference in the present, with the Perfect and the Progressive defining the event time: in (22a), the Perfect construction, the event time precedes the reference time; in (22b), the Progressive construction, the event time overlaps the reference time. In (23), the relations between event time and reference time are just the same as described for (22); the difference is that we have a temporal reference in the past: in (23a), the event time precedes a past time reference; in (23b), the event time overlaps a past time reference.17 In the next section, an approach to ‘know’ in Korean is brought into the discussion. The purpose is to show that the semantic restrictions involving the Perfect and this particular verb have been attested in a different language. Therefore, it is not surprising that the Perfect State related to sabe risposta ‘have known the answer’ is distinct from the ones obtained in other situations. There are, however, some differences between Choi (2010) and the current proposal.

4.  Cross-linguistic idiosyncrasies of ‘know’ In her study about the different types of stative predicates in Korean, Choi (2010) shows that the Perfect marker in this language, -ess, when combined with an emotive verb like ‘love’, salangha, has a past reading (24), but when combined with the mental verb ‘know’, al, it may have a present reading (25). (24) Juno-nun Yuna-lul olaecene/*cikum salangha-ss-ta. Juno-top Yuna-acc a long time ago / now love-pft-dec ‘Juno loved Yuna a long time ago/*now.’ (25) Minho-nun ku sasil-ul olaecene/cikum al-ass-ta. Minho-top the fact-acc a long time ago / now know-pft-dec ‘Minho found out / is aware of the fact a long time ago / now.’

17.  In Pratas (2011a,b), these perfect/progressive distinctions have been represented in the terms defined in Demirdache and Uribe-Etxebarria (2007), who follow Klein’s (1995) notions of tense (a relation between reference time and the time of utterance) and aspect (a relation between reference time and event time).



‘I know the answer’ 

For Choi (2010), the relevant generalization is the following: atelic predicates + -ess = existential past; telic predicates + -ess = result state. She then proposes that Korean ‘know’ behaves as a telic predicate, an idea that is reinforced by the following contrast: whereas ‘love’+-ess is compatible with expressions of the type ‘for X time’ and incompatible with the ones of the type ‘in X time’ (thus, the combination is atelic), ‘know’+-ess is compatible with expressions of the type ‘in X time’, which could lead to the conclusion that Korean ‘know’ is telic. The test with ‘since X time’, however, shows that al ‘know’+-ess patterns with atelics (26). (26) Juno-nun ku pati-ihu ku sasil-ul al-ass-ta. J-top the party-since the fact-acc know-pft-dec ‘Juno has known the fact since the party.’

Choi (2010) argues that ‘know the fact’ (along with states like ‘be old’)18 is an inchoative state: it is compatible with ‘since X time’ adverbials because it denotes the inception of a state and it is compatible with ‘in X time’ adverbials because it describes a time of transition. The author follows Bar-el’s (2005) study for Squamish Salish in assuming that, like all (underived) Squamish Salish stage-level states, the Korean inchoative states have an inception (initial boundary), but no culmination (final boundary) in their meanings. In Capeverdean, sabe ‘know’ + null Perfect is also compatible with any of these adverbial expressions. The current proposal, however, predicts a different description of these facts. Consider the following examples: (27) E sabe risposta na tres minutu. 3sg know answer prep three minute ‘He got to know / found out the answer in three minutes.’ (28) E sabe risposta desdi simana pasadu. 3sg know answer prep week last ‘He has known the answer since last week.’

I argue that, in (27), the ‘in X time’ expression has a similar function as ‘yesterday’ adverbials; it cancels any temporal anchor on the Perfect State, and we have a past interpretation, such as in kume pexe ‘ate the fish’. In other words, there is nothing that distinguishes this from other telic predicates in the Perfect. Therefore, 18.  I do not include these verbs in the discussion here, since in Capeverdean ‘be old’ and ‘get old’ are denoted by different expressions: the first one is a state (either stage-level, sta bedju, or individual-level, e bedju), and the second one is not; it includes a culmination of the ‘get’ type, fika bedju or bira bedju. Therefore, the latter, both meaning ‘get old’, are true inchoatives in the sense defined in Smith (1991: 44): ‘John got angry’, ‘Mary became tired’.

 Fernanda Pratas

the notion of an inchoative state is not relevant here. In (28), the ‘since X time’ expression reinforces the temporal anchor on the Perfect State, here analyzed as “a state due to the occurrence of the situation mentioned” (Smith 1991: 148). We are informed that the relevant prior situation occurred last week, but the temporal reading of the sentence is present. If this is on the right track, we do not need the independent notion of an inception to the state of knowing. This notion accounts for the fact that this state has an initial boundary; however, at least for Capeverdean, it does not explain the possible present reading of its bare form. We may contrast this with another example, with a basic-level stative. (29) N *(ta) kridita na Nhor Des desdi anu pasadu. 1sg  tma believe prep sir god prep year last ‘I believe in God since last year.’ [I am a believer] [present]

In (29), the initial boundary of the state is also expressed, and yet it needs ta to have a present interpretation. Furthermore, the initial boundary of present sabe ‘know’ may seem very clear in (28), because of the ‘since X time’ expression, but it is much less clear in the absence of this expression, like in (30). (30) Pedru so ten 4 anu mas e sabe konta ti 20. Pedru only have 4 year but 3sg know count until 20 ‘Pedru is only 4 years old but he can count until 20.’

Consequently, I argue that this notion is intrinsic to the properties of the particular type of Perfect State available for sabe ‘know’. In this section I have brought Choi’s (2010) analysis of ‘know’ in Korean into the discussion. This has been done to show that the semantic restrictions involving the Perfect and this particular verb have been attested in a different language. There is, however, a crucial distinction between her analysis and the one proposed in this paper: the null Perfect proposal described here accounts for all types of situations with lexical verbs, be they telic or atelic. The distinct temporal interpretations depend on the nature of the Perfect State. Additionally, it dispenses with the independent notion of an inception to the state of knowing, for this notion, by itself, does not account for the fact that other bare statives with an inception description cannot have a present interpretation.

5.  Final remarks Stativity, as a lexical property of some predicates, does not account for the following temporal contrast in Capeverdean: N sabe risposta ‘I know the answer’



‘I know the answer’ 

(present) vs. N kume pexe ‘I ate the fish’ (past). In fact, some predicates that show a semantic behavior typical of states pattern with eventives in this respect. This is the case with N kridita na Nhor Des ‘I believed in God’ and N lenbra nha mosindadi ‘I remembered my youth’ (past). The current proposal accounts for this puzzle in the following way: all allegedly bare forms of lexical verbs in simple sentences, be they eventive or stative, are marked by a null Perfect morpheme. Perfect sentences denote a state located at reference time, which is due to the prior occurrence of a closed situation (Smith 1991: 147). Thus, they have a perfective aspect. I argue that only for sabe risposta ‘know the answer’ may a Perfect State be the direct result of the past eventuality (Smith 1991). I argue that this direct result nature explains why this Perfect State is part of the event structure. When the Perfect State is, as Portner (2011) puts it, an abstract state of the event’s ‘having occurred’ (Parsons 1990; ter Meulen 1995), it does not participate in the event structure. This Perfect proposal has only tackled the distinct temporal readings of ­Capeverdean predicates. In future studies, the different interpretations of the ­Perfect and their modal implications shall be described.

References Baptista, Marlyse. 2002. The Syntax of Cape Verdean Creole, the Sotavento varieties. Amsterdam/ Philadelphia: John Benjamins. Bar-el, Leora. 2005. Aspectual Distinctions in Skwxwú7mesh. Ph.D. dissertation, University of British Columbia. Bickerton, Derek. 1981. Roots of Language. Ann Arbor: Karoma. Bickerton, Derek. 1984. The language bioprogram hypothesis. The Behavioral and Brain Sciences 7, 173–88. Borik, Olga & Fernanda Pratas. 2008. “Stativity and Temporal Interpretation in CV”. Paper presented at Formal Approaches to Creole Studies (FACS), Tromsø, November 2008. Choi, Jiyoung. 2010. “The puzzle of stative predicates in Korean”. Ms. 7th Workshop on Syntax and Semantics (WoSS7), Université de Nantes. Comrie, Bernard. 1976. Aspect. Cambridge: Cambridge University Press. Demirdache, Hamida & Miryam Uribe-Extebarria. 2007. The syntax of time arguments. Lingua 117, 330–366. Dowty, David R. 1979. Word meaning and Montague grammar. Dordrecht: Kluwer. Gehrke, Berit & Nino Grillo. 2009. “How to become passive”. Exploration of Phase Theory: features, arguments, and interpretation at the Interfaces ed. by Kleanthes Grohmann, 231–268. Berlin: De Gruyter. Kamp, Hans & Uwe Reyle. 1993. From discourse to logic: Introduction to modal theoretic semantics of natural language, formal logic and discourse representation theory. Dordrecht: Kluwer. Klein, Wolfgang. 1995. “A time relational analysis of Russian aspect.” Language 68, 525–552.

 Fernanda Pratas ter Meulen, Alice. 1995. Representing Time in Natural Language: The Dynamic Interpretation of Tense and Aspect. Cambridge, Mass: MIT Press. Moens, Marc. 1987. Tense, Aspect and Temporal Reference. Ph.D. dissertation, Edinburgh University. Moens, Marc & Mark Steedman. 1988. “Temporal ontology and temporal reference”. Computational Linguistics, 15–28. Parsons, Terence. 1990. Events in the Semantics of English: A Study of Subatomic Semantics. Cambridge, Mass: MIT Press. Portner, Paul. 2003. Linguistics and Philosophy 26, 459–510. Portner, Paul. 2011. “Perfect and progressive”. Semantics: An International Handbook of Natural Language Meaning ed. by Claudia Maienborn, Klaus von Heusinger and Paul Portner, 1217–1261. Berlin: Mouton de Gruyter. Pratas, Fernanda. 2007. Tense Features and Argument Structure in Capeverdean Predicates. Ph.D. dissertation, Universidade Nova de Lisboa. Pratas, Fernanda. 2010. “States and Temporal Interpretation in Capeverdean”. Romance Languages and Linguistic Theory 2008 – Selected papers from Going Romance Groningen 2008 ed. by Bok-Bennema, Reineke, Brigitte Kampers-Manhe & Bart Hollebrandse, 215–231. Amsterdam/Philadelphia: John Benjamins. Pratas, Fernanda. 2011a. “Tense and Aspect: interactions in Capeverdean.” Paper presented at Chronos 10, Birmingham, April 2011. Pratas, Fernanda. 2011b. “Temporal and event interpretation. in Capeverdean”. Paper presented at TMEI (Temporal, modal and event interpretation in natural language discourse), Açores, June 2011. Reichenbach, Hans. 1947. Elements of Symbolic Logic. New York: Macmillan. Rendall, Helderyse. (in preparation). Aspects of Passives in Capeverdean. MA dissertation, Universidade Nova de Lisboa. Silva, Izione S. 1985. Variation and Change in the Verbal System of Capeverdean Crioulo. Ph.D. dissertation, Georgetown University. Singh, Mona. 1998. “The Semantics of the Perfective Aspect”. Natural Language Semantics 6, 171–199. Smith, Carlota. 1991. The Parameter of Aspect. New York: Kluwer. Smith, Carlota. 2003. “Aspectual entities and tense in discourse.” The Syntax, Semantics and Acquisition of Aspect ed. by P. Kempchinsky and S. Roumyana. Dordrecht: Kluwer. Suzuki, Miki. 1994. “The markers in Cape Verdean CP”. Ms, CUNY. Taylor, Barry. 1977. Tense and continuity. Linguistics and Philosophy 1: 2, 199–220. van de Vate, Marleen. 2011. Tense, Aspect and Modality in a radical creole: The case of Saamáka. Ph.D. dissertation, University of Tromsø. Veenstra, Tonjes. 2008. “Creole Genesis: The Impact of the Language Bioprogram Hypothesis.” The Handbook of Pidgin and Creole Studies ed. by Silvia Kouwenberg & John Victor Singler, 219–241. Oxford: Blackwell. Vendler, Zeno. 1957. “Verbs and times”. The Philosophical Review 66: 2, 143–160. [reprinted in Vendler, Zeno. 1967. “Verbs and times”. Linguistics in philosophy ed. by Zeno Vendler, 97–121. Ithaca: Cornell University Press.] Vlach, Frank. 1981. “The Semantics of the Progressive”. Syntax and Semantics 15: Tense and Aspect ed. by P. Tedeschi & A. Zaenen, 271–292. New York: Academic Press.

Stressed vowel duration and stress placement in Italian What paroxytones and proparoxytones have in common* Stefano Canalis & Luigia Garrapa This paper examines stressed vowel duration and word stress placement in Italian. It presents experimental data showing that absolute duration is slightly longer in paroxytones than in proparoxytones (as previously reported), but relative duration (computed as the ratio between stressed vowel duration and the duration of the immediately post-tonic vowel) remains constant. The data also suggest that proparoxytones have a word-final secondary stress. The paper also examines the status of weight-sensitivity in Italian, arguing (on the basis of stress placement in loanwords, acronyms and non-standard pronunciations) that it is no longer productive. The same data also show that even-syllabled words tend to receive penultimate stress, while odd-syllabled words tend to be antepenultimate. The paper argues that it is the result of a conflict between preference for penultimate stress and the need to parse all syllables. Keywords:  vowel duration; word stress; Italian phonology

1.  Introduction The interplay between word stress and vowel duration in Italian has been the subject of much phonetic and phonological research and debate. While a consensus exists on several aspects of this issue, some points remain controversial. In some cases, controversy is not limited to the analysis and explanation of a widely accepted body *  We would like to thank our speakers for participating in the experiment, and Birgit Alber, Roberta D’Alessandro, Jacopo Garzonio, Maria Giavazzi, Pavel Iosad, Martin Krämer, G ­ iovanna Marotta, Bruce Mayo, Marc van Oostendorp, Diana Passino and three anonymous reviewers for their comments and suggestions. Needless to say, all remaining faults are our own. Although this paper is the result of a joint effort, Sections 1, 2.4, 3.1, 3.2.1 to 3.2.2, 4, 5.2.1, 6 and 7 were written by Stefano Canalis, and Sections 2.1 to 2.3, 3.2, 5.1 and 5.2 by Luigia Garrapa.

 Stefano Canalis & Luigia Garrapa

of evidence, but concerns empirical data as well. One of the most debated problems is the nature of stressed vowel duration in proparoxytones. On the phonetic side, some authors claim that it is the same as in paroxytones, whereas others present data supporting the opposite view. On the phonological side, various metrical analyses of proparoxytones have been proposed (ternary feet, extrametricality of the final syllable, and so on). In this paper we present the results of an experimental study on stressed vowel duration in Italian (§ 3), and discuss their implications for a phonological analysis of Italian stress (§ 4). It will be shown that vowel duration follows a parallel pattern in both paroxytones and proparoxytones (although absolute duration is slightly longer in the former), and that in proparoxytones the final vowel is longer than the penultimate. We will also discuss the role of syllable weight and word length in influencing stress position when stress is not lexical (§ 5). We interpret our data and observations as motivating an analysis of Italian metrical structure in terms of binary syllabic feet in both paroxytones and proparoxytones, the final syllable of proparoxytones being parsed as a degenerate foot. We will also argue that different stress patterns in Italian arise from the interaction between a tendency to stress odd-syllabled words as proparoxytones to parse all syllables, and a tendency favouring penultimate stress (§ 6). Before presenting our own experiment and analysis, we review some basic facts about Italian stress and vowel duration in § 2.1 to 2.3, and briefly discuss previous analyses and explanations in § 2.4. 2.  Italian stress position and vowel duration 2.1  Stress position In Italian primary stress falls on one of the last three syllables, counting from the right word edge. As the examples in (1) to (3) show, both open and closed syllables are eligible stress bearers. Enclitics are ignored by stress assignment principles; as a result, stress may be on the fourth or fifth from last syllable if clitics adjoin to a verb (4). (1) Oxytones a. [me.'ta] metà ‘half ’ b. [e.'lit]1 élite ‘elite’

1.  Consonant-final Italian words are usually loanwords or acronyms.



Stressed vowel duration and stress placement in Italian 

(2) Paroxytones a. [a.'mi:.ko] amico ‘friend’ b. [fi.'nɛs.tra] finestra ‘window’ (3) Proparoxytones a. ['ɛ:.zi.to] esito ‘result’ b. ['ven.di.ta] vendita ‘sale’ (4) a. ['da:.te=me=lo] datemelo ‘give-imp it to me’ b. ['ɔk.ku.pa=te=ne] occupatene ‘take-imp care of it’

Final, penultimate and antepenultimate stress are all possible, but this does not mean they are equally likely. On the contrary, their relative proportions vary greatly. Word-final stress is not frequent (~2 percent of Italian lexicon); penultimate stress is the most common option (about 80 percent), while antepenultimate stress is found in about 18 percent of words (cf. Thornton et al. 1997).2 Syllable weight appears to play a role in determining stress location. If the penultimate syllable is closed, it is almost always stressed (e.g. 2b). However, a handful of ‘exceptions’ to this generalization exist (5). (5) a. ['man.dor.la] b. ['pɔ:.lit.tsa] c. ['a:.ris.ta] d. ['a:.kan.to] e. [ka.'nɛ:.der.lo] f. ['ɔ:.tran.to] g. ['ta:.ran.to] h. ['lɛ:.van.to] i. ['ka:.or.le] j. ['lɛ:.pan.to]

mandorla ‘almond’ polizza ‘insurance policy’ arista ‘chine of pork’ acanto ‘acanthus’ canederlo ‘kind of dumpling’ Otranto town name Taranto town name Levanto town name Caorle town name Lepanto town name

2.2  Stress is contrastive With the exception of the restriction to the last three syllables, and possibly the role of syllable weight, stress position is not subject to other absolute constraints and is therefore largely unpredictable. This is clearly demonstrated by the existence of minimal pairs distinguished only by stress position.

2.  The data refer to frequency in the lexicon; frequency of usage may be partially different. For example several oxytones are high-frequency words (as sì “yes”, no “no”, qui “here”, però “but”), which could increase the frequency of final stress.

 Stefano Canalis & Luigia Garrapa

(6) a. ['ka.:pi.to] b. [ka.'pi:.to] c. ['prin.ʧi.pi] d. [prin.'ʧi:.pi] e. ['aŋ.ko.ra] f. [aŋ.'ko:.ra] g. ['sa:.ra] h. [sa.'ra]

càpito3 ‘(I) happen to’ capìto ‘understood’ prìncipi ‘princes’ princìpi ‘principles’ àncora ‘anchor’ ancóra ‘still/again’ Sàra ‘Sarah’ sarà ‘she/he/it will be’

2.3  Stressed vowel duration ‘Stress’ is a fairly abstract phonological entity. In many languages it has no simple, straightforward and unidimensional phonetic correlate; its phonetic realization is usually implemented by a complex and language-specific combination of intensity, duration, pitch, to name only the most important parameters (among many others, see Lehiste 1970; Hayes 1995: 5–9). Italian stress is no exception, as differences in duration, intensity and pitch have been reported as physical correlates of word stress (among others, see ­Bertinetto 1981). Among them, duration appears to be the most important dimension in the perception of Italian stress. Bertinetto (1981: 80–86), using synthetic speech stimuli, manipulated duration, intensity and pitch one at the time, keeping the other two parameters constant. When Italian speakers were asked to locate stress position in the stimuli presented to them, by far the most reliable acoustic cue was duration, followed by intensity and finally by pitch. Alfano et al. (2009) report broadly similar findings with respect to duration and pitch (they do no discuss intensity). They also show to what extent the phonetic cues of stress may be language-specific even when closely related languages are compared: in their study the manipulation of duration and pitch yields very different results for ­Spanish speakers, who apparently rely more on pitch than duration in order to detect stress position. Given the major role played by duration as a phonetic correlate of Italian stress, in the following sections we will mainly focus on this vocalic property (with a brief digression on unstressed vowel reduction, § 3.2.2 and 4.2), although a more comprehensive account of Italian stress should include at least intensity and pitch as well. Vowel duration in Italian can vary significantly. However, it is never contrastive, being largely predictable on the basis of several parameters. They

3.  Grave accents mark word stress in Italian spelling, and are therefore not to be interpreted as secondary stresses.



Stressed vowel duration and stress placement in Italian 

include segmental ones, like vowel height, manner of articulation and place of articulation of the adjacent consonants, but the most important4 are stress position and syllable structure. All other things being equal, vowels in non-final stressed open syllables (as [a] in 7a to 7b) are longer than any other vowels ([a] in 7c to 7e). (7) a. ['ka:.re] care ‘dear-f-pl’ b. ['ka:.ri.ko] carico ‘load (noun)’ c. [ka.'ri:.ne] carine ‘dear-f-pl-dim’ d. ['kar.te] carte ‘papers’ e. [me.'ta] metà ‘half ’

This is not to be taken to mean that vowels in any other environment are equally short, but the standard interpretation (e.g. Canepari 1977; Nespor & Vogel 1986) is to consider non-final stressed vowels in open syllables as lengthened and all the other vowels as short. In fact, labelling a vowel as ‘long’ or ‘short’ sometimes crucially depends on the criteria adopted by the analyst. Measurements usually show that word-final stressed vowels are shorter than non-final ones, but some authors (Vayra 1994; van Santen & D’Imperio 1999) report the former to be often followed by a phase of glottalization or breathy voice. If this portion is included in the duration of the vowel, word-final stressed vowels in van Santen and D’Imperio’s (1999) corpus are slightly longer than antepenultimate, and slightly shorter than penultimate stressed vowels. 2.4  Explanations of vowel duration The problems raised by the durational difference between penultimate and antepenultimate stressed vowels are even deeper, as they involve not only the choice of measurement criteria, but explanations too. As mentioned in the previous section, the standard position is to consider stressed vowels in non-final open syllables (be they penultimate or antepenultimate) as lengthened, and all the other vowels as short. However, many works (among others, Marotta 1985; Farnetani & Kori 1990; D’Imperio & Rosenthall 1999; van der Veer 2006; Hajek et al. 2007) report that stressed vowels in open syllables are shorter in proparoxytones than in paroxytones. For example, the overall mean duration of

4.  Leaving aside the influence of higher prosodic domains. Stressed vowels undergo ­significant lengthening only when the words they belong to are in a prominent prosodic ­position, that is sentence-final and/or phrase-final; see Farnetani and Kori (1990), Bertinetto (2010: 486–487) and the references therein.

 Stefano Canalis & Luigia Garrapa

[a] in D’Imperio & Rosenthall’s (1999: 4) study is 177 ­milliseconds in fàte ‘you-pl do’ and 149 milliseconds in fàtele ‘do-imp them’. Several proposals have been made to explain this asymmetry. Farnetani and Kori (1990) suppose that it depends on the distance from major prosodic boundaries: stressed vowels would be longer when closer to a major prosodic boundary. Since a penultimate stressed vowel in a sentence final-word is closer to the right prosodic boundary than an antepenultimate in the same context, their hypothesis predicts the former to be longer. Marotta (1985) and van der Veer (2006) suggest instead a word compression effect. It has been observed in several languages that when overall word duration increases, vowel duration tends to decrease; and – by definition – paroxytones are at least disyllabic, while proparoxytones are at least trisyllabic. In fact, in van der Veer’s (2006) experimental data the formula word length (in milliseconds)/number of syllables in the word accounts for more variance of stressed vowel duration than stress position does (47.7 vs. 34.3 percent; van der Veer 2006: 51). Marotta (1985) observes, however, that her own data only partially support the compression effect; it is attested in stress-initial words, but not in stress-medial or stress-final ones (for example, in her experiments the stressed vowel of pesàte ‘you-pl weigh’ is consistently shorter than that in pesàtemi ‘weigh-imp me’). To complicate the picture further, Landi and Savy (1996: 69) find very little support for a word compression effect – in their data, stressed vowels are on average longer in four-syllable words than in two- and three-syllable words (although their study is not readily comparable to the others, since it lumps together vowel durations of proparoxytones, paroxytones and oxytones). The problem of the durational difference between paroxytones and proparoxytones is also addressed by D’Imperio and Rosenthall (1999), who challenge both explanations (distance from major boundaries and compression effect) and offer an alternative solution. Concerning the compression effect, they observe that in their study it does not affect all stress positions equally, and “penultimate syllable duration stays the same when word length increases by adding syllables to the left of the word’ (D’Imperio & Rosenthall: 9). ­Concerning the distance from prosodic boundaries, they observe that their and Marotta’s (1985) data come from words uttered in carrier sentences where the target word is never in absolute final position (and hence never immediately preceding the sentence’s right boundary); however, antepenultimate stressed vowels are shorter than penultimate (cf. their fate/fatele example mentioned above) in a statistically significant way in their corpus. They conclude that previous hypotheses are not convincing, and also that it is not possible to treat the two stress positions as a homogeneous category; rather, lengthening must come from two different sources.



Stressed vowel duration and stress placement in Italian 

Since they assume both syllabic and moraic feet for Italian, they ascribe difference in stressed vowel duration to a difference in foot form. Although they rank syllabic feet as more optimal than moraic feet, a moraic foot leaves the wordfinal syllable unparsed in paroxytones, which satisfies their constraint NonFinal (“the head foot of the prosodic word must not be final”, D’Imperio & Rosenthall 1999: 13); in proparoxytones the same constraint is best satisfied by a syllabic foot. Therefore they make a distinction between ‘phonological’ and ‘phonetic’ duration: in their opinion, a penultimate stressed vowel is significantly lengthened to create an additional mora, and hence form a bimoraic foot. On the contrary, in proparoxytones stressed vowel duration is phonetic, since the foot is syllabic. In the latter case duration is a phonetic correlate of stress (together with intensity and pitch), and thus does not need be as prominent as in paroxytones. 3.  Experiment The available evidence on vowel duration is sometimes conflicting, with different studies reporting diverging results and at the same time adopting different criteria, assumptions and methods. As an attempt to overcome this state of affairs, we designed our own experiment. Previous analyses have assumed that the most important dimension of vowel duration is either its absolute value (e.g. D’Imperio & Rosenthall 1999) or the ratio between duration and word length (e.g. Marotta 1985; van der Veer 2006). ­Following an argument at the core of various trends of research in prosodic matters (among many others, Lehiste 1970), we think it is useful to start from the viewpoint that stress and duration are first and foremost relative properties. A vowel is not ‘stressed’ or ‘long’ in itself, but only with respect to less prominent or shorter vowels. Consequently, relative values may be more informative and revealing than absolute ones, as far as stress and duration are concerned. Interestingly, other aspects of Italian phonology already provide evidence for the hypothesis that perceptual invariants may be based on relational rather than absolute properties. Pickett et al. (1999) show that the contrast between Italian singleton and geminate consonants is captured by a relational measure: “the ratio between consonant duration and preceding vowel duration discriminate[s] between singletons and geminates both within and across speaking rates’ (p. 135), whereas the (absolute) parameter of closure duration alone fails to discriminate singletons and geminates across speaking rates. However, the available empirical evidence on vowel duration shows that a relative value, namely the ratio between stressed vowel duration and number of syllables within a word, is at best only partially significant in explaining

 Stefano Canalis & Luigia Garrapa

the duration patterns of Italian. The indicator we propose to adopt is the ratio between stressed vowel duration and duration of the immediately post-tonic vowel (in (8) an actual example from our data – speaker F2’s first tokens of the word crede and credere – is given). As the two vowels are in adjacent syllables it is reasonable to assume that their respective durations are closely related.

(8) a. Vowel duration in paroxytones (e.g. ['kre.de] ‘(s/he/it) believes’)



Duration ratio =

Duration of ['e]

=

97 ms

Duration of [e]

=

51 ms

= 1.9

b. Vowel duration in proparoxytones (e.g. ['kre.de1.re2] ‘to believe’‌)

Duration ratio =

Duration of ['e]

=

106 ms

Duration of [e1]

=

54 ms

= 1.96

Additionally, some previous studies report the immediately post-tonic vowel of proparoxytones to be shorter than the following vowel (e.g. Marotta 1985: 135). Since this fact might be relevant in understanding the prosodic properties of proparoxytones, we measured the duration of all their post-tonic vowels as well. 3.1  Design of the experiment Six speakers (four females and two males, aged 27–35) from different Italian regions (Apulia, Latium, Piedmont, Veneto (2 speakers) and Tuscany) produced 6 paroxytones and 6 proparoxytones within the carrier sentence Dico [word] piano ‘(I) say [word] slowly’ (in order to minimize the effect of higher level prosody, avoiding sentence-final lengthening), reading it at a natural pace. Age, sex and place of origin of the speakers merely depend on their availability. All paroxytones were disyllabic and all proparoxytones were trisyllabic. To neutralize the role of segmental factors on vowel duration, each paroxytone was chosen so as to be identical to the first two syllables of a proparoxytone (9); each stimulus was repeated three times. The recordings were made in a silent room using an Audio Technica AT2020 USB microphone, recorded on a PC with the software Audacity, and then analysed with Praat (Boersma & Weenink 2010). Vowel duration was measured from the offset of the preceding consonant to the onset of the following consonant, excluding transitions. (9) Paroxytones Proparoxytones a. crede ['kre:.de] ‘(s/he/it) believes’ credere ['kre:.de.re] ‘to believe’ b. cene ['ʧe:.ne] ‘dinners’ cenere ['ʧe:.ne.re] ‘ash’ c. scrive ['skri:.ve] ‘(s/he/it) writes’ scrivere ['skri:.ve.re] ‘to write’ d. coni ['kɔ:.ni] ‘cones’ conici ['kɔ:.ni.ʧi] ‘conical-pl’ e. fate ['fa:.te] ‘do-imp/fairies’ fatele ['fa:.te.le] ‘do-imp them-f’ f. dite ['di:.te] ‘say-imp’ ditele ['di:.te.le] ‘say-imp to them-f’



Stressed vowel duration and stress placement in Italian 

3.2  Results

60 80 100

Paroxytones Proparoxytones

0

20

40

Milliseconds

130

160

The results of the experiment show that there is some preference for longer stressed vowels in penultimate syllables, but not in a clear-cut way. For instance the mean duration of the stressed vowels of crede and credere (Figure 1) is clearly longer in the former word for three speakers (F2, F3 and F4), but two other speakers (F1 and M1) display very little difference, and for one speaker (M2) the stressed vowel of credere is longer. Stressed vowels in the other word pairs have comparable differences.

F1

F2

F3 F4 Speakers

M1

M2

Figure 1.   Mean ['e] duration in crede and credere for all speakers

Table 1.  Mean ['e] duration in crede and credere for all speakers Speakers Word type Paroxytones Proparoxytones

F1

F2

F3

F4

M1

M2

74 ms 71 ms

114 ms 97 ms

153 ms 125 ms

115 ms 103 ms

117 ms 110 ms

98 ms 124 ms

An ANOVA test (with stress position, speaker and word as factors) shows that the speaker and word effects are highly significant in accounting for duration differences ([F(5, 144) = 72.7664, p < 0.001] and [F(11, 144) = 12.7357, p < 0.001]), but stress position is not [F(1, 144) = 2.85, p = 0.0935]. Only the interaction of the speaker and word factors is highly significant [F(46, 144) = 2.7996, p < 0.001]. These results thus seem to offer an ambiguous picture: the hypothesis that stressed vowels are longer in paroxytones is not confirmed, but is not wholly disproved either (in most cases paroxytones have longer stressed vowels, and the

 Stefano Canalis & Luigia Garrapa

‘V/V Ratio

0.0 0.6 1.0 1.5 2.0 2.5 3.0 3.5 4.0

p-value of the ANOVA test is not substantially above the 5 percent significance level). However, stressed vowel duration shows a different pattern if computed with the formula proposed above and exemplified in (8). If relative duration is taken into account, stress position becomes irrelevant: with regard to the words crede and credere, three speakers have higher ratios in paroxytones, but for the other three the opposite is true (Figure 2). A similar situation holds for the other word pairs. This is confirmed by an ANOVA test (again with stress position, speaker and word as factors); the effect of stress position on the vowel ratio is not statistically significant [F (1, 144) = 0.0863, p = 0.7694], whereas the speaker and word factors are ([F (5, 144) = 18.2362, p < 0.001], [F (11, 144) = 12.3113, p < 0.001]) as well as their interaction [F (46, 144) = 2.0565, p < 0.001).

Paroxytones Proparoxytones

F1

F2

F3 F4 Speakers

M1

M2

Figure 2.  Mean duration ratio (stressed V / post-tonic V) in crede and credere for all speakers

Table 2.  Mean duration ratio in crede and credere for all speakers Speakers Word type

F1

F2

F3

F4

M1

M2

Paroxytones Proparoxytones

1.62 1.09

2.25 1.91

1.89 2.24

2.08 1.99

1.84 2.09

1.92 2.3

3.2.1  Post-tonic vowels of proparoxytones: Duration The second aspect we wanted to investigate was post-tonic unstressed vowel duration in proparoxytones. Our data (Figure 3) show that for all speakers but one the second post-tonic (i.e. final) vowel of the word credere is longer than the first post-tonic (i.e. penultimate) vowel, although only slightly so. Over all proparoxytones, the mean difference is 9.3 ms in favour of the final vowel. An ANOVA test



Stressed vowel duration and stress placement in Italian 

80 100 120 140 160 60

Stressed V Penult V Final V

0

20

40

Milliseconds

c­ onfirms that vowel position is highly significant (F[1, 144] = 29.7793, p < 0.001) in accounting for duration differences among post-tonic vowels. Speaker and word factors are significant as well, as their interaction.

F1

F2

F3 F4 Speakers

M1

M2

Figure 3.   Mean V durations in credere

Table 3.  Mean V durations in credere Speakers Vowels

F1

F2

F3

F4

M1

M2

Stressed V

71 ms

97 ms

125 ms

103 ms

110 ms

124 ms

1st post-tonic V 2nd post-tonic V

65 ms 60 ms

51 ms 52 ms

57 ms 80 ms

52 ms 56 ms

53 ms 61 ms

55 ms 58 ms

3.2.2  Post-tonic vowels of proparoxytones: Centralization Shortening is a common form of vowel reduction, but by no means the only one. Another property frequently associated with reduced vowels is centralization; for example Italian unstressed vowels not only are shorter than stressed ones, but are also more centralized (Albano Leoni et al. 1995). Given this relationship between vowel reduction and centralization, and the finding that the final vowel of Italian proparoxytones is slightly longer than the penultimate (§ 3.2.1), we would expect the former to be less centralized as well. Figures 4 and 5 show F1 and F2 values (measured at the midpoint of the vowel) of penultimate and final /e/’s in the proparoxytones of two of our speakers. Both figures indicate that /e/ in penultimate position has higher F1 and lower F2 than word-finally, resulting in a more centralized mid front vowel.

 Stefano Canalis & Luigia Garrapa F2 (Hz) 2400

2000 1800 1600

1400

1200 300 400

600

F1 (Hz)

500 Penult [e] Final [e] Mean of penult [e]’s Mean of final [e]’s

700 800 900

Figure 4.   Formant frequencies of speaker F2’s unstressed /e/’s in proparoxytones F2 (Hz) 2400

2000 1800 1600

1400

1200 300 400

600

F1 (Hz)

500

Penult [e] Final [e] Mean of penult [e]’s Mean of final [e]’s

700 800 900

Figure 5.   Formant frequencies of speaker M1’s unstressed /e/’s in proparoxytones

4.  Interpretation of the results and analysis 4.1  Stressed vowels As seen in § 2.4, the evidence about absolute stressed vowel duration is not always consistent, at least when the alleged difference between paroxytones and proparoxytones is considered. Our data in § 3.2 are no exception. Three speakers have longer stressed vowels in paroxytones, whereas the other three have very similar



Stressed vowel duration and stress placement in Italian 

durations or even longer vowels in proparoxytones. Antepenultimate stressed vowels may be shorter than penultimate, but it is not a strong tendency, and there is a substantial amount of inter- and intra-speaker variation. Furthermore, our ratios (Figure 2, Table 2) indicate that also the immediately post-tonic vowel is slightly shorter in proparoxytones than in paroxytones. This means that whatever antepenultimate shortening may exist, it is counterbalanced by shortening of the following vowel. Therefore, invariance emerges once relative (rather than absolute) values of vowel duration are taken into account: the ratio between stressed vowel duration and post-tonic vowel duration is not affected by stress position. 4.2  Duration and centralization of post-tonic vowels The results reported in § 3.2.1 and 3.2.2 suggest that proparoxytones have a final secondary stress (not a new idea, as it dates back to Camilli (1965) at least). In terms of metrical grid this finding is represented as follows: (10) * * * * * * σ́́ σ σ̀

An alternative explanation exists which denies a ‘real’ secondary stress. According to this interpretation, this secondary stress is not an inherent property of proparoxytones. It rather derives from their interaction with the larger phrasal context, only occurring in order to avoid a lapse of three unstressed syllables if the first syllable of the next word is unstressed (Nespor 1993: 169). However, in our carrier sentence, proparoxytones (and paroxytones) are followed by a primary stress (the first syllable of the word piàno), thus the different durations of the final and penultimate vowels cannot be due to the phrasal context. The presence of a following word in the carrier sentence also prevents the vowels examined from receiving sentence-final vowel lengthening. Another way to explain the durational difference between the penultimate and final vowels of proparoxytones is to interpret it as the result of a shortening of the immediately post-tonic vowel, rather than to a lengthening of the final vowel. Indeed, such a phenomenon is widely attested in Italian dialects: in many of them the penultimate vowel of proparoxytones is reduced to [ә], [i] or [e], whereas /a, e, i, u/ or /a, e, i, o/ contrast word-finally in both paroxytones and proparoxytones; sometimes it completely assimilates to the following vowel, or underwent syncope (see Rohlfs (1966: 171–175) for examples). In our opinion this is not an alternative to our explanation; on the contrary, we believe it is precisely a consequence of the metrical structure we assume for

 Stefano Canalis & Luigia Garrapa

proparoxytones. Syncope, vowel reduction, phonological neutralization of vowel contrasts and complete assimilation to another vowel are typologically associated with weak, unstressed prosodic positions rather than secondary or even primary stress; the very fact that post-tonic vowel reduction in Italian dialects is restricted to, or more radical in, the penultimate vowel of proparoxytones than in word-final vowels suggests that the former is likely to be less prosodically prominent that the latter. This is entirely consistent with the metrical representation put forward in (10): it predicts that any vowel reduction occurring in unstressed syllables is expected to be more intense in the penultimate syllable of proparoxytones than in other positions. It may be useful to add that in some Italian dialects the vowel between a secondary stress and main stress (a mirror image of the prosodic environment in (10)) underwent syncope (Rohlfs 1966: 175); therefore vowel reduction between two prosodically stronger ­syllables is independently attested. 5.  Principles of stress assignment Understanding how vowel duration is realized in Italian not only sheds light on a problem of phonetics-phonology interface – the phonetic implementation of the abstract phonological category ‘word stress’. As seen in § 2.4, data on vowel duration are also relevant – and sometimes crucial – for several hypotheses about the metrical representation of Italian. For example, phonologists who assume that ­Italian is sensitive to syllabic quantity usually interpret lengthening in stressed open syllables as stemming from the need to form a bimoraic foot. Conversely, possible lack of lengthening in stressed open syllables poses a problem to the ­quantity-sensitivity hypothesis. Manipulated vowel stimuli with higher pitch but identical duration can still be perceived as stressed (§ 2.3) – only with less frequency than vowels having longer duration but identical pitch – and stressed syllables within words in nonprominent prosodic positions show rather weak lengthening (­Footnote 4); these data make it difficult to consider increased duration as a necessary property of stressed vowels in open syllables. More specific hypotheses about the phonology of Italian stress are also based on the behaviour of vowel duration; for example D’Imperio and Rosenthall (1999) see differences in stressed vowel duration as an important argument to assign different metrical representations to paroxytones and proparoxytones. With respect to vowels not carrying (primary) stress, our experiment suggests that the final vowel of proparoxytones is longer than the penultimate, but this result is unexpected in any analysis which represents the latter as the nucleus of a weak, completely unstressed syllable.



Stressed vowel duration and stress placement in Italian 

Since so many aspects of vowel duration have a direct bearing on whether ­Italian stress is quantity-sensitive or not, and on the metrical representation to adopt, it is worth examining what the other aspects of these issues are. In fact, vowel duration is but one of the areas where paroxytones and pro­ paroxytones differ. First, they seem to require two distinct metrical representations. As seen in § 2.4, D’Imperio and Rosenthall (1999) adopt moraic feet for paroxytones and syllabic feet for proparoxytones; Nespor (1993) and Bafile (1999) adopt syllabic binary feet for paroxytones and syllabic ternary feet for proparoxytones; den Os and Kager (1986) assume that the final syllable of pro­ paroxytones is lexically extrametrical. In general, it seems that whatever type of metrical representation fits one stress position, it cannot fit the other, because of the different number of syllables in the word. Second, it is unclear whether the default, unmarked stress position is penultimate or antepenultimate. Frequency is in favour of the former option (cf. § 2.1), but in some words stress has shifted from penultimate to antepenultimate (14). This problem also overlaps with the alleged role of syllable weight: if closed syllables count as heavy, (…) CVC.CV words should be paroxytones and (…) CV(C). CV.CV words should be proparoxytones. 5.1  Non-lexical stress and syllable weight In order to understand stress assignment in Italian, the status of syllable weight needs to be considered. However, given the lexical nature of Italian stress, it is difficult to test which principles determine stress position. Nevertheless, a testing ground to compare the predictions of different explanations is provided by loanwords, acronyms and brand names. Since these words do not arise via native morphological processes and are often first introduced through the written medium, it is safe to say that most of them do not have lexical stress. Despite the contrastive role of stress position in Italian, Italian orthography does not obligatorily mark stress (save for oxytones). This means that readers usually have no information about stress position when they encounter an unknown word while reading; it is reasonable to assume that this situation allows the phonological principles governing non-lexical stress assignment to surface. With respect to the role of syllable weight, it has already been pointed out by several authors (e.g. Bafile 1999; Marotta 1999) that proparoxytones with a penultimate CVC syllable, although on the whole rare in the lexicon (§ 2.1), are increasingly attested in the Italian pronunciation of foreign place names (11a to 11c), loanwords (11d), acronyms (11e) and brand names (11f to 11i). This fact is a counterargument to the claim that closed syllables count as heavy, and more broadly casts doubts on the presence of quantity-sensitivity in Italian.

 Stefano Canalis & Luigia Garrapa

(11) a. ['is.tam.bul] b. ['am.ster.dam] c. ['maɲ.ʧes.ter] d. ['ka:.ber.net] e. ['a:.duz.bef] f. ['nɔ:.ver.ka] g. ['ɛ:.ter.nit] h. ['fi:.nin.vest] i. ['pi:.kan.to]

Istanbul Amsterdam Manchester cabernet Adusbef Noverca Eternit Fininvest (Kia) Picanto

This pattern is all the more revealing when stress position is not the same as in the donor language (Istánbul, Amsterdám, cabernét). As for the native lexicon, it is uncontroversial that this pattern is rare, but rarity may be due to diachrony rather than to a synchronic restriction. The reasons can be summarised as follows: 1. An overwhelming part of the Italian lexicon comes from Latin. 2. In Latin, CVC (and CVV) penultimate syllables always received stress. 3. Italian has preserved Latin stress position in the overwhelming majority of words. As a logical consequence of points 1, 2 and 3, in Italian the overwhelming majority of words having a CVC penultimate syllable have penultimate stress – although these facts say nothing about the minority of words not coming from Latin,5 nor about synchronic constraints in modern Italian. This suggests that frequency alone is not very informative about Italian stress assignment, and novel words as in (11) to (13) may offer a more reliable indicator of active patterns. 5.2  Stress assignment and number of syllables Most words in (11) and (14) are trisyllabic. This seems to be more than accidental: both loanwords and the native lexicon display a tendency towards antepenultimate stress in trisyllabic words, in spite of the higher overall frequency of penultimate stress (and despite the frequent claim that that is the default stress position). As for trisyllabic loanwords and acronyms with an open penultimate syllable, they, too, tend to receive antepenultimate stress:6 5.  Indeed, several words in (5) were borrowed from Greek (polizza, acanto, Lepanto, etc.), even though they entered Italian many centuries ago and are no longer perceived as loanwords. 6.  In the case of loanwords the ‘correct’ original stress position (as Gorbachév or Bengási) may be adopted by, for example, TV speakers, but antepenultimate stress is attested in e­ veryday speech.



(12) a. ['mar.ke.tiŋg] b. ['kɔ:.da.kons] c. ['kɔ:me.kon] d. ['bɛ:.ne.luks] e. ['gɔr.ba.ʧov] f. ['bɛŋ.ga.zi]

Stressed vowel duration and stress placement in Italian 

marketing CODACONS (acronym) COMECON (acronym) Benelux (acronym) Gorbaciov Bengasi

On the contrary, quadrisyllables usually preserve (13a to 13f), adopt (13g to 13h), or even change to (13i to 13j), penultimate stress. (13) a. [sin.sin.'na:.ti] Cincinnati b. [fi.la.'dɛl.fja] Philadelphia c. [ko.pe.'na:.gen] Copenhagen d. [ku.ro.'sa:.wa] Ku.ro.sa.wa e. [gor.ba.'ʧɔ:.va] (Raisa) Gorbaciova f. [so.li.'dar.noʃ] Solidarnosc g. [a.di.'kɔn.sum] Adiconsum (acronym) h. [e.na.'zar.ko] Enasarco (acronym) i. [to.po.'la:.nek] Topolanek (Czech politician; Czech Tópolanek) j. [ʃa.ra.'pɔ:.va] Sharapova (Russian Sharápova)

Similar tendencies can be observed in native trisyllables and quadrisyllables. Several trisyllables display variation between penultimate and antepenultimate stress position, the former being the etymological, standard pronunciation and the latter being a common non-standard variant (where ‘non-standard’ mainly refers to the diastratic rather than diaphasic or diatopic dimension: the non-standard form is more likely among less educated speakers).7 This tendency also includes words having a CVC penultimate syllable (14k to 14l). (14) Standard pronunciation Non-standard pronunciation a. i.cò.na ì.co.na ‘icon’ b. li.bì.do lì.bi.do ‘libido’ c. mas.cà.ra màs.ca.ra ‘mascara’ d. mol.lì.ca mòl.li.ca ‘breadcrumb’ e. pu.dì.co pù.di.co ‘modest’ f. ru.brì.ca rù.bri.ca ‘notebook’ g. scor.bù.to scòr.bu.to ‘scurvy’

7.  Most of the information about what the standard and non-standard variants are comes from Canepari (1999) and the DOP, two pronunciation dictionaries of Italian. Other examples come from the authors’ personal experience; more anecdotal evidence may also come from the Internet, as shown by forum debates over the pronunciation of e.g. Bengasi (http://usenet. it.rooar.com/showthread.php?t=2552876),  cabernet  (http://www.cookaround.com/yabbse1/ archive/index.php/t-34163.html), Agusta (http://forum.mitoclub.com/topic/18637-mv-agusta/).

 Stefano Canalis & Luigia Garrapa

h. ta.fà.no tà.fa.no ‘gadfly’ i. cu.cù.lo cù.cu.lo ‘cuckoo’ j sa.lù.bre sà.lu.bre ‘healthy’ k. ca.rìs.ma cà.ris.ma ‘charisma’ l. A.gùs.ta À.gus.ta surname8

In a few cases the proparoxytonic variant has completely suppressed the other possibility; for instance the words rècluta ‘recruit’ and èsile ‘feeble/slender’ are now proparoxytone for all speakers, but were reclùta (Rohlfs 1966: 106) and esìle (DOP: sub voce) until the nineteenth century. In other cases dictionaries still report the paroxytone variant as the ‘correct’ one, but acknowledge that it has almost completely fallen out of use at the expense of the proparoxytone variant (as vàluto ‘I evaluate’ rather than valùto). Additionally, the tendency to stress trisyllables on the initial syllable seems to give rise to a small number of stress alternations: when a one-syllable prefix is added to some disyllabic words, stress moves to the prefix (prefixes are never lexically stressed in Italian; therefore, the alternations in (15) cannot be due to lexical specification of the prefix). (15) a. fìdo ‘trustworthy’ ìnfido ‘untrustworthy’ (along with infìdo) b. pàri ‘even’ dìspari ‘odd’

The tendency towards stressing trisyllables on the initial rather than the second syllable is no more than that – a tendency, not an exceptionless pattern. Examples of the opposite change do exist (16); nevertheless, they are less frequent, and several cases may be ascribed to analogical effects (for instance the much higher frequency of words ending in [ˈite] than in unstressed [ite] might favour termìte over tèrmite (16d)). (16) Standard pronunciation Non-standard pronunciation a. dàr.se.na dar.sè.na ‘dock’ b. gó.me.na go.mè.na ‘cable’ c. lù.bri.co lu.brì.co ‘lewd’ d. tèr.mi.te ter.mì.te ‘termite’

Native quadrisyllables display less fluctuation in stress position. Sometimes penultimate stress is the etymologically motivated option, while antepenultimate stress is the popular pronunciation (17); but at least as frequently, stress position changes from antepenultimate to penultimate (18).

8.  Also the name of a helicopter manufacturing company and a motorcycle manufacturing company.



Stressed vowel duration and stress placement in Italian 

(17) Standard pronunciation Non-standard pronunciation a. al.ca.lì.no al.cà.li.no ‘alkaline’ b. scan.di.nà.vo scan.dì.na.vo ‘Scandinavian’ c. per.sua.dé.re per.suà.de.re ‘to persuade’ (18) Standard pronunciation Non-standard pronunciation a. a.nò.di.no a.no.dì.no ‘ineffective’ b. ve.lò.dro.mo ve.lo.drò.mo ‘cycling track’ c. ip.pò.dro.mo ip.po.drò.mo ‘race track’ d. cal.lì.fu.go cal.li.fù.go ‘corn plaster’

5.2.1  Further data on stress shift The examples in (14) to (18) are among the most common and well-known examples of stress shift. To test more formally whether the number of syllables actually influences stress placement, we created a corpus of words with stress variation. We selected all words from letters A to C in Canepari (1999) that have at least three syllables and show variable stress, excluding proper names (surnames, place names, brand names, etc.). Since our hypothesis assumes the number of syllables to be the independent variable, we discarded all variation in stress position that is accompanied by a change in the number of syllables.9 It is not always obvious which of two variants is the standard. After reporting what he deems to be the most neutral variant of a word, Canepari (1999) classifies less neutral variants as either ‘traditional’, ‘almost equally acceptable’, ‘tolerated’, ‘careless’, ‘intentionally displaying one’s knowledge’ and ‘elevated’. Many words in the second category are learned words of Ancient Greek origin, and the traditional explanation for the acceptability of both stress patterns is that one of them follows the original Greek accent, while the other results from the Latin adaptation of the Greek term. This is true of many words in the sixth category (‘elevated’) too, which also include stress variants attested only in poetry. Since it is unclear whether an unambiguous standard exists for these words, they were discarded, while the other four classes were used to build our list of stress variants. Table 4 shows that the number of syllables in a word influences non-­standard stress placement, confirming the tendency reported above: among words with

9.  This may arise for two reasons: (1) an ambiguity in Italian spelling conventions, which represent both vowels [i, u] and glides [j, w] as 〈i〉 and 〈u〉 (this means that e.g. leccornia ‘delicacy’ may be read as standard [lek.kor.'ni.a] – quadrisyllabic and stressed on [i] – or as non-standard [lek.'kɔr.nja] – trisyllabic and stressed on [ɔ]); (2) the variable realization of /i , u / as either [i , u ] or [j , w ] – this means that e.g. diuresi  ‘diuresis’ can be realized as [di.u.'rɛ.zi] or [di.'u.re.zi] – quadrisyllabic – but also as [dju. 'rɛ.zi] or ['dju.re.zi] – trisyllabic).

 Stefano Canalis & Luigia Garrapa

variation in stress position, trisyllables become proparoxytones more often than paroxytones (57 vs. 31), while four-syllable words exhibit a weak preference for penultimate stress (30 paroxytones vs. 21 proparoxytones). The difference between trisyllables and quadrisyllables is statistically significant (χ2 = 6.3734, df = 1, p = 0.01158). Table 4.  Word length

Word becomes paroxytone Word becomes proparoxytone

3 syllables

4 syllables

5 syllables

6 syllables

31 57

30 21

5 6

0 2

5.3  Provisional conclusions and discussion of previous explanations Examining the evidence provided by loanwords, acronyms and sociolinguistic variation, stress position appears not to be influenced by syllabic weight; even words having a penultimate closed syllable – like Istanbul or carisma – may be proparoxytones.10 On the other hand, the number of syllables in a word seems to play a role, at least as a tendency; trisyllables favour antepenultimate stress, while quadrisyllables slightly favour penultimate. Words in Table 4 with more than four syllables are too few to detect any tendency, but – also considering that the overwhelmingly majority of disyllabic loanwords and acronyms receives penultimate stress rather than final – it may be inferred that an odd number of syllables makes antepenultimate stress likely, while an even number favours penultimate stress. Alternative accounts for the data reported in § 5.1–5.2 exist. A traditional explanation for the tendency towards proparoxytonic stress appeals to non-­ phonological factors: words judged by speakers as not fully native (loanwords, but

10.  Interestingly, recent research on stress assignment in Spanish (a language which has frequently been argued to be sensitive to syllable quantity) has reached similar conclusions. Bárkányi (2002) presented speakers of Spanish with written sentences that contained trisyllabic nonce words; when they were asked to indicate where the word would be stressed were it a Spanish word, their answers frequently did not conform to the hypothesis of quantity sensitivity. She concludes that quantity sensitivity is a historical heritage rather than an active rule or constraint. In the experiment performed by Alvord (2003) nonce words with antepenultimate stress and a closed penultimate syllable were overwhelmingly accepted by Spanish speakers. To mention another similarity, a variation roughly similar to that exemplified in (14) is also attested in Spanish (e.g. Spanish non-standard ícona vs. standard icóna).



Stressed vowel duration and stress placement in Italian 

also learned words and acronyms) would form a special subclass and receive antepenultimate stress. This would be due either to hypercorrection in case of learned words (Rohlfs 1966: 441),11 or to a minor stress rule for non-native or otherwise unusual words (Loporcaro & Bertinetto 2005: 143) in case of loanwords, brand names and acronyms. Several facts support this hypothesis. Many learned words of Latin origin are proparoxytones, and the influence of Ancient Greek (§ 5.2.1) is usually invoked to justify antepenultimate stress; in fact, many of the proparoxytones in Table 4 are learned words. It is therefore plausible that a tendency exists to assign antepenultimate stress to unknown words felt as learned or exotic. However, the two hypotheses – ‘learnedness’ of antepenultimate stress, and role of the number of syllables in a word – are not mutually exclusive, and the former alone cannot explain all the data. Firstly, words preferably receiving antepenultimate stress form a rather heterogeneous class, some of them clearly belonging to the core lexicon (like mollica ‘breadcrumb’ (14d)). Secondly (and crucially), it does not explain why antepenultimate stress is more frequent in trisyllabic than in four-syllabic words (the number of syllables obviously being a purely phonological factor). For example, ‘correct’ Bengàsi (which has an Italian-like phonotactics) may be realized as Bèngasi, whereas Copenhagen (a phonotactically ‘exotic’ word for Italian ­speakers, having a word-final consonant) always preserves penultimate stress (12f and 13c respectively). Interestingly, the two joined hypotheses can explain why even-syllabled words overwhelmingly favour penultimate stress if disyllabic, but only slightly so if quadrisyllabic: in the latter case both tendencies are at work, and since their goals are conflicting they cancel each other out. Another potential counterargument to our conclusion that the number of syllables in a word plays a role comes from the nonce word test in Krämer (2009). In this test 12 Italian speakers were asked to read several nonce words (without indication of stress position) in a carrier sentence. Penultimate stress was more frequent in four-syllable words than in trisyllables, but ratios were quite different from ours. Nonce words made of four light syllables (LLLL) were nearly always (91.7 percent) realized as paroxytones, and penultimate stress was the most frequent choice also in trisyllables, especially when the second-tolast syllable was closed (this result was interpreted by Krämer (2009) as a case for weight-­sensitivity): 100 percent in LHL words, 71 percent in HLL words, and 55.1 ­percent in LLL words.

11.  A very recent discussion of the role of learnedness as a cue for antepenultimate stress (and of several others aspects of Italian stress) can be found in Hayes (2012).

 Stefano Canalis & Luigia Garrapa

The lower ratio of paroxytones reported in § 5.2.1 (58.8 percent for quadrisyllables and 35.2 percent for trisyllables) is probably to be ascribed to the tendency to stress learned words as proparoxytones. While many words taken from Canepari (1999) are learned, and therefore subject to this influence, Krämer’s nonce words have no meaning (by definition) and bear no obvious phonological resemblance to any existing Italian learned word, therefore are less likely to be treated as such by speakers. With regard to two other results reported by Krämer (2009) – the preference for penultimate stress also in trisyllables and the role of weight-sensitivity – they apparently conflict with our conclusions that trisyllables favour antepenultimate stress and syllable weight is no longer relevant. In our opinion, the difference between stress assignment in Krämer’s nonce word test and loanwords/­acronyms may be due to the role of word similarity. Other experiments by Giraudo and ­Montermini (2010) have shown that stress assignment in Italian-like nonce words is not dependent on purely prosodic conditions alone. Segmentally and morphologically similar words influenced the choice of stress placement in their nonwords, even when the words were prosodically identical (i.e. same number of syllables, same distribution of open and closed syllables). Their findings suggest that, since words with antepenultimate stress and a closed penultimate syllable are rare in the lexicon, it is unlikely that analogical stress assignment is based on them. Another similarity effect, not discussed by Giraudo and Montermini (2010), may be caused by rhyme; when asked to read an unknown word, speakers could be influenced by words in the lexicon that can rhyme with it. To make an example, one of the HLL nonce words used by Krämer (2009) is frampeco; a search in Canepari (1999) for nouns12 longer than two syllables and ending in /eko/ or /ɛko/ reveals that none of them is a proparoxytone. They include copèco ‘copeck’, metèco ‘metic’, trichèco ‘walrus’, cercopitèco ‘guenon’ and related words (antropopitèco, paleopitèco), names of pre-Columbian civilizations (aztèco, miztèco, olmèco, z­ apotèco) and brand names (Corèco, Ivèco). If the other word classes are ­considered, proparoxytones are attested (as intrìnseco ‘intrinsic’), but they are outnumbered by ­paroxytones (e.g. accèco ‘I blind’, arrèco ‘I cause’, imprèco ‘I swear’, etc.). On the contrary, segmental and morphological similarity – and, more generally, any kind of analogical relationship – can be less easily inferred from loanwords and acronyms. Paradoxically, they may well be a better testing ground than

12.  Because words in Krämer’s (2009) experiment were presented as nouns, within the carrier sentence Ho visto due ____ “I saw two ____” .



Stressed vowel duration and stress placement in Italian 

existing Italian words or Italian-like non-words for investigating stress assignment, as they minimize the role of similarity effects. 6.  Feet and stress assignment in Italian The combined information provided by the data about vowel duration (§ 3 and 4) and stress placement (§ 5) can help understand the foot inventory of Italian and the principles governing stress assignment. 6.1  Foot types The above results support a specific interpretation of the metrical structure of Italian. The behaviour of stress in loanwords, acronyms and non-standard pronunciations suggests that feet are only syllabic; moraic feet are ruled out by weightinsensitivity of CVC syllables (§ 5.3). There is also no reason to postulate either moraic or syllabic feet depending on vowel duration, because (relative) vowel duration is constant across paroxytones and proparoxytones (§ 4.1). The only remaining candidate is the syllabic foot. It is rarely disputed that the basic rhythm of Italian is trochaic rather than iambic.13 Paroxytones can thus be represented with the syllabic trochee (19). (19)    *    * * (σ́ σ)

However, the status of the final syllable of proparoxytones still remains unaccounted for. Assuming that it is extrametrical (den Os & Kager 1986) does not explain why its vowel is longer than the preceding one. In this respect, it may be useful to point out that stressed monosyllables are licit words in Italian (20). (20) a. gru ‘crane’ b. re ‘king’ c. qui ‘here’ d. (io) so ‘(I) know’ e. (io) do ‘(I) give’

14.  But cf. Marotta (1999: 108–11) for arguments for the existence of iambic feet in Italian.

 Stefano Canalis & Luigia Garrapa

This means that any account of Italian metrical structure has to accept the degenerate (i.e. monosyllabic) foot in its foot inventory (as already pointed out by Nespor (1993), Marotta (1999)). If the ratio of the stressed vowel to its next vowel is the same across paroxytones and proparoxytones, given the representation in (19) above for paroxytones, the most obvious solution for proparoxytones is to assume that they have an ­identical binary syllabic foot. As for the lengthening of the final vowel in proparoxytones, this prominence is captured if the vowel is assumed to form a (­degenerate) foot of its own. If proparoxytones are represented as made of a trochaic foot plus a degenerate foot, both their parallelism with paroxytones (because of the same duration ratio) and their final secondary stress follow (21): (21)  *  *  *  * *  * (σ́́ σ) (σ̀)

6.2  Non-lexical stress The assumption that only binary and degenerate syllabic trochees exist in Italian has an additional, interesting implication: stress position in words without lexical stress can be given a new explanation. One of the most puzzling aspects of stress assignment is the preference for antepenultimate stress in odd-syllabled words. Several facts point at penultimate stress as the most basic position: frequency, but also the binary foot obviously being less marked than the unary foot (which we assume to be present in proparoxytones). Antepenultimate stress in odd-syllabled words can be seen as the result of a conflict between two demands: one is that all syllables be parsed into feet, and the other is that there be only binary feet. A trisyllabic proparoxytone, as in (21) above, has one degenerate foot, but parses all syllables; if it were a paroxytone it would leave the initial syllable unparsed. 7.  Conclusions and further directions Our experimental results suggest that penultimate and antepenultimate stressed vowels have the ‘same’ duration in Italian. This is often true with respect to absolute values, and even more so with respect to relative values: the ratio between stressed vowel duration and post-tonic vowel duration is not affected by stress position. The experimental data also indicate that a final secondary stress is present in proparoxytones.



Stressed vowel duration and stress placement in Italian 

Moreover, the status of quantity-sensitivity has been scrutinized, arguing (on the basis of stress placement in loanwords, acronyms and non-standard pronunciations) that it is no longer active. Taken together, the data on vowel duration and on stress placement suggest that feet in Italian are always syllabic in both paroxytones and proparoxytones, with unary feet in the final syllable of proparoxytones (and in oxytones). A tendency to realize odd-syllabled words as proparoxytones emerged. This stress pattern can be accounted for as the result of the interaction between two constraints: (1) a preference for binary feet over degenerate feet, (2) and the need to parse all syllables – at the cost of making the final syllable a degenerate foot, which is manifested by the secondary stress found in the experiment. Our study leaves open several questions for future research. First, the results are preliminary, and further experiments and data are needed to corroborate them. Second, the results show the number of syllables plays a role in stress assignment, but as a tendency rather than an exceptionless regularity. However, the issue of how to formally represent its probabilistic nature within a model of phonological variation was left unexplored. Finally, the focus of our paper is on the phonetic and phonological properties of Italian stress, but stress assignment seems to be influenced by non-phonological factors as well. They plausibly include at least analogy, morphology, and learnedness.

References Albano Leoni, Federico, Caputo, Maria Rosaria, Cerrato, Loredana, Cutugno, Francesco, Maturi, Pietro & Savy, Renata. 1995. “Il vocalismo dell’Italiano: analisi di un campione televisivo.” Studi Italiani di Linguistica Teorica e Applicata 24 (2): 405–411. Alfano, Iolanda, Savy, Renata & Llisterri, Joaquim. 2009. “Sulla realtà acustica dell’accento lessicale in italiano ed in spagnolo: La durata vocalica in produzione e percezione.” In AISV 2007. La fonetica sperimentale. Metodo e applicazioni. Atti del 4o Convegno Nazionale AISV, Luciano Romito, Vincenzo Galatà & Rosita Lio (eds), 22–39 [CD-ROM]. Torriana: EDK Editore. Alvord, Scott M. 2003. “The Psychological Unreality of Quantity Sensitivity in Spanish: Experimental Evidence”. Southwest Journal of Linguistics 22 (2): 1–12. Bafile, Laura. 1999. “Antepenultimate Stress in Italian and Some Related Dialects: Metrical and Prosodic Aspects”. Rivista di Linguistica 11 (2): 201–229. Bárkányi, Zsuzsanna. 2002. “A Fresh Look at Quantity Sensitivity in Spanish”. Linguistics 40 (2): 375–394. Bertinetto, Pier Marco. 1981. Strutture prosodiche dell’italiano. Firenze: Accademia della Crusca. Bertinetto, Pier Marco. 2010. “Fonetica italiana”. In Enciclopedia dell’Italiano, Raffaele Simone (ed), vol. I, 478–494. Roma: Istituto della Enciclopedia Italiana. Boersma, Paul & Weenink, David. 2010. Praat: Doing Phonetics by Computer [Computer program]. Version 5.1.44, retrieved from http://www.praat.org/.

 Stefano Canalis & Luigia Garrapa Camilli, Amerindo. 1965. Pronuncia e grafia dell’italiano. Third edition. Firenze: Sansoni. Canepari, Luciano. 1977. Introduzione alla fonetica. Torino: Einaudi. Canepari, Luciano. 1999. DiPI. Dizionario di Pronuncia Italiana. Bologna: Zanichelli. D’Imperio, Mariapaola & Rosenthall, Sam. 1999. “Phonetics and Phonology of Italian Main Stress”. Phonology 16 (1): 1–28. DOP = Migliorini, Bruno, Tagliavini, Carlo & Fiorelli, Piero. 2010. Dizionario italiano multimediale e multilingue d’ortografia e di pronunzia. Third edition revised, updated and ­augmented by Piero Fiorelli and Tommaso Francesco Borri, 2 vols. Roma: Rai Eri. Also available at http://www.dizionario.rai.it/. Farnetani, Edda & Kori, Shiro. 1990. “Rhythmic Structure in Italian Noun Phrases: A Study on Vowel Duration”. Phonetica 47 (1–2): 50–65. Hayes, Bruce. 1995. Metrical Stress Theory: Principles and Case Studies. Chicago: Chicago ­University Press. Hayes, Bruce. 2012. “How Predictable is Italian Word Stress?” Handout of a talk given at National Chiao Tung University, May 11, 2012. Krämer, Martin. 2009. The Phonology of Italian. Oxford: Oxford University Press. Giraudo, Hélène & Montermini, Fabio. 2010. “Primary Stress in Italian: Linguistic and Experimental Issues”. Lingue e linguaggio 9 (2): 112–129. Hajek, John, Stevens, Mary & Webster, Georgia. 2007. “Vowel Duration, Compression and Lengthening in Stressed Syllables in Italian”. Proceedings of the XVI International Congress of Phonetic Sciences. Saarbrücken. 1057–1060. Landi, R. & Savy, Renata. 1996. “Durata vocalica, struttura sillabica e velocità d’eloquio nel parlato connesso”. In Atti del XXIV Convegno Nazionale dell’Associazione Italiana di Acustica, Alessandro Peretti & Paolo Simonetti (eds.), 65–70. Padova: Arti Grafiche Padovane. Lehiste, Ilse. 1970. Suprasegmentals. Cambridge, Mass.: MIT Press. Loporcaro, Michele & Bertinetto, Pier Marco. 2005. “The sound pattern of Standard Italian, as compared with the varieties spoken in Florence”. Journal of the International Phonetic Association 35 (2): 131–151. Marotta, Giovanna. 1985. Modelli e misure ritmiche. La durata vocalica in italiano. Bologna: Zanichelli. Marotta, Giovanna. 1999. “Degenerate Feet nella fonologia metrica dell’italiano”. In Fonologia e morfologia dell’italiano e dei dialetti d’Italia. Atti del XXXI Congresso della Società di Linguistica Italiana, Paola Benincà, Alberto Mioni & Laura Vanelli (eds), 97–116. Roma: Bulzoni. Nespor, Marina. 1993. Fonologia. Bologna: Il Mulino. Nespor, Marina & Vogel, Irene. 1986. Prosodic Phonology. Dordrecht: Foris. Pickett, Emily R., Blumstein, Sheila E. & Burton, Martha W. 1999. “Effects of Speaking Rate on the Singleton/Geminate Consonant Contrast in Italian”. Phonetica 56 (3–4): 135–157. den Os, Els & Kager, René. 1986. “Extrametricality and Stress in Spanish and Italian”. Lingua 69 (1–2): 23–48. Rohlfs, Gerhard. 1966. Grammatica storica della lingua italiana e dei suoi dialetti. Fonetica. Torino: Einaudi. Revised and updated translation of Historische Grammatik der italienischen Sprache und ihrer Mundarten. Lautlehre. Bern: Francke. van Santen, Jan & D’Imperio, Mariapaola. 1999. “Positional Effects on Stressed Vowel Duration in Standard Italian.” Proceedings of the XIV International Congress of Phonetic Sciences. San Francisco. 241–244.



Stressed vowel duration and stress placement in Italian 

Thornton, Anna Maria, Iacobini, Claudio & Burani, Cristina. 1997. BDVdB. Una base di dati sul vocabolario di base della lingua italiana. Second edition. Roma: Bulzoni. Vayra, Mario. 1994. “Phonetic Explanations in Phonology: Laryngealization as the Case for Glottal Stops in Italian Word-Final Stressed Syllables”. In Phonologica 1992: Proceedings of the 7th International Phonology Meeting, Wolfgang U. Dressler, Martin Prinzhorn & John R. Rennison (eds), 275–294. Torino: Rosenberg & Sellier. van der Veer, Bart. 2006. The Italian ‘Mobile Diphthongs’: A Test Case for Experimental Phonetics and Phonological Theory. Utrecht: LOT.

Serial prosodification and voiced stop geminates in Catalan* Francesc Torres-Tamarit & Clàudia Pons-Moll Meertens Instituut/KNAW / Universitat de Barcelona

1.  Introduction The main purpose of this paper is to show how the process of voiced stop gemination that applies in Central Catalan finds a straightforward explanation in Harmonic Serialism. In Catalan, root-final clusters involving a labial or a velar voiced stop followed by an alveolar lateral (/bl/, /ɡl/)1 surface as geminates ([b.bl], [ɡ.ɡl]) (Bermúdez-Otero 2000; Bonet & Lloret 1998; Fabra 1912/1982; ­Mascaró 1987, 2003; Recasens 1991, 1993; Colina 1995; Jiménez 1997; Wheeler 1979, 1980, 2005; Pons-Moll 2004, 2008, 2011). Otherwise, if these clusters precede a vowel belonging to the root, voiced stops spirantize and the cluster is parsed as a complex onset ([.βlV], [.ɣlV]). It is argued that gemination is only triggered when the voiced stop is syllabified in coda position (Mascaró 1987), and in order to fix an ill-formed rising sonority intersyllabic contact (Colina 1995; Jiménez 1997; Pons-Moll 2004, 2008). The insertion of an epenthetic schwa or the presence of a vowel-initial derivational or inflectional suffix do not block gemination, although the presence of these vowels make up the phonological context that could bleed the application of the gemination process, that is, the voiced stop syllabified as the first element of a complex onset. In order to explain these facts, this paper develops a theory of serial syllabification in H ­ armonic Serialism based on Elfner (2009) and suggests that the binary operation core syllabification can create complex degenerate syllables and cannot operate with two adjacent segments if one of them, but not the other, is integrated into a prosodic category higher than the syllable. This means that prosodic categories *  This research has been supported by a Rubicon postdoctoral fellowship from the Netherlands Organisation for Scientific Research awarded to the first author (446-11-022). We are grateful to Ricardo Bermúdez-Otero, Joan Mascaró, John J. McCarthy, the editors of this volume, and four anonymous reviewers for helpful comments on previous versions of this paper. 1.  Catalan, as many other Romance languages, does not allow tautosyllabic [dl] and [tl] clusters syllabified in onset position.

 Francesc Torres-Tamarit & Clàudia Pons-Moll

create opaque domains for syllabification. This assumption, together with serial prosodification, guarantees that vowels outside the root, either epenthetic or inflectional, are not available for syllabification purposes when the root is first syllabified. This paper is organized as follows. The data is presented in § 2. The theoretical background on which the analysis is based is presented in § 3; this section includes both a brief explanation of Harmonic Serialism and prosodification (§ 3.1), and presents a theory of serial syllabification in Harmonic Serialism with some implications for the prosody-morphology interface (§ 3.2). The Harmonic Serialism analysis is presented in § 4. General conclusions are drawn in § 5.

2.  Data In Central Catalan, voiced stops in underlying /bl/ and /ɡl/ clusters undergo a process of gemination provided that those clusters are root-final, as in (1).2 Note that the examples in (1) surface with a peripheral schwa because tautosyllabic coda clusters with a flat or rising sonority profile are prohibited in Catalan and repaired through schwa epenthesis, with the exception of clusters in which the second consonant is /s/.3 (1) /pɔbl/ [�pɔb.bl6] ‘town’ /dobl/ [�dob.bl6] ‘double’ /pusibl/ [pu.�sib.bl6] ‘possible’ /seɡl/ [�seɡ.ɡl6] ‘century’

2.  Mascaró (1987) points out that these alternations are allomorphic in nature: [di.ˈab. bl6] ‘devil’ ~ [di.6.ˈβɔ.lik] ‘devilish’; [ˈmɔb.bl6] ‘piece of furniture’ ~ [mu.βi.ˈlja.ɾi] ‘furniture’; [ˈnɔb.bl6] ‘noble’ ~ [nu.βi.ˈlja.ɾi] ‘relative to nobility’; [bu.ˈlub.bl6] ‘voluble’ ~ [bu.lu.βi.li.ˈtat] ­‘volubility’. 3.  Following other scholars, the schwa in (1) is treated as an epenthetic vowel. Its appearance is easily explained as a fixing strategy to an otherwise unsyllabifiable rising-sonority cluster of consonants. It is true that this schwa is also found in words such as [ˈɔm+6] ‘man’, where positing epenthesis would be unmotivated, but positing a process of epenthesis in (1) is supported by the fact that the unmarked masculine morph in Catalan is a zero morph ([ˈdob.bl6] double.masc and [dub.ˈbl+a] ‘to double’). The schwa in (2), however, corresponds to the unmarked feminine morph in Catalan.



Serial prosodification and voiced stop geminates in Catalan 

The presence of an inflectional suffix such as the feminine morph does not block gemination, because the consonantal cluster is actually root-final.4 (2) /reɡl+6/ [�reɡ.ɡl6] ‘rule’ /kobl+6/ [�kob.bl6] ‘stanza’

Elsewhere, that is, when those clusters are not root-final, voiced stops undergo spirantization and they are syllabified as the first element of a complex onset. These root-internal clusters are always followed by a vowel belonging to the root. (3) /ɛɡluɡ+6/ [�ɛ.ɣlu.ɣ6] ‘eclogue’ /pɾublɛm+6/ [pɾu.�βlɛ.m6] ‘problem’ /ublid+a+ɾ/ [u.βli.ˈða] ‘to forget’ /publi/ [�pu.βli] ‘Publius’

When the second element of the cluster is a flap (/bɾ/, /ɡɾ/), voiced stops do not geminate, but they spirantize. (4) /pɔbɾ/ [�pɔ.βɾ6] ‘poor’ /aɡɾ/ [�a.ɣɾ6] ‘sour’

Although the aforementioned data correspond to the general pattern described for Central Catalan, /bl/ and /ɡl/ clusters can also be subject to a process of devoicing and be syllabified in onset position ([ˈpɔ.pl6]) or undergo a two-step process of devoicing and gemination ([ˈpɔp.pl6]) (Mascaró 1976). Other dialects such as Majorcan Catalan seem to have generalized gemination in those contexts in which the cluster is not root-final (Mascaró p.c.), and Western Catalan completely lacks geminates of that type. It is not the purpose of this paper, however, to address all this dialectal variation.

3.  Theoretical background 3.1  Harmonic Serialism and prosodification Harmonic Serialism (Prince & Smolensky 1993/2004; Elfner 2009, to appear; Jesney to appear; Kimper 2011; McCarthy 2000, 2007a, b, 2008a, b, 2010a, b, c, 2012, Pruitt 2010) is a non-stratal derivational version of Optimality Theory

4.  Voiced stop geminates are found in all kinds of derivatives, in both derivational and inflectional contexts: [ˈpub.bl+ik] ‘public’, [dub.bl+6.ˈɣ+a] ‘to fold’, [6.#r6ɡ.ˈɡl+a] ‘to fix’.

 Francesc Torres-Tamarit & Clàudia Pons-Moll

(Prince & Smolensky 1993/2004). In Harmonic Serialism Gen is restrained by a gradualness condition on candidate generation by which candidates only introduce one single modification with respect to the (latest) input, until convergence on the fully faithful candidate is reached (i.e. no further harmonic improvement is possible). An inescapable consequence of gradualness is the need for a Gen → Eval → Gen … loop, given that output forms are often the result of applying more than one phonological operation. In Harmonic Serialism Eval imposes the same constraint hierarchy at every step of the derivation. The perdurability of the constraint hierarchy in Harmonic Serialism contrasts with Stratal OT (Kiparsky 2000; Bermúdez-Otero to appear), in which the three recognized levels of phonological evaluation (stem, word and phrase) show a different ranking of the constraint set. Defining gradualness, that is, exploring what it means to introduce a single phonological operation at a time, is one of the main research interests in Harmonic Serialism. With regard to syllabification, two different views have been discussed so far in the literature within Harmonic Serialism. If syllabification in tautomorphemic sequences is never contrastive, then faithfulness constraints on syllabification must be excluded from the theory of Con. If gradualness is defined in terms of faithfulness, then syllabification freely applies with other phonological operations (McCarthy 2008b, Elfner to appear, Pruitt 2010). The other view on syllabification departs from an operation-based definition of gradualness. In an operation-based definition of gradualness, not only feature-changing operations correspond to a single step in the derivation, but also structure-building operations, such as syllabification, footing, or parsing a lexical word into a prosodic word. In this perspective, syllabification is also subject to the gradualness requirement on Gen (Jesney to appear, Pater 2012; Elfner 2009). This paper shares the latter view, in which Gen performs prosody-building operations, including syllabification, in a stepwise manner. If prosodification is subject to the gradualness requirement on Gen, then prosody-building operations cannot co-occur with other prosody-building or structure-changing operations. 3.2  A theory of serial syllabification in Harmonic Serialism This subsection develops a theory of serial syllabification in Harmonic Serialism along the lines of Elfner (2009). In Elfner (2009), a set of three syllable formation operations are proposed: (a) project syllable, which creates a syllable (X) from a segment X, where X can be either moraic (Xμ) or not (X); (b) adjunction, which takes a segment X and adjoins it to a syllable (Yμ) or (Y) to the left or the



Serial prosodification and voiced stop geminates in Catalan 

right, where X can be either moraic (Xμ) or not (X);5 and (c) core syllabification which builds a binary syllable (XYμ), where Y is a moraic syllable head and X its dependent.6 The proposal made in this paper maintains this small set of syllable-building operations, but proposes to modify the formulation of core syllabification, as shown in (5). (5) Core syllabification (modified Elfner’s 2009 version) From two segments X and Y, create a binary syllable (XY(μ)), where Y is either a moraic head or a non-moraic dependent of the syllable node, and X is a non-moraic dependent of the syllable node.7

This new formulation of core syllabification has two advantages. The first one is that core syllabification is parallel to the operation project syllable because the second parsed segments can be either moraic or not. The second advantage is that it allows the creation of a complex minor syllable if the second segment is a non-moraic segment. This type of configuration, along the lines

5.  An anonymous reviewer interestingly points out that the operation of adjunction allows for moraic onsets. Typologically, moraic onsets do not seem to exist (but see Topintzi 2006). The fact that moraic onsets can be generated given the existence of the adjunction of a moraic consonant to the left of an already syllabified moraic nucleus does not necessarily mean that a Harmonic Serialist grammar predicts moraic onsets if there are universal constraints against them. The operation of adjunction is thus not precisely an instance of the duplication problem. Con should be enough to discard them. 6.  Elfner (2009) argues that core syllabification is necessary in order to discard unattested stress assignment patterns. Her argument is as follows. Imagine a ranking in which ParseSegment dominates Onset and Onset dominates No-Coda. Without core syllabification, an input like /pata/ would be mapped as (pat)(a). Although it is true that a later derivational step would be able to resyllabify the coda as the onset of the following syllable, HS would be able to predict a language where the placement of stress is sensitive to the presence of onset consonants if stress assignment precedes resyllabification. For instance, in a hypothetical language with final stress except in the presence of a heavy syllable, and with the ranking ParseSegment » Onset » No-Coda, /pata/ might be stressed as (pá)(ta) because stress would be assigned to the intermediate form (pát)(a), and /paa/ might be stressed as (pa)(á). Elfner points out that a stress system like this does not seem to occur. However, further investigations about the duplication problem, which is inherent in a constrained-based model of grammar that incorporates a finite set of operations, are of priority. 7.  Notice that in the new formulation of the constraint core syllabification, the first segment X is always a non-moraic dependent of the syllable node. This means that core syllabification cannot create a (VμCμ) syllable at once. This configuration emerges derivationally as the result of applying sequentially project syllable and then adjunction.

 Francesc Torres-Tamarit & Clàudia Pons-Moll

of Hayes (1989), can only be interpreted as a minor, degenerate or moraless syllable containing a complex onset, given that only onsets are immediately dominated by the syllable node while codas are always dominated by a mora. These complex minor syllables potentially violate a universal fixed hierarchy of markedness constraints based on a more specific version of the markedness constraint against minor syllables Syllable-Head, *Complex-SyllableHead. Those constraints refer to the relative harmony of the intrasyllabic sonority profile of complex onsets in minor syllables. The activity of these constraints will be explained in more detail in § 3. The operation of resyllabification defined in (6) will also be crucial in this paper. (6) Resyllabification Take a parsed segment and change its syllabic affiliation parsing it to an already existing syllable, either to its right or to its left.

Resyllabification is a cost-free operation that is not correlated with any faithfulness constraint violation. Finally, gemination is interpreted as the result of inserting an association line from a previously syllabified root node to a following or previous syllable node. Gemination violates the faithfulness constraint Dep-Link. In § 3 the activity of these constraints will be explained in more detail. 3.3  The domain of syllabification In this paper we assume the standard idea that the input of phonology as an interpretative component of a generative grammar is a set of morphs that stands in a hierarchical morphosyntactic representation, from which linear precedence relations directly follow. Phonological linear immediate precedence relations, or adjacency, come from two different sources. On the one hand, the linear immediate precedence relation between two segments x and y can be established in the lexicon if (a) both x and y are a substring of the same morph; (b) x precedes y in the underlying representation of that morph; and (c) there is no z such that x precedes z and z precedes y. On the other hand, the linear immediate precedence relation between two segments x and y can be inherited from morphosyntax if x is the last segment in the underlying representation of a morph M1, and y is the first segment in the underlying representation of another morph M2, and M1 precedes M2 after morphosyntax. In order for core syllabification to apply, the segments x and y must stand in a linear immediate precedence relation. This is implicit in Elfner’s (2009) formulation of core syllabification.



Serial prosodification and voiced stop geminates in Catalan 

However, we also propose to constrain the applicability of core syllabification with respect to another universal condition, formalized in (7), which must be understood as an inherent property or feature of Gen. (7) Gen-restraint core syllabification in Harmonic Serialism Let (x, y) stand for a set of segments in a phonological linear immediate precedence relation. Let PCat1 and PCat2 stand for prosodic categories higher than the syllable, where PCat1 is lower than PCat2. Gen cannot create a binary syllable (xy) if: ∃ PCat1 s.t. x XOR y ∈ PCat1 & ¬ ­ ∃ PCat2 s.t. x ∧ y ∈ PCat2

As stated in (7), core syllabification is blocked when one of the two segments that stand in a phonological linear immediate precedence relation, but not the other, is dominated by a prosodic category higher than the syllable, and there is no other higher prosodic category that dominates both of them. This means that the presence of a prosodic category higher than the syllable creates an opaque domain for core syllabification. Following Elfner (2009), the constraints enforcing prosodification are those in (8) and (9).

(8)  Parse-Segment (Prs-Seg): assign one violation mark for every segment that is not associated with a syllable. (Elfner 2009)

(9) Parse-Syllable (Prs-Syll) Assign one violation mark for every syllable that is not associated with a prosodic word.8 (based on Elfner 2009)

4.  Harmonic Serialism analysis 4.1  Root-internal clusters This subsection presents a Harmonic Serialism account of those forms that include a root-internal /bl/ cluster that is always followed by a vowel belonging to the root. In these cases, the voiced stop undergoes spirantization and surfaces in onset p ­ osition along with the following lateral (/publi/ → [ˈpu.βli]).

8.  An orthodox definition of Parse-Syllable should refer to metrical feet. It is assumed here that syllables are directly parsed into prosodic words for ease of exposition. For discussions on foot parsing in Harmonic Serialism, see McCarthy (2008b) and Pruitt (2010).

 Francesc Torres-Tamarit & Clàudia Pons-Moll

The first step of the Harmonic Serialism derivation appears in tableau (12), where the relevant syllable-building operations are included. At this step of the derivation, candidates (a) and (h) are the winners, which are the ones that show the application of core syllabification projecting a mora. These are the candidates that minimally violate Parse-Segment, and violate neither *σ/O,R,9 which assigns one violation mark for every moraic obstruent or sonorant as a syllable head, nor *Complex-Syllable-Head, which is violated by candidate (g). The cover constraint *Complex-Syllable-Head, which will be split into two more specific constraints in the next subsections, is a markedness constraint that stands in a stringency relation with the less stringent constraint Syllable-Head. Both constraints are defined in (10) and (11) below. (10) Syllable-Head (Syll-Head)  Assign one violation mark for every syllable that does not dominate at least one mora. (Elfner 2009) (11) *Complex-Syllable-Head (*Complex-Syll-Head) Assign one violation mark for every complex syllable that does not dominate at least one mora.

All the other candidates are harmonically bounded by the winner. Ties are common in Harmonic Serialism when the same operation is applicable at different loci. For expository reasons, candidate (a) is taken as the input to the next step of the derivation, the one with left-to-right parsing. Taking candidate (h) would yield the same result. Parentheses mark syllable boundaries. (12) Step 1    /publi/

*σ/O,R

3

     a.  →(puµ)bli     b.  (pµ)ubli

Prs-Seg

W1

        c.  p(uµ)bli

SyllHead

*Comp-    Operations Syll-Head     core s with m

W4

    s with m

W4

    s with m

     d.  pu(bµ)li

W1

W4

    s with m

        e.  pub(1µ)i

W1

W4

    s with m

9.  Elfner (2009) proposes a universal fixed hierarchy of markedness constraints, namely *σ/O » *σ/R » *σ/V, that explain the cross-linguistic preference for high sonority segments to head syllables.



Serial prosodification and voiced stop geminates in Catalan 

   /publi/

*σ/O,R

Prs-Seg

       f.  publ(iµ)

W4

     g.  pu(bl)i

3

       h.  →pub(liµ)

3

       i.  (p)ubli

W4

        j.  publi

W5

SyllHead

*Comp-    Operations Syll-Head         s with m

W1

W1

        core s without m         core s with m

W1

        s without m         Ø

At the second step of the derivation (tableau 13), applying again core syllabification minimally violates Parse-Segment and does not violate any other constraint. The most harmonic candidate, candidate (a), harmonically bounds all the other candidates. (13) Step 2    /publi/

*σ/O,R

Prs-Seg

SyllHead

*Comp-    Operations Syll-Head

1

      a.  →(puµ)b(liµ)

       core s with m        adjunction

      b.  (puµbµ)li

W2

       c.  (puµ)(b)li

W2

W1

       d.  (puµ)(bl)i

1

W1

         s without m W1

       core s without m

       e.  (puµ)(blµ)i

W1

1

       f.  (puµ)(bµ)li

W1

W2

         s with m

W3

        Ø

      g.  (puµ)bli

       core s with m

At the next step (tableau 14), the input contains only one unparsed segment. The winning candidate, candidate (a), is the one in which the unparsed voiced stop is parsed as the first element of a complex onset to the second syllable. Applying onset adjunction is more harmonic than applying coda adjunction, as candidate (b) shows, because No-Coda dominates *Complex-Onset. *Complex-Onset is dominated by Syllable-Head, as the comparison with candidate (c) illustrates. The fully faithful candidate, candidate (d), is also ruled out because it fatally violates Parse-Segment, which dominates *Complex-Onset. The selected candidate also violates a low-ranked markedness constraint against postvocalic heterosyllabic voiced stops. We use an ad hoc constraint *V.b… for ease of exposition. The satisfaction of this markedness constraint will trigger spirantization.

 Francesc Torres-Tamarit & Clàudia Pons-Moll

(14) Step 3    /(puµ)b(liµ)/

*σ/O,R Prs- Syll- No- *Comp- *V.b...    Operations Seg Head Coda Ons

       a.  →(puµ)(bliµ) W1

        b.  (puµbµ)(liµ) W1

       c.  (puµ)(b)(liµ) W1

       d.  (puµ)b(liµ)        e.  (puµ)(bµ)(liµ)

W1

1

1

       adjunction

L

L

        adjunction

L

1

       s without m

L

L

       Ø

L

1

        s with m

At the fourth step (tableau 15), spirantization applies in order to remove the violation of *V.b…. This constraint dominates the faithfulness constraint Ident [continuant], which assigns one violation mark for every corresponding segments in the input and the output with a different specification of the feature [continuant]. No-Coda also dominates Ident [continuant]. (15) Step 4    /(puµ)(bliµ)/

No- *Comp- *V.b... Id    Operations Coda Ons [cont]

       a.  →(puµ)(βliµ)

1

        b.  (puµ)(bliµ)

1

       c.  (puµbµ)(liµ)

W1

W1

L

1

         spirantization

L

      Ø

L

        resyllabification

The derivation converges at the next step of the derivation (tableau 16), where no harmonic improvement is achievable. (16) Step 5: convergence    /(puµ)(βliµ)/

No- *Comp- *V.b... Id    Operations Coda Ons [cont]

        a.  →(puµ)(βliµ)

1

       b.  (puµ)(bliµ)

1

       c.  (puµβµ)(liµ)

W1

L

       Ø W1

W1

     fortition         resyllabification

To sum up, when the cluster /bl/ appears root-internally and followed by a vowel belonging to the root, gemination is blocked because the voiced stop is syllabified as the first element of a complex onset at step 3 of the derivation. The process of



Serial prosodification and voiced stop geminates in Catalan 

gemination, as it will be made clear in the next subsection, can only apply as long as the voiced stop is parsed in coda position. 4.2  Root-final /bl/ clusters The analysis of the cases in which a /bl/ cluster appears root-finally is given in this subsection. The first step of the derivation, in which a core syllable is created, is omitted here. The difference between an input containing a /bl/ cluster root-finally and an input containing the same cluster root-internally arises at the second step of the derivation (tableau 17). Given the absence of a vowel in an input such as /pɔbl/, core syllabification, as candidate (d) illustrates, does not represent a harmonically-improving step because *σ/O,R dominates ParseSegment. The winning candidate is then candidate (a), the one that parses the voiced stop in coda position to the previously existing syllable. However, there is the possibility of building a syllable that parses the /bl/ cluster together, as candidate (c) shows. This operation would completely satisfy Parse-Segment. However, this potential candidate is ruled out because it fatally violates a markedness constraint that, for clarity of exposition, is written here as *(bl)Syllable-Head, which assigns one violation mark for every complex minor syllable with a (b/ɡl) complex onset. This paper argues for the existence of a universal fixed hierarchy of sonority-driven markedness constraints on possible complex onsets in minor syllables that stand in a stringency relation with Syllable-Head. Following Pons-Moll (2008, 2011), we assume that laterals are less sonorous than flaps in Romance. That universal constraint hierarchy is based on the Sonority Dispersion Principle (Clements 1990), according to which the more sonority distance between the segments in a complex onset, the better. Given that flaps are more sonorous than laterals, a complex minor syllable like (b/ɡɾ) will always be more harmonic than a complex minor syllable like (b/ɡl). This is expressed by ranking *(b/ɡl)-Syllable-Head over *(b/ɡɾ)Syllable-Head. As can be seen in the following tableau, *(bl)-SyllableHead also outranks Parse-Segment. (17) Step 2    /(pɔµ)bl/

*σ/O,R

*(b1)Prs-Seg Syll- No- *V.b...    Operations Syll-Head Head Coda

       a.  →(pɔµbµ)l         b.  (pɔµ)bl        c.  (pɔµ)(bl)

W1

1

1

        adjunction

W2

L

       Ø

L

L

W1         core s       without m

 Francesc Torres-Tamarit & Clàudia Pons-Moll

       d.  (pɔµ)(blµ)

W1

L

L

W1        core s        with m

       e.  (pɔµ)(bµ)l

W1

1

L

W1        s with m

L

W1         s     without m

1

       f. (pɔµ)(b)l

W1

At the third step of the derivation (tableau 18), the unsyllabified lateral is parsed into a single minor syllable, as candidate (a) shows. This is the most harmonic candidate at this stage of the derivation because all the segments have been parsed into syllables. Parse-Segment thus dominates Syllable-Head. The winning ­candidate violates a constraint not presented yet, namely Syllable-Contact (Syll-Cont) (see, among others, Gouskova 2004), which prohibits heterosyllabic clusters with a flat or rising sonority profile. Applying again coda adjunction to (pɔb) would result in a complex coda with an intrasyllabic rising sonority profile, *(pɔbl), as candidate (c) illustrates, which is banned by Sonority-Sequencing (Baertsch 2002). The markedness constraint Sonority-Sequencing (Son-Seq) militates against complex codas in which the first element is less sonorous than the second one. The last candidate is also ruled out because it violates the higherranked constraint *σ/O,R. The next tableau demonstrates that both SonoritySequencing and Parse-Segment dominate Syllable-Contact and SyllableHead, which is dominated by *σ/O,R. (18) Step 3    /(pɔµbµ)l/

*σ/O,R

SonSeq

Parse- Syll- Syll- No-    Operations Seg Cont Head Coda

       a.  →(pɔµbµ)(l) W1

       b.  (pɔµbµ)l W1

      c.  (pɔµbµl)        d.  (pɔµbµ)(lµ)

W1

1

1

1

         s without m

L

L

1

         Ø

L

L

1

         adjunction

1

L

1

         s with m

At the next step (tableau 19), the winning candidate is the one that removes the violation of Syllable-Head by epenthesizing a schwa, which is correlated with a Dep-V violation. Gemination, on the one hand, and resyllabification, on the other hand, are not harmonically improving operations at this stage of the derivation because of the high ranking of *(bl)-Syllable-Head, which rules out candidate (c). The high ranking of a faithfulness constraint against mora deletion, Max-μ (Lin 1997), which is violated by candidate (d), is also included in tableau (19). The constraint *(bl)-Syllable-Head dominates Syllable-Contact.



Serial prosodification and voiced stop geminates in Catalan 

(19) Step 4    /(pɔµbµ)(l)/

Max-µ

*(bl)SyllSyll-Head Cont

l

     epenthesis

W1

L

   Ø

L

W1

L

        gemination

L

W1

L

          resyllabification

       a.  →(pɔµbµ)(l6µ)

l

        b.  (pɔµbµ)(l)

l W1 W1

       c.  (pɔµbµ)(bl)        d.  (pɔµ)(bl)

W1

Syll- Dep-V   Operations Head

At the fifth step of the derivation (tableau 20), gemination is able to apply in order to avoid a rising sonority profile between the two heterorganic consonants. Resyllabification is blocked by the activity of Max-μ. The winning candidate, candidate (a), thus violates the low-ranked markedness constraint Dep-Link, which assigns one violation mark for every root node multiply linked to higher prosodic tiers only in the output. (20) Step 5    /(pɔµbµ)(l6µ)/

Max-µ

SyllCont

       a.  →(pɔµbµ)(bl6µ) W1

        b.  (pɔµbµ)(l6µ)        c.  (pɔµ)(bl6µ)

W1

Dep-    Operations Link

NoCoda

*CompOns

l

l

l

       gemination

l

L

L

     Ø

L

1

L

     resyllabification

Convergence is met at the next step of the derivation, omitted here. The analysis proposed so far has demonstrated that the opaque interaction between gemination and schwa epenthesis, which stand in a counterbleeding relation, is straightforwardly captured by Harmonic Serialism, where processes are applied in a step-wise manner under the same constraint hierarchy. 4.3  Root-final /bɾ/ clusters At this point of the discussion, those inputs containing a /bɾ/ cluster root-finally can be compared with those ones containing a /bl/ cluster. The crucial difference between an input like /pɔbl/ and an input like /pɔbɾ/ is that in the former case, as has been demonstrated, the creation of a complex minor syllable is not possible to build at the second step of the derivation (tableau 17) because of the ranking *(bl)-Syllable-Head » Parse-Segment. However, if Parse-Segment dominates *(bɾ)-syllable-head, then a binary complex minor syllable with an

 Francesc Torres-Tamarit & Clàudia Pons-Moll

empty nucleus emerges as the most harmonic candidate at the second step of the derivation for inputs with a root-final /bɾ/ cluster. Then, if the voiced stop is not syllabified in coda position, but in onset position, there is no chance for ­gemination to apply later in the derivation, because Syllable-Contact is already satisfied. The second step for /pɔbɾ/ is illustrated below. (21) Step 2    /(pɔµ)bɾ/

No- *Comp- *V.b...    Operations Coda Ons

Prs *(bɾ)SyllSeg Syll-Head Head

       a.  →(pɔµ)(bɾ )

2

l

l

l

      core s         without m

L

L

      Ø

L

L

     adjunction

        b.  (pɔµ)bɾ

W2

L

L

       c.  (pɔµbµ)ɾ

W1

L

L

W1

At step 3 (tableau 22), an epenthetic schwa is inserted because this is the most harmonic operation that compels the satisfaction of *(bɾ)-Syllable-Head, given the ranking *(bɾ)-Syllable-Head » Dep-V. (22) Step 3    /(pɔµ)(b)/

Dep- *Comp- *V.b...    Operations V Ons

*(bɾ)SyllSyll-Head Head

       a.  →(pɔµ)(bɾ6µ) W1

        b.  (pɔµbµ)(ɾ)        c.  (pɔµ)(bɾ)

W1

l

l

l

    epenthesis

L

L

L

       resyllabification

l

l

        Ø

W1

Later on (tableau 23), spirantization applies in order to satisfy *V.b…, the constraint against postvocalic heterosyllabic voiced stops. The derivation converges at the next step of the derivation, not shown here. (23) Step 4    /(pɔµ)(bɾ6µ)/

NoCoda

*CompOns

      a.  →(pɔµ)(bɾ6µ)

l

       b.  (pɔµ)(bɾ6µ)

l

       c.  (pɔµbµ)(ɾ6µ)

W1

L

*V.b...

W1

Id    Operations [cont] l

      spirantization

L

     Ø

l

       resyllabification



Serial prosodification and voiced stop geminates in Catalan 

In this subsection it has been shown that the ranking between the constraints *(bl)-Syllable-Head and *(bɾ)-Syllable-Head with respect to ParseSegment, namely *(bl)-Syllable-Head » Parse-Segment » *(bɾ)-SyllableHead, explains the asymmetry between those inputs containing a root-final cluster /bl/ and those containing /bɾ/. The fact that *(bl)-Syllable-Head is higher-ranked forces the voiced stop to be syllabified in coda position, at the expense of violating Parse-Segment, which triggers gemination later on in the derivation in order to fix an intersyllabic rising sonority contact. 4.4  V  oiced stop plus lateral root-final clusters followed by overt inflectional suffixes As stated in § 2, voiced stop gemination also stands in a counterbleeding relation with morphological affixation. The presence of a vowel-initial derivational or inflectional suffix does not block gemination, although the presence of vowelinitial suffixes introduce the phonological context that could bleed the application of gemination by allowing core syllabification to apply to those sequences. As introduced in § 3.3, in this paper it is proposed that prosodic categories higher than the syllable create opaque domains for syllabification. This fact explains why an input like /reɡl+6/, consisting of a root followed by the inflectional feminine morph, as opposed to /publi/, where the last vowel belongs to the underlying lexical form of the root, escapes spirantization and undergoes gemination if first the root is parsed into its own prosodic word. The vowel belonging to the feminine morph cannot be integrated into a syllable together with the last consonant of the root at the steps of the derivation in which syllabification applies. This is so because there is a prosodic word dominating the root but not the affix, which creates an opaque domain for syllabification. This situation emerges if a prosody-morphology interface constraint requiring the right edge of the root to be aligned with some prosodic word, Align-Right (Root, Prosodic Word) (Al-R (√, ω)) dominates Parse-Segment. If the alignment constraint is first satisfied, the root in /reɡl+6/ is syllabified the same way /pɔbl/ is, where the voiced stop is parsed as a syllable coda and the lateral forms a single minor syllable. The vowel belonging to the suffix is parsed into its own syllable, yielding the following intermediate representation: (reμɡμ)(l)+(6μ). At that point, a prosodic word dominating all the syllables is the most harmonic candidate given the ranking Parse-Syllable » Syllable-Head, as tableau (24) illustrates. The winning candidate is thus candidate (a), with a recursive prosodic word.10 Square brackets mark prosodic word

10.  We could assume that the last syllable of the string is adjoined to the already existing prosodic word, instead of being adjoined to a recursive prosodic word that dominates the inner

 Francesc Torres-Tamarit & Clàudia Pons-Moll

boundaries and the symbol ‘+’ indicates that there is a prosodic word boundary separating the root and the affix. This prosodic word boundary creates an opaque domain that prevents core syllabification to operate with the last consonant of the root and the vocalic suffix. (24)    /[(reµ:µ)(l)]+(6µ)/

PrsSyll

SyllCont

SyllHead

l

l

       a.  →[(reµ:µ)(l)] (6µ)         b.  [(reµ:µ)(l6µ)]+(6µ)

W1

l

L

       c.  [(reµ:µ)(l)]+(6µ)

W1

l

l

Dep-    Operations V     w building W1

      epenthesis      Ø

At this point of the derivation, syllabification is sensitive to the whole string of segments dominated by the prosodic word. Syllable-Head must thus be satisfied. Among the alternatives, conflating the single minor syllable together with the onsetless syllable into one syllable is the most harmonic one, given that this operation is not correlated with any violation of a faithfulness constraint, as opposed to inserting an epenthetic vowel, correlated with a Dep-V violation. This is demonstrated in the tableau (25). (25)    /[[(reµ:µ)(l)](6µ)/

SyllCont

      a.  →[[(reµ:µ)(l6µ]]

l

       b.  [[(reµ:µ)(l6µ)](6µ)]

l

     c.  [[(reµ:µ)(l)](6µ)]

l

SyllHead

Dep-      Operations V       w building W1

W1

      epenthesis      Ø

At that point of the derivation, the input /[[(reμɡμ)(l6μ)]]/ is parallel to the input /(pɔμbμ)(l6μ)/ in tableau (20). Gemination applies at the next step to satisfy Syllable-Contact. The final ranking of the whole set of constraints presented so far appears in (26) as a Hasse diagram.

prosodic word. But saying this would force us to assume that Align-Right(Root, Prosodic Word) is only violated in the absence of input syllables, in order to make this constraint active before syllabification, and inactive after syllabification, which seems a stipulative assumption.



Serial prosodification and voiced stop geminates in Catalan 

(26) ∗(bl)-Syll-Head

∗(bɾ)-Syll-Head

Al-R(√,ω)

∗σ/O,R

Prs-Seg

Son-Seq

Prs-Syll

Syll-Head

Dep-V

∗V.b...

Max-μ

Syll-Cont

No-Coda

∗Comp-Ons

Dep-Link

Id[cont]

5.  Conclusions This paper has presented a Harmonic Serialism analysis of voiced stop gemination in Catalan and has shown how this serial model without strata is able to derive some opaque forms which show counterbleeding interactions between voiced stop gemination, schwa epenthesis and suffixation. The analysis rescues two essential ideas from previous literature on voiced stop gemination in Catalan: gemination only applies when voiced stops are parsed in coda position (Mascaró 1987) as a strategy to avoid a rising syllable contact (Bermúdez-Otero 2000; Colina 1995; Jiménez 1997; Pons-Moll 2004, 2008, 2011). The data analyzed in this paper require a crucial ordering between different phonological operations: syllabification, epenthesis and gemination. The interaction between those operations can be straightforwardly accounted for in Harmonic Serialism if prosody-building operations count as a single step (Elfner 2009). Two different proposals about syllabification have been made that explain the asymmetries between voiced stop plus lateral root-final clusters, with or without inflectional suffixes, on the one hand, and voiced stop plus tap clusters, on the other hand. First, a theory of serial syllabification in

 Francesc Torres-Tamarit & Clàudia Pons-Moll

­ armonic Serialism based on Elfner (2009) has been developed. We have proH posed that the binary operation core syllabification can create complex minor syllables and cannot operate with two adjacent segments if one of these segments, but not the other, is integrated into a prosodic category higher than the syllable. This situation allows for positing a universal fixed hierarchy of markedness constraints disfavoring those complex onset configurations based on the Sonority Dispersion Principle (Clements 1990), according to which a complex minor syllable like (bl) is more marked than a complex minor syllable like (bɾ), given that taps behave as more sonorous than laterals in Romance (Pons-Moll 2008, 2011). The constraint ranking *(bl)-Syllable-Head » Parse-Segment » *(bɾ)-Syllable-Head explains the difference between [ˈpɔb.bl6], with gemination, and [ˈpɔ.βɾ6], with spirantization. Second, the asymmetry between /publi/ → [ˈpu.βli], with spirantization, and /reɡl+6/ → [ˈreɡ.ɡl6], with gemination, is explained resorting to the idea that the presence of a prosodic word boundary creates an opaque domain for syllabification operations. This restriction on Gen about syllable formation operations together with the ranking AlignRight(Root, ProsodicWord) » Parse-Segment ensures that words with voiced stop plus lateral root-final clusters with overt suffixes (i.e. /reɡl+6/) behave like words without overt morphs (i.e. /pɔbl/), which show gemination, instead of behaving like words with a final vowel belonging to the root (i.e. /publi/), which undergo spirantization.

References Baertsch, Karen. 2002. An Optimality Theoretic approach to syllable structure: the split margin hierarchy. Ph.D. dissertation, Indiana University. Bermúdez-Otero, Ricardo. 2000. “Catalan obstruent gemination.” Manuscript, University of Manchester. Bermúdez-Otero, Ricardo. To appear. Stratal Optimality Theory. Oxford: Oxford University Press. Bonet, Eulàlia & Lloret, Maria-Rosa. 1998. Fonologia catalana. Barcelona: Ariel. Clements, George N. 1990. “The role of the sonority cycle in core syllabification.” In Papers in Laboratory Phonology I: Between the grammar and physics of speech, John Kingston & Mary Beckman (eds), 283–333. Cambridge: Cambridge University Press. Colina, Sonia. 1995. A constrained-based analysis for syllabification in Spanish, Catalan and Galician. Ph.D. dissertation, University of Illinois. Elfner, Emily. 2009. “Syllabification and stress-epenthesis interactions in Harmonic Serialism.” Manuscript, University of Massachusetts. Elfner, Emily. To appear. “Stress-epenthesis interactions in Harmonic Serialism.” In Harmonic Grammar and Harmonic Serialism, John J. McCarthy & Joe Pater (eds). London: Equinox. Fabra, Pompeu. 1912. Gramática de la lengua catalana. Barcelona: L’Avenç. Gouskova, Maria. 2004. “Relational hierarchies in OT: The case of syllable contact.” Phonology 21 (2): 201–250.



Serial prosodification and voiced stop geminates in Catalan 

Hayes, Bruce. 1989. “Compensatory lengthening in moraic phonology.” Linguistic Inquiry 20 (2): 253–306. Jesney, Karen. To appear. “Positional Faithfulness, non-Locality, and the Harmonic Serialism solution.” In Proceedings of the North East Linguistic Society 39, Suzi Lima, Kevin Mullin & Brian Smith (eds). Amherst: GLSA. Kiparsky, Paul. 2000. “Opacity and cyclicity.” The Linguistic Review 17 (2–4): 351–366. Jiménez, Jesús. 1997. L’estructura sil·làbica del valencià. Ph.D. dissertation, Universitat de València. Kimper, Wendell A. 2011. “Locality and globality in phonological variation.” Natural Language & Linguistic Theory 29 (2): 423–465. Lin, Yen-Hwei. 1997. “Syllabic and moraic structures in Piro.” Phonology 14 (3): 403–436. Mascaró, Joan. 1976. Catalan phonology and the phonological cycle. Ph.D. dissertation, Massachusetts Institute of Technology. Mascaró, Joan. 1987. “Syllable-final processes in Catalan.” In Studies in Romance Language, Carol Neidle & Rafael A. Núñez Cedeño (eds), 163–180. Dordrecht: Foris. Mascaró, Joan. 2003. “Comparative markedness and derived environments.” Theoretical Linguistics 29 (1–2): 113–122. McCarthy, John J. 2000. “Harmonic Serialism and parallelism.” In Proceedings of the North East Linguistic Society 30, Masako Hirotani, Andries Coetzee, Nancy Hall & Ji-yung Kim (eds), 501–524. Amherst: GLSA. McCarthy, John J. 2007a. “Restraint of analysis.” In Freedom of analysis, Sylvia Blaho, Patrik Bye & Martin Krämer (eds), 203–231. Berlin and New York: Mouton de Gruyter. McCarthy, John J. 2007b. Hidden generalizations: Phonological opacity in Optimality Theory. London: Equinox. McCarthy, John J. 2008a. “The gradual path to cluster simplification.” Phonology 25 (2): 271–319. McCarthy, John J. 2008b. “The serial interaction of stress and syncope.” Natural Language & Linguistic Theory 26 (3): 499–546. McCarthy, John J. 2010a. “An introduction to Harmonic Serialism.” Language and Linguistics Compass 4 (10): 1001–1018. McCarthy, John J. 2010b. “Studying Gen.” Journal of the Phonetic Society of Japan 13 (2): 3–12. McCarthy, John J. 2010c. “Autosegmental Spreading in Optimality Theory.” In Tones and Features (Clements memorial volume), John Goldsmith, Elizabeth Hume & Leo Wetzels (eds), 341–373. Berlin and New York: Mouton de Gruyter. McCarthy, John J. 2012. “Pausal phonology and morpheme realization.” In Prosody matters: Essays in honor of Lisa Selkirk, Toni Borowsky, Shigeto Kawahara, Takahito Shinya & Mariko Sugahara (eds), 341–373. London: Equinox. Nespor, Marina & Vogel, Irene. 1986. Prosodic phonology. Dordrecht: Foris. Pater, Joe. 2012. “Serial Harmonic Grammar and Berber syllabification”. Prosody matters: Essays in honor of Elisabeth Selkirk.” In Toni Borowsky, Shigeto Kawahara, Takahito Shinya & Mariko Sugahara (eds), 43–72. London: Equinox. Pons-Moll, Clàudia. 2004. Els contactes consonàntics en balear: descripció i anàlisi. Ph.D. dissertation, Universitat de Barcelona. Pons-Moll, Clàudia. 2008. “Regarding the sonority of liquids: Some evidence from Romance.” Paper presented at the 38th Linguistic Symposium on Romance Languages, UrbanaChampaign, April 2008. Pons-Moll, Clàudia. 2011. “It is all downhill from here: The role of Syllable Contact in Romance languages.” Probus 23 (1): 105–173.

 Francesc Torres-Tamarit & Clàudia Pons-Moll Prince, Alan & Prince, Smolensky. 1993/2004. “Optimality Theory: constraint interaction in generative grammar.” Manuscript, Rutgers University and University of Colorado, Boulder. Malden /Oxford: Blackwell, 2004. Pruitt, Kathryn. 2010. “Serialism and locality in constraint-based metrical parsing.” Phonology 27 (3): 481–526. Recasens, Daniel. 1991. Fonètica descriptiva del català. Barcelona: Institut d’Estudis Catalans. Recasens, Daniel. 1993. Fonètica i fonologia. Barcelona: Enciclopèdia Catalana. Topintzi, Nina. 2006. A (not so) paradoxical instance of compensatory lengthening. Journal of Greek Linguistics 7: 71–119. Wheeler, Max W. 2005. The phonology of Catalan. Oxford: Oxford University Press. Wheeler, Max W. 1979. Phonology of Catalan. Oxford: Blackwell. Wheeler, Max W. 1980. “Phonological variation in contemporary Catalan.” Miscel·lània Aramon i Serra 2: 603–621.

Interfacing information and prosody French wh-in-situ questions Viviane Déprez*, Kristen Syrett & Shigeto Kawahara

*Laboratoire sur le Langage, le Cerveau et la Cognition (L2C2-CNRS) / Rutgers University We present experimental evidence bearing on Cheng and Rooryck’s (2000) proposal that French wh-in-situ questions are licensed by an intonational morpheme also present in yes-no questions and their claim that such questions are ungrammatical without a rising contour. While most participants produced a rising contour, not all did; when they did, the slope was not as steep as in yes-no questions. Our findings support C&R’s proposal, admitting the central role of information structure. We support a view of question formation in French in which information structure, syntax, and prosody form a tight relationship: the shape of the syntactically-designated contour is affected by pragmatic information. We present a theoretical account appealing to movement through givenness-marking that explains the observed pitch compression. Keywords:  wh-questions; French interrogatives; prosody; information structure; givenness

1.  Introduction French has both moved wh-questions and wh-in-situ questions, as indicated in (1a–b). (1) a. Elle est allée où en Allemagne? she is.3sg go-pst-f where in Germany ‘Where did she go in Germany?’ b. Où est-ce-qu’ elle est allée en Allemagne? where ques she is.3sg go-pst-f in Germany ‘Where did she go in Germany?’

In an influential paper that provided an elegant account for wh-in-situ questions in French, Cheng and Rooryck (2000) (henceforth C&R) claimed that these questions manifest the following properties. (We note here other researchers making similar claims.)

 Viviane Déprez, Kristen Syrett & Shigeto Kawahara

A. They occur in particular discourse contexts characterized as ‘strongly presuppositional’, which foster positive expectations and exclude negative answers (Chang 1997); B. They only occur in root clauses and cannot be embedded (Boeckx 2000; Bošcović 2000); C. They are sensitive to intervention effects (Boeckx 2000; Bošcović 2000; ­Zubizarreta 2003); D. They are syntactically licensed by an intonation morpheme merged in C that induces an obligatory sentence-final rising intonation contour identical to the one found in purely intonational yes-no questions, such as the one in (2). (2) Elle est allée en Allemagne? she is.3sg go-pst-f in Germany ‘Did she go to Germany?’

Of these four claims, only (C) remains unchallenged. The discourse and syntactic claims in (A) and (B) have been questioned by attested corpus examples featuring negative answers or embedded instances of in-situ questions (Adli 2006; Hamlaoui 2009; Mathieu 2009). However, the possibility of dialectal distinctions (Zubizarreta 2003) or other semantic nuances (Baunaz & Patin 2009) still leaves this issue open. C&R’s claim about the obligatory rising intonation morpheme in (D) has been challenged in a number of papers. However, these challenges have relied primarily on informal intuitions (Hamlaoui 2010; Mathieu 2009), and occasionally on previous experimental studies, which were not specifically designed to investigate the prosody of in-situ questions (Delattre 1966; Beyssade et al. 2007; a.o.) and/or omitted important design variables, thereby casting doubts on the solidity of their results (Adli 2006; cf. also Wunderli 1983, 1984). As a consequence, the validity of both C&R’s claims and the objections to them remain controversial. In this paper, we first summarize the results of an experimental investigation reported on in Déprez, Syrett and Kawahara (2012) (henceforth DS&K), which specifically targeted the intonation of French wh-in-situ questions, while also taking into account some key aspects of their discourse felicity conditions. We then discuss the impact of these results on the claims in (D), the heart of C&R’s account. Drawing on recent works by Kučerová (2007) and Wagner (2005, 2006), we also discuss how a deeper understanding of information structure brings new insights on how pragmatic, prosodic and syntactic factors interact in the licensing of these questions. C&R have claimed that French wh-in-situ questions are characterized by an obligatory rising contour that is identical to that of yes-no questions. In DS&K, we address this claim experimentally with a production study of French in-situ questions. In this paper, we rely on these empirical findings to ground a more general, theoretical discussion of the information structure and syntactic licensing of wh-in-situ questions.



Interfacing information and prosody in French wh-in-situ 

The structure of the paper is as follows. In Section 1, we summarize our experimental findings on the prosody of French wh-in-situ questions. In Section 2, we discuss the discourse conditions that have been found to affect the felicity of these questions. In Section 3, we bring these two strands of research together, reflecting upon how they interact with one another and what role these distinct factors play in licensing these questions. We end with predictions concerning how our findings could play out in the particular theoretical model for question licensing drawn by Richards (2006, 2010). 2.  The prosody of French in-situ questions: An experimental investigation C&R proposed that French wh-in-situ questions are syntactically licensed by an abstract intonation morpheme in C. This morpheme plays a role largely identical to the one played by overt question particles in wh-in-situ languages such as Japanese or Korean (Chang 1997). On this view, wh-in-situ questions – be they apparently optional (French, Portuguese) or more generally obligatory (Chinese, Japanese) – are syntactically licensed in a unified way: through the presence of a specialized complementizer that marks the clause as an interrogative and licenses the in-situ question term. C&R claim that the French intonation morpheme is phonetically realized as an obligatory sentence-final rising contour comparable to the one commonly found in purely into national yes-no questions. This proposal makes the ­following two empirical predictions for the intonation of French wh-in-situ questions. First, they should consistently manifest a detectable sentence-final rising contour. Second, their final rise should be qualitatively identical to the one observed in purely intonational yes-no questions. We tested these predictions experimentally in DS&K, which we briefly review here. (See DS&K for further details.) 2.1  Experimental design 2.1.1  Participants and procedure 12 French native speakers (5 F, 7 M; age range: early 20s – late 50s)1 were recorded in a sound-attenuated booth using a high quality microphone. Stimulus sentences were presented one at a time on a computer screen.

1.  Speakers were from French regions, and were second language speakers of English, having been in the States for a time ranging between a few months to several years. Our data revealed no effect of time spend in the United States or region.

 Viviane Déprez, Kristen Syrett & Shigeto Kawahara

2.1.2  Stimuli The stimuli consisted of seven sentence types: (1) declarative, (2) yes-no ­question with est-ce que, (3) yes-no question without est-ce que in declarative form, (4) wh-in-situ question (content), (5) wh-in-situ question (echo), (6) moved wh-­question with est-ce que, (7) moved wh- question without est-ce que and with subject-­auxiliary inversion. The first three served as the baseline of intonation contours, as predictions of a rise or lack of rise were clear. The fourth was our target construction. Each sentence was preceded by a short context presenting a choice scenario. This setup was intended to make the target sentence felicitous by providing most of the question information content, except for the questioned constituent itself. A sample context with three types of target sentences is given in (3). (3) Pour participer à un test de psychologie, to participate-inf in a test of psychology ‘To participant in a psychology test,’ Emma devait placer un rond, un carré ou un triangle Emma must-pst place a circle a square or a triangle ‘Emma had to place a circle a square or a triangle’ sur un tableau. on a board ‘on a board.’ Le psychologue a demandé: the psychologist has ask-pst ‘The psychologist asked:’ [Additional sentence for yes-no question only:] Emma a pris le rond et l’ a placé Emma has take.pst the circle and it has place.pst sur le tableau. on the board ‘Emma took the circle and placed it on the board.’ a. moved wh- question (wh- phrase underlined) Quel élément a-t-elle mis au milieu? which shape has she put-pst in.the middle ‘Which shape did she place in the middle?’ b. wh-in-situ question Elle a mis quel élément au milieu? she has put-pst which shape in.the middle ‘She placed which shape in the middle?’ c. yes-no question Elle a mis cet élément au milieu? she has put-pst this shape in.the middle ‘She placed that shape in the middle?’



Interfacing information and prosody in French wh-in-situ 

There were five such discourse contexts for each of the seven sentence types, resulting in a total of 35 items. To prevent the sentence-final contour from being merged with the pitch accent on the wh-term, and to prevent potential sentence-final creakiness from disrupting the F0 contour, the in-situ question items were always followed by PPs that were either complements or adjuncts of the verbs as in (3b). Lexical items used sonorants as much as possible to obtain an unperturbed into national contour. Sentence types were blocked to avoid the influence of minimal pair members on each other (e.g. declaratives never appeared alongside purely intonational ­yes-no questions). Participants were randomly assigned to one of three conditions in a Latin square format. The order of the sentences was randomized within each block per participant. All recordings were inspected for naturalness by native French speakers. See DS&K for discussion of the naturalness judgment task and additional coding of files. 2.1.3  Analysis Two types of data were collected for analysis: (1) categorical perception of sentencefinal rise/fall and (2) quantitative measure of intonation contour (fundamental frequency, or F0). For the first measure, three research assistants independently coded sound files for rise/fall. These files were then given to two additional coders with little to no knowledge of French for double-blind verification. Agreement between this second group of coders was 90%. All disagreements were resolved by discussion while listening to the files, still blind. For the second measure, all the target sentences were annotated in Praat (Boersma & Weenink 1999–2011). Target sentences were divided into 40 ­windows with equal duration beginning with the onset of the wh-word or corresponding determiner and ending at the termination of the sentence. The average F0 was calculated for each window, allowing us to quantify the sentence-final intonational contour for each sentence. We then performed statistical analyses of these values using R. 2.2  Results Analysis 1. The percentage of perceived final rise is presented in Figure 1. The three baseline cases patterned as predicted, with declarative sentences manifesting little to no final rise, and yes-no questions showing a final rise in nearly every instance. These differences are significant from each other and from chance level (by proportion and binomial tests). At first glance, the abovechance (p ∀; ∀>∃

(ii) At least two judges awarded me every medal.

2>∀; ∀>2



(iii) Almeno un amico ha presentato a Maria ogni ragazza. At least one friend aux.3sg introduced to Maria every girl ‘At least one friend introduced every girl to Maria’. ∃>∀; ∀>∃ Thus, the double object construction differs from the causative construction as far as the scope interaction between the subject and the object are concerned. This is a desired result if the causative structure and the double object structure are structurally different in the way proposed here.

 Francesco Costantini

thus arguments of the causative head, whereas the object (if any) is an argument of the causativized predicate. In much the same way, Ippolito (2000) proposes that the causee is merged outside the causativized vP. She claims that it is hosted in the specifier of an applicative head (Appl), which assigns benefactive (or malefactive) θ-role and checks Dative Case. ApplP dominates the causativized vP and is dominated by the causative phrase, which hosts the causer. (35) [causeP causer [cause' vcause [ApplP causee [Appl' Appl [vP … object]]]]]

As far as quantification is concerned, Alsina’s and Ippolito’s hypotheses suggest different interpretations depending on what may be considered as a quantificational domain, i.e. either the causativized vP or ApplP, or both, or none. To determine whether the causativized vP or ApplP count as quantificational domains is outside the scope of this article. However, since both phrasal domains are compatible with the semantic type and might be quantificational (see Pylkkänen 2002), let us consider three possible scenarios. (i) Let us suppose that in the structure in (35), the causativized vP counts as a quantificational domain. The object is adjoined to vP by QR, whence it cannot take scope over the causee.11 (36) … [ApplP causee [Appl' Appl [vP object [vP … object ]]]]

This is, however, contrary to facts. (ii) Let us suppose that ApplP, though not vP, counts as a quantificational domain. The object undergoes QR, adjoins to ApplP, and may take scope over the causee. (37) [causeP causer [cause' vcause [ApplP object [ApplP causee [ … object]]]]]

Under Ippolito’s theory the causee checks Dative Case in [spec; ApplP]. Consequently, the causee need not raise outside the causative vP and cannot take scope over the causer and the object, which is, again, contrary to the facts. A weaker version of Ippolito’s proposal, whereby the causee does not check Dative Case against the applicative head and raises à la Kayne (2004) in order to check Case, would get the desired results. Nonetheless, this would be a costlier theory, since the theories that hypothesize a structure as in (31) get the same results with less assumptions (under these theories, ApplP is not necessary and the “affectedness constraint” may be explained as in Folli and Harley (2007)).

11.  The object does not take scope over the causee unless one assumes that the object moves bypassing the causee (but not the causer!). This is clearly an ad hoc solution, however.



On the argument structure of the causative construction 

(iii) Finally, let us suppose that neither the causativized vP nor Ippolito’s ApplP are quantificational domains, and the first eligible site for QR is the causative vP. In that case, Alsina’s and Ippolito’s theories would be equivalent to Zubizarreta’s (1985). To conclude, theories that posit a structure akin to (35) for the causative structure do not seem to capture the relevant generalizations discussed above. 4.2.3  Folli and Harley (2007) Folli and Harley (2007) propose an argument structure akin to (31), which I will repeat here:12 (38)

vcauseP causer vP

vcause causee





v

object

As for Case checking, however, they claim that the arguments of a causative construction check Case in situ via abstract Agree. The data discussed in Section 3.2 appear to disfavor such a claim, at least with respect to the causee. Under their hypothesis, since the causee need not move in order to check Case, it does not take scope over the causer and the object, as it does not raise. (39) [causeP causer … [vP object [vP causee[dat] [v' v [ … object ]]]]]

This conclusion is however incompatible with the data discussed above.13 To sum up, the structure in (31) and Kayne’s analysis about how the causee checks Case appears to capture straightforwardly the relevant generalizations.

12.  Similarly to Guasti (1993, 1996), they assume however that the [spec; vP] is on the right in order to explain the unmarked word order. (i) [vP [v' v compl ] spec] 13.  The object, on the other hand, may check Case in situ. It may even check Case by moving, at the condition that the Case checking site be lower than [spec; vP] and [spec; causeP].

 Francesco Costantini

4.3  A remark on word order In Section 3.2 I have discussed causative constructions displaying the marked word order (V>causee>object). This was done on the assumption that the marked word order is not derived via movement and preserves the merging structural relations between arguments.14 Since movement affects scope interactions (see Section 2), the marked word order is the most suitable to compute scope interactions in order to establish whether two constituents are coarguments. As a matter of fact, the unmarked word order shows different scope interactions than the marked word order. If the word order is unmarked, the object can outscope the causer, which is not true if the word order is marked (compare sentence (23)). (40) Almeno un insegnante ha fatto leggere ogni libro At least one teacher aux.3sg made read every book ∃>∀; ∀>∃

a Gianni. to Gianni ‘At least one teacher made Gianni read every book.’

The object must c-command the causer at some level of the representation in the derivation of (40). Among the technical solutions that may be implemented to account for those facts, the most promising is that the causativized vP is raised outside the causative vP, as originally proposed by Burzio (1986) and later by Ippolito (2000) to account for the adjacency of causative verb and infinitive. (41) a. [causeP causer [ vcause [vP … v … object]]] b. [[vP … v … object] [causeP causer [ vcause [vP … v … object]]]]

By QR, the object is adjoined to the causativized vP at LF. (42) vP object

vP …v object





vcauseP causer vcause

vP

14.  In fact even under the unmarked word order some movement must occur, but it concerns the infinitive verb only.

(i) causer > vcause > causee > Vinf > (object) (ii) causer > vcause > Vinf > causee > (object)

(merging ordering) (‘surface’ marked ordering)

As this point is tangential to the main topic of this article, I leave it for further investigation.



On the argument structure of the causative construction 

Assuming Kayne’s (1994: 16) definition of c-command (“X c-commands Y iff X and Y are categories and X excludes Y and every category that dominates X dominates Y”), since the object is dominated by a segment of vP and not by the category vP, it c-commands at LF (and can take scope over) the causer.

5.  Conclusion I have resorted to the properties of QR in order to establish the argument structure of the causative construction in Italian. Following Hornstein (1995), J­ ohnson and Tomioka (1998), Johnson (2000) and Bruening (2001), I have assumed that QR is local and obeys superiority, and that A-movement reconstructs. Given these natural assumptions, the scope interactions between arguments show that the causative construction is not like a double object construction in its argument structure properties. Rather, the causee and the object are arguments of the causativized predicate, whereas the causer is introduced outside the causativized vP, arguably by the causative head, as previously proposed by Kayne (1975), Burzio (1986), Guasti (1993, 1996), Cinque (1998), Folli and Harley (2007). Moreover, the data I have presented suggest that the causee moves outside the causative vP in order to check Dative Case, as previously claimed by Kayne (2004).

References Alsina, Alex. 1992. “On the argument structure of causatives.” Linguistic Inquiry 23: 517–555. Belletti, Adriana & Rizzi, Luigi. 1988. “Psych-verbs and -theory.” Natural Language & ­Linguistic Theory 6: 291–352. Bruening, Benjamin. 2001. “QR obeys superiority: frozen scope and ACD.” Linguistic Inquiry 32: 233–273. Burzio, Luigi. 1986. Italian syntax: A government and binding approach. Dordrecht: Reidel. Cecchetto, Carlo. 2004. “Explaining locality condition of QR: Consequences for the theory of phases.” Natural Language Semantics 12: 345–397. Chomsky, Noam. 2000. “Minimalist inquiries: The framework.” In Step by Step: Essays in Minimalist Syntax in Honor of Howard Lasnik, Robert Martin, David Michaels, Juan Uriagereka (eds), 89–155. Cambridge, Mass.: MIT Press. Chomsky, Noam. 2001. “Derivation by phase.” In Ken Hale: A Life in Language, Michael Kenstowicz (ed.), 1–52. Cambridge, Mass.: MIT Press. Chomsky, Noam. 2004. “Beyond explanatory adequacy.” Structures and beyond, Adriana Belletti (ed.), 104–131. Oxford: Oxford University Press. Cinque, Guglielmo. 1998. “The interaction of passive, causative, and ‘restructuring’ in Romance.” University of Venice Working Papers in Linguistics 8/2: 29–51. Collins, Chris & Thráinsson, Höskuldur. 1993. “Object shift in double object construction and the theory of Case.” Papers on Case and agreement II. MIT Working Papers in Linguistics 17, Colin Phillips (ed.), 131–174.

 Francesco Costantini Folli, Raffaella & Harley, Heidi. 2007. “Causation, obligation, and argument structure: On the nature of little v.” Linguistic Inquiry 38: 197–238. Fox, Danny. 2000. Economy in semantic interpretation. Cambridge, Mass.: MIT Press. Giorgi, Alessandra & Longobardi, Giuseppe. 1991. The syntax of noun phrases. Cambridge: Cambridge University Press. Guasti, Maria Teresa. 1993. Causative and perception verbs. Turin: Rosenberg and Sellier. Guasti, Maria Teresa. 1996. “Semantic restrictions in Romance causatives and the incorporation approach.” Linguistic Inquiry 27: 294–313. Heim, Irene & Angelika Kratzer. 1998. Semantics in generative grammar. Oxford: Blackwell. Hornstein, Norbert. 1995. “Putting truth into Universal Grammar.” Linguistics and Philosophy 18: 381–400. Ippolito, Michela. 2000. Remarks on the argument structure of Romance causatives. Ms., MIT, Cambridge, Mass. Johnson, Kyle, 2000. “How far will quantifier go?” In Step by step, Roger Martin, David Michaels, & Juan Uriaguereka (eds), 187–210. Cambridge, Mass.: MIT Press. Johnson, Kyle & Tomioka, Satoshi. 1998. “Lowering and mid-size clauses.” Proceedings of the 1997 Tübingen workshop on reconstruction, Graham Katz, Shin-Sook Kim & Winhart Haike (eds), 185–206. Kayne, Richard. 1975. French syntax. Cambridge, Mass.: MIT Press. Kayne, Richard. 1994. The antisimmetry of syntax. Cambridge, Mass.: MIT Press. Kayne, Richard. 2004. “Prepositions as probes.” The cartography of syntactic structures. Vol. 3, Structures and beyond, Adriana Belletti (ed.), 192–212. New York: Oxford University Press. Lasnik, Howard & Saito, Mamoru. 1991. “On the subject of infinitives.” Papers from the 27th regional meeting of the Chicago Linguistic Society, Lise M. Dobrin, Lynn Nichols & Rosa M. Rodriguez (eds), 324–343. Pylkkänen, Liina. 2002. Introducing arguments. Ph.D. diss., Department of Linguistics. ­Cambridge/Mass., MIT. Stepanov, Arthur & Stateva, Penka. 2007. “When QR disobeys superiority.” Linguistic Inquiry 38, 176–185. Sazbolcsi, Anna (ed.). 1997. Ways of scope taking. Dordrecht: Kluwer. Zubizarreta, Maria Luisa. 1985. “The relation between morphophonology and morphosyntax: The case of Romance causatives.” Linguistic Inquiry 16, 247–289.

Index

A adjunction  118–119, 123–126, 128, 205 affectedness constraint  204, 216 agreement  1–2, 6, 9, 14, 17–19, 22, 139, 177, 179–180, 182, 188, 200–202 Albanian  5–6, 8 allomorphy  25, 27 contextual allomorphy  21–22, 24–27, 30, 34–38 stem allomorphy  21, 23, 27, 35, 38 alternation  8, 21–24, 27–30, 32–38 stem alternation  21–22, 24, 27, 34, 37–38 analogy  111 argument structure  53, 203–205, 211, 213, 215, 217, 219 aspect  41–42, 47, 53, 79, 81–82, 85, 173 atelic  65, 67, 81, 83–84 B Basque  14 blocking  8, 21, 188, 192

case checking  203, 206, 212, 214, 217 Catalan  15, 40, 51, 115–116, 131, 158, 163–164, 166–171 causatives  203–205, 213, 215 causee  203–205, 209–219 causer  203–205, 209–219 c-command  164, 208, 215, 218–219 clitic  1, 3–9, 11–13, 15–17, 88, 168, 179–180, 182, 191–193, 195 clitic doubling  177–178, 180–182, 191, 194, 198 clitic left dislocation (CLLD)  177, 179, 184–187, 193–198, 200 clitic right dislocation  181 coargument  208, 213 coda  115–116, 119–120, 123, 125–126, 128–131 competition  25 creole  66, 68–69, 71, 73, 79, 155 culmination  65, 67–68, 76–78, 80, 83 culminated process  67–68, 76, 78, 80 cyclicity  25–26, 31–32

C Capeverdean  65–66, 68–71, 73, 75–77, 79–80, 83–85, 155–165, 167–169, 172 case  1–6, 8–10, 12, 17–18, 177, 180, 182, 198, 200, 204, 216–217 accusative case  1, 4–10, 13, 53, 66, 181–182, 200 dative case  1–6, 8–17, 203, 215–216, 218 genitive case  2–6, 13

D degree variable  54–58, 60–62 deverbal nouns  41, 63 diacritic  11, 31, 34 Differential Object Marking (DOM)  1, 12, 15, 17, 19, 177–180 Kayne’s generalization  177, 179–183, 186, 188, 200 distinctiveness condition  181 accusative Marker drop  182 diphthongization  21, 28, 30–33

Distributed Morphology (DM)  2, 6, 25, 177 ditransitive  4, 7, 12–13, 15–17, 156, 181 E ellipsis  168, 173, 177–179, 182, 186–189, 191–193, 195–200 TP ellipsis  155, 164, 166–169, 172–173, 179, 187–188, 193, 195–196 VP ellipsis  155, 158, 162, 164–169, 172–173, 195 vP ellipsis  188 pseudo-stripping  193 scope parallelism condition  192–193, 195 VP fragments  191–192, 195–197 Elmo Generalization  187–189, 191, 197 event  6–7, 13, 41, 43–51, 53, 57–63, 65, 67–69, 71, 75, 78–82, 85, 143, 204, 214–215 eventives  7, 43, 47, 65–68, 70, 72–77, 79–81, 85 event structure  65, 67–69, 78–80, 85 extrametricality  88 F faithfulness constraint  118, 120, 124, 127, 130 focus  35, 143, 145–146, 148, 150–151, 162, 178–179, 185–187, 191–193, 195–198 focus fronting  178, 182–183, 188, 196, 198 foot  88, 93, 100–101, 109–111 iambic foot  109 trochaic foot  109–110

 Index French  8, 16, 135–137, 139, 143–152, 158, 164–166, 168–169, 171–172, 211 functional category  157, 160, 172 G Galician  166–168, 171 gemination  115–117, 120, 125, 127–129, 131–132 generalized quantifier (GQ) givenness  135, 145, 147–151 G-operator  148–150 gradualness  118 Greek modern Greek  8, 11, 13, 102, 105 ancient Greek  105, 107 H Harmonic Serialism  115–118, 121–122, 127, 131–132 hierarchy  5, 59–61, 118, 120, 122, 125, 127, 132, 155–156, 163, 167 I inclusion  1, 3, 7, 15, 59–60, 75 inclusiveness  6, 16 information structure  135–136, 143, 145–148, 150, 152 intensity  90, 93 island  177–179, 183, 185–187, 189, 191–192, 196–197, 200 adjunct island  184–185, 196–197 interrogative island  184, 190, 194 island repair  177–179, 182–183, 185–187, 192–193, 195–197, 200 noun complement clause  183, 185, 190, 195 relative island  178, 184–185, 189, 196 sentential subject  178, 183–184, 190–191, 194 Italian  1–2, 6, 8–9, 11–13, 87–90, 93–94, 97, 99–105, 107–111, 158, 166, 168, 172, 205, 207, 209, 211, 218

L Latin  2, 6, 22–23, 102, 105, 107 learnedness  107 licensing head  157, 162, 165, 168 locality  21–22, 24–25, 27, 34–39, 150, 187, 205–207 locative  11, 14, 16–17, 44, 47–49, 51 M manner  44–45, 47, 49, 52–53, 59 markedness  120, 122–123, 125–127, 132 modifier  46, 51 mora  93, 100–101, 109, 118–120, 122, 127 morpheme  21–25, 28, 30–31, 34–39, 67–68, 72–73, 77–78, 81, 85, 159, 214 intonation morpheme  136–137, 143–145, 147–148, 150–151 Morpheme Interaction Conjecture (MIC)  37–39 temporal morpheme  66, 69–70 zero morpheme  65, 79, 81 morphology  1–3, 6, 9, 13, 17–18, 21, 23–24, 26, 30, 129, 168, 170–171, 173, 188, 200 morphophonology (MP)  22, 24, 28, 30–31, 34–35, 37–38, 62, 214 M-word  32 N nominalization  41–43, 47, 53, 59–63 Null Perfect  65, 68, 73, 78–81, 83–85 O obligation effect  204 onset  94, 115, 117, 119–121, 123, 125, 128, 130, 132, 139, 141–142 P parametric variation  155, 160–161, 163–165

person  1, 5–18, 22, 30, 33, 50–51, 66, 166, 180–181 Person Case Constraint (PCC)  1, 10–17 phase  26 phonology  22–38, 93, 100–101, 107–108, 111, 115, 118, 120–121, 129, 131, 160 pitch  90, 93, 100, 139, 141–143, 147 pitch compression  153, 151 polarity  155, 160, 162, 165–167, 171, 173 polar answers  157, 159, 161–163, 165, 169, 171–177 positive answerhood  143, 145–146 Portuguese  68–69, 137, 155, 157, 159, 161, 163–164, 166–169 Brazilian Portuguese  30, 164–165, 172 European Portuguese  157, 164–166, 172 presupposition  136, 143, 145–146, 148, 150 process phrase (ProcP)  59–63 prosody  91–94, 100, 108, 115–116, 118, 121, 127, 129–130, 132, 135–137, 146–152 prosodic domain  91, 149–151 pruning  36 pseudo-stripping see ellipsis Q quantificational domain  216 quantity  56, 100, 106 quantity sensitivity  100–101, 106, 111 R raising  21, 32–34 Quantifier Raising (QR) reference time  67–68, 73–74, 79–80, 82, 85 Relativized Minimality  10, 14–17 (anti-) repair effects  177, 179, 181–183, 185–190, 191–195, 197–201

Index  resumption  178–179, 183, 186, 188, 193–195, 198, 200 PF resumption  179, 186, 200 rising contour  135–137, 141, 144, 147–148 Romanian  2, 5–6 root  26–31, 33–38, 56, 62, 115–117, 120–121, 125, 127, 129–130, 132, 136 root-final  115–117, 125, 128–129, 132 S scope  30–31, 69, 80, 148, 162, 192–193, 195, 197, 203, 205–208, 210–219 scope parallelism condition see ellipsis semantics  42, 54, 58–62, 69, 74, 79, 81–85, 136, 148, 199, 205, 212, 215 sluicing  178, 187, 192–193, 195–197 Spanish  17, 21–24, 27–29, 32–35, 37–38, 41–42, 45, 47, 51, 62, 90, 106, 150, 158, 161, 163, 166, 168–169, 171–172, 178, 180–181, 195, 200 River Plate Spanish  177–182, 186, 188, 200 spirantization  117, 121, 123–124, 128–129, 132 split accusativity  1, 5–6, 10, 12, 15, 17

state  7, 41–44, 46–47, 49–56, 59–62, 65–71, 74–80, 83–85 D-state  41–42, 44–46, 50–63 K-state  41–42, 44–47, 50–54, 56–59, 61–62 perfect state  65, 68–69, 78–80, 82–85 state phrase (StateP)  59-63 statives  42, 47, 69, 71, 74, 79, 81, 84 stem  21–23, 27–30, 32–33, 35–36, 38, 62, 118 stem alternation  21–24, 26–27, 32–35, 37–39 stem storage (SS)  22–24, 27–28, 30, 33–34, 38 stress  29–31, 77, 87–93, 95–96, 99–111, 119 primary stress  88, 99–100 secondary stress  87, 99–100, 110–111 superiority  205, 207–208, 212, 215, 219 syllabification  115–118, 120, 125, 128–132 core syllabification  119–123, 125, 129–130, 132 syllable  31, 33, 87–89, 91–95, 99–111, 115, 118–123, 125–126, 128–132, 148 degenerate syllable  115 project syllable  118–119 syncretism  2, 5–6, 8, 62

T telic  61, 67, 78, 81, 83–84 temporal reading  45, 50, 52, 65–66, 72, 77–80, 84–85 tense  22, 25–26, 29–30, 33, 35–37, 39, 66, 73–77, 79, 82 TMA markers  159–161, 163, 165 V Vocabulary Insertion (VI)  25, 27, 34–36 Vocabulary Item  23, 25, 27, 29 voiced stop  115–117, 121, 123, 125, 128–129, 131–132 vowel  28–29, 31–33, 91–100, 105, 109–110, 115–117, 121, 125, 129–130, 132 vowel duration  87–88, 90–94, 96, 98–101, 109–111 vowel reduction  90, 97, 100 VP fragments see ellipsis V-relatedness  155, 160–161, 168 W Weak Cross Over (WCO)  179, 198–200 wh-questions  135, 138, 141, 145, 147, 152 wh-in-situ questions  135–152 Z zero operator  65, 67–68, 78, 81 zonal inclusion  3, 7