Available online 29 January 2015

e meaningerature on

www.elsevier.com/locate/lingua

Available online at www.sciencedirect.com

ScienceDirect

Lingua 157 (2015) 36--53verb meaning has focused on analysing verb uses when they describe events in the physical world. Consequently, theverbs combinatorial capacity has been assumed to be restricted by default to physical entities. As a matter of fact,however, any randomly chosen text that is representative of a language, for instance a press article, demonstrates that therange of combinatorial possibilities of a verb vastly surpasses the domain of physical entities. In this context it has longbeen acknowledged, but hardly further developed, that the choice of arguments can strongly affect the meaning of theWhat belongs to the meaning of a verb has long been a controversial issue in linguistic theory. Certainly, thof a verb determines the range of situations in the world that it can be used to describe. Much of the linguistic litAbstract

For many instances of verb-object combinations involving change of state verbs, different kinds of internal arguments seem to triggerdistinct interpretations of the verb phrase. These are usually divided into literal uses, such as romper la ventana break the window orcortar el papel cut the paper and figurative uses such as romper el desarrollo interrupt the development or cortar la circulacion cut offthe circulation.

Based on an extensive manual annotation of corpus (the used corpus is the El Pas Corpus consisting of all the El Pas newspaperissues from 1976 to 2007 and is hosted at the Insitut Universitari de Linguistica Aplicada (IULA) at the University Pompeu Fabra) datainvolving verb-object combinations with Spanish change of state verbs, I argue that combinations like romper el desarrollo or cortar lacirculacion, far from representing frozen idiom chunks, exemplify very productive compositional patterns. The frequency and naturalnesswith which change of state verbs take both physical and abstract entities as objects raises the question of how verbs express theirmeaning and makes this kind of data especially relevant for a theory of the lexicon as well as of composition.

I provide a clear inventory of the typical combinatorial patterns of romper and cortar and I show that their combinatorial behaviour ismuch more diverse than usually acknowledged. I then argue that these facts need to be addressed by the compositional system, ratherthan by postulating homophonic lexical entires (Dowty, 1979; Alonso Ramos, 2011) or contextualist accounts (Recanati, 2005).

For a proposal I turn to Modern Type Theories, which allow me to incorporate a richer notion of lexical semantics within compositionalsemantics. These theories thus allow me provide an insightful compositional account of what has long been considered non-compositional, namely combinations of change of state verbs with objects denoting abstract entities. 2015 Elsevier B.V. All rights reserved.

Keywords: Meaning shift; Change of state verb; Modern type theory

1. IntroductionSpanish change of state verbs in composition with atypicaltheme arguments: Clarifying the meaning shifts

Alexandra Anna SpalekUniversitat Pompeu Fabra, Roc Boronat, 138, 08018 Barcelona, Spain

Received 24 September 2013; received in revised form 24 December 2014; accepted 26 December 2014E-mail address: alexandra.spalek@upf.edu.

http://dx.doi.org/10.1016/j.lingua.2014.12.0070024-3841/ 2015 Elsevier B.V. All rights reserved.

verb phrase. Marantz (1984:25), for example, observed that just about every simple transitive English verb expresses awide range of predicates depending on the choice of the direct object. He illustrated this observation with the verb throw

ection preferences based on naturally occurring data available throughr

t theme arguments of change of state verbs easily range from physicals i ocesses and states, as in the case of romper break.1 This effect of the

r eoretical horizon that stretches from theories that resolve the problems in

A.A. Spalek / Lingua 157 (2015) 36--53 37the lexicon itself by arguing for a sense enumerative lexicon (Dowty, 1979; Alonso Ramos, 2011), to theories thatunderspecify verb meaning (Pustejovsky, 1995) and which favour resolving the problem in composition, up tocontextualist accounts on the other extreme, which claim that lexical meaning changes in context.

1 It is worth pointing out that all the corpus examples used throughout this paper have been additionally validated by Spanish native speakers.argument fo ces one to take a position on the th

object to d stinct kinds of eventualities like prWhat corpus data generally illustrate is tha

corpo a.

which I extensively tracked their argument selas in example (1), where baseball-throwing seems to trigger some other kind of predication than boxing-match-throwing orparty-throwing do.

(1) a. throw a baseballb. trow support behind a candidatec. throw a boxing matchd. throw a partye. throw a fit

It is worth mentioning that the data in example (1) do not seem to represent frozen idiom chunks, since sentence internalmodification is allowed in all cases. The possibility of inserting new elements between the verb and the object suggeststhat these are cases of free composition. What the examples put forward by Marantz (1984) thus show very clearly is thatthe nature of the objects in transitive verbs has strong semantic effects on the verb phrase. I will refer informally to thisphenomena as meaning shifts.

More recently researchers working with very large amounts of corpus data (Hanks and Jezek, 2008) have alsounderscored the importance of the semantics of the theme argument for the way the verb is understood. Among manyother things, these researchers make a basic and yet much understudied observation, namely that word meanings arestrongly context sensitive even with respect to their closest context, namely argument selection. Extending an observationby Asher (2011:viii) to these cases, it can be said that when word meanings are combined, the meaning of the result candiffer from what standard compositional semantics together with the words lexicographic definitions have led us to expect.

This effect of the argument on the interpretation of the verb phrase is directly related to a broader issue with thecompositionality of examples like the ones illustrated in (1). The question that arises is whether throw has a meaning alongthe lines of move the arm and hand and release an object. If that is the case, the verb can be argued to restrict itsargument selection for physical entities only. Accordingly, composition with other kinds of arguments can be claimed tocall for additional composition rules so as to accommodate the presupposition of a physical entity.

Yet another way of tackling the combinatorial variety of verbs, as also pointed out by Marantz (1984:25), is to try todistinguish the basic or literal uses of the verb in (1) from its metaphorical and idiomatic use. This is the path taken bycognitive linguists from the tradition of Lakoff and Johnson (1980). Such an approach principally distinguishes betweenuses like the one in (1-a) and the remaining combinations of throw. In this paper, however, I address these facts as aproblem of composition and seek to situate their analysis within the theoretical linguistics literature.

In the theoretical literature, the questions raised by this broad combinatorial capacity of verbs, with its effects on thesemantics of a verb phrase, have found few answers. On the one hand, theoretical linguists have avoided the problem byeither considering that lexical items enter the compositional process after disambiguation, as is the case for the Montagueschool linguists, or have mainly limited their data to the set of examples that apply to the physical world. On the other hand,many other approaches, fundamentally belonging to the Meluk school, have classified this kind of data as non-compositional in nature (Alonso Ramos, 2011). The two approaches basically rely on two distinct ways of viewing thelexicon and the process of composition. While in the first case a sense enumerative lexicon guarantees a neat process ofsemantic composition, in the second case examples like the ones illustrated in (1) are mostly consigned to theunfortunately too broad and too uncertain terrain of figurative speech, metaphors, idioms and collocations.

The challenge posed by verb-object combinations, like the one illustrated in (1), thus lies in analysing what kind ofcombinatorial capacity a verb actually has, whether it naturally ranges beyond the domain of physical entities, and if so, ifthese combinations can be constrained by rules.

For the present investigation I focus on the combinatorial capacity of change of state verbs in Spanish. Moreconcretely, I use a sample of two crucial members of this class, namely romper break and cortar cut, for

I argue against approaches that postulate homophonic lexical entries as well as against contextualist theories thatestablish the meaning in discourse context and instead account for the rich combinatorial paradigm of verbs through theidentification of the core meaning of the analysed verbs that is maintained throughout the different combinatorial patterns.

Th

e i o d h e n

p ,

efined to describe events of separation in material integrity involving some kind ofd Keyser, 1986; Levin, 1993). This definition perfectly applies to examples like (3) for

n

A.A. Spalek / Lingua 157 (2015) 36--5338(3) [. . .] la polica corto las cadenasthe police cut the chains

The police cut the chains. [El Pas Corpus]

Nonetheless, a look at corpus data reveals that these verbs actually also take a whole range of theme arguments that lackmaterial properties, as illustrated in examples (4) and (5), a fact not accounted for by the definitions provided in the literature.

(4) El cortejo llego hasta el cementerio y en ningn momento se rompio el silencio.the procession arrived until the cemetery and in no moment REFL broke the silenceThe procession reached the cemetery and at no time was the silence broken. [El Pas Corpus]

(5) El tcnico britnico afirma que su decision corta definitivamente cualquier especulacion sobrethe technician British claims that his decision cuts definitively any speculation aboutun posible acuerdo con el Barcelona.a possible agreement with the BarcelonaThe British soccer coach claims that his decision definitively cuts off any speculation about a possible agreementwith the Barcelona Football Club. [El Pas Corpus]

Generalising over the corpus data, romper can denote events ranging from a sense of destruction of the physical integrityof an object, as in (6) and (7), to denoting events that affect the existence of states2, as in examples (8), or processes (9).

2 For this study I classify a state as the presence of a property in time. A more detailed discussion concerning the question what kind of abstractnouns can be classified as states lies beyond the scope of this paper.the Spanish equivale t cortar.

motion, contact and effect Hale an

Similarly, the verb cut has been d

((2) Juan rompio el jarron.Juan broke the vaseJuan broke the vase.physical make-u of the theme as in (2).

felicitously account for th uses of th Spanish

take English break as a verb that lexicalises a particular change in the material integrity of an entity. Such definitions do

e e equivalent romper, where the verb does in fact describe a change in thelanguages, like Spanish romper and cortar, as typical members of this class: these verbs have been described as verbsthat denote actions which bring about a change in the physical state of an entity. Focusing first on verbs of breaking,Achard (2006), for example, describes French rompre as a verb of destruction, while Rappaport Hovav and Levin (2005)Bohnemeanimeyng sr, 2007; Majifts, neverthid et al., 2eless, ha008). Tve reche divedversity of com little attentiobinatn. Coorialnsid patternser break of change and cut af state d theirverbs ancounterp theartscorrelatedin variousre, 1967; Guerssel et al., 1985; Levin, 1993; Haspel ath, 199 ; Achar , 2006; Ra paport Hovav and Levin, 2005;(Fillmo

e class of change of state verbs has been widely studied with respect to their argument alternation possibilities

m 3 d p

Th2. Empirical generalisationsFinally sec ion 5 concludes.t

Section 4 uts forward an analysis in erms of restrictedp t predication elaborated for two Spanish change of state verbs.

the possible accoun s for meaning shifts and argues fo

taking paper is structured as follows. Section 2 offers an overview of the empirical generalisations over the attested data,into account both the syntactic alternations as well as the range of possible theme arguments. Section 3 surveys

t r an underspecified view of the lexicon as an unifying account.eIn addition to underspecified lexical entries, I use dependent types extending a suggestion used by Luo (2011a) to capturethe effect of the theme on the interpretation of the verb. This account provides a better handle on the combinatorialcapacity of change of state verbs and how they behave in composition.

(6) Juan rompio la ventana.Juan broke the window

Juan broke the t-shirt

t lived in the first monthss hich interrupted the positive evolution experienced in the

meaning of the overall result of a romper verb phrase, allowing for a destruction result reading, as in example (6), up to ad (9) respectively.3

i s r objects denoting some kind of

as in (11) through (13).

o

d the parliamentary processes. [El Pas Corpus]

o n notes the interruption of an action under performance or an ongoing process, such

A.A. Spalek / Lingua 157 (2015) 36--53 39

h I ussion of the use of romper with natural phenomena, such as las olas rompen en las rocas.3 For t e sake of space leave out the discas entrega delivery or respiracion breathing. But even state-denoting nominals, such as silencio silence are common

as objects of cortar verb phrases.In these kinds f combi ations cortar deThe call for elections interrupte(13) La convocatoria electoral corto el proceso parlamentario.

the call electoral cut the process parliamentaryThe sobbing was taking his breath away and caused a lump in his throat. [El Pas Corpus](12) [. . .] los sollozos le cortaban la respiracion y le anudaban la voz en la garganta

the sobs to him cut the breathing and to him knot the voice en the throatFrance wo ld stop delivering of TA members. [El Pas Co pus] u E r

France cut.IMP the delivery f ETA members(11) Francia cortara la entrega de etarras.(10) [. . .] quin sostendr las tijeras que corten la cinta inauguralwho support.FUT the scissors that cut.SUBJ the band inaugural

who will hold the scissors that will cut the inaugural ribbon. [El Pas Corpus]

Corpus evidence, however, shows that cortar also appears in combinations with all kinds of action nouns and processes,incision event, such as in example (10).

Sim lar ob ervations apply to co tar, which is typically described as applying to physicalceasing result reading that affects a particular state or a process, as in examples (8) anfirst months. [El Pas Corpus]

These changes in the semantic type of the theme argument contribute very decisively to influencing a change in theIn the la t trimester the sales of housing stagnated, w

broke he evolution positive that REFL had

rompio la evolucion positiva que se haba vivido en los primeros meses.(9) En el ltimo trimestre se produjo un estancamiento en la venta de viviendas que

in the last trimester REFL produced a stagnation in the selling of livings thatFraga broke his political silence yesterday. [El Pas Corpus](8) Fraga rompio ayer su mutismo poltico.

Fraga broke yesterday his silence politicalJuan tore the t-shirt.Juan broke the window.

(7) Juan rompio la camiseta.

.t r

. Thus, a location like highway in example (16) can refer to the event of traffici w sition with cortar the event referred to is the blocking of the highway, rather than

nal Madrid-Andalusiak adrid-Andalusia highway [El Pas Corpus]

rgument also involves different argument realisation possibilities. For instance, e e that has considered change of state verbs and particularly the break-verb class the

e

romper impose few restrictions on the causep restrictions, as the following contrast illustrates.

A.A. Spalek / Lingua 157 (2015) 36--5340Againstargume recent, soived me vwisdom, erb-objecaccot comrding to binationswhi seech bream to imk-typosee ver very cbs like oncreteThe choice of the intern l argument al o has con equ nces for the

sugges ed for Italian anticausatives (Chierchia, 2004).

a s s e choice of the possible cause of romper-events.

tThe norm broke by itself.

It is worth emphasising at this point that, since the clitic se can be used to form reflexive passives in Spanish, the idiomaticanaphor por s solo/a is used in (20) to deactivate the reflexive passive reading. Por s solo/a in this case contributes toconstruing norma as the sole cause of the event by identifying the cause and the theme as one unique argument, a testThe norm REFL brok by self alone

(20) #La norma se rompio por s sola.(19) Juan rompio la norma.Juan broke the normJuan broke the norm.The window broke.

the window REFL broke(18) La ventana se rompio.core class of causative-anticausative alternating verbs (Jespersen, 1927; Levin, 1993; Levin and Rappaport Hovav, 1995;Reinhart, 1996), certain object choices of romper preclude the anticausative variant. This fact is illustrated by the followingcontrast, where norma norm does not allow for an anticausative form.

(17) Juan rompio la ventana.Juan broke the windowJuan broke the window.despite the oft n-stated gen ralisation

the alternation in the type f object a

phrase and that romper break and cor

oSummarising, corpus data show that the different kinds of internal arguments trigger distinct interpretations of the verb

tar cut can easily describe both physical as well as abstract changes. Interestingly800 people who bloc ed the M

800 persons who cut the natio(16) [. . .] 800 personas que cortaro literally cutting the highway.n la nacional Madrid-Andalucacirculat ng on the high ay. In compo

rather than effectively denoting one

In many cases the entity in theme poThe Joventut basketball team ended its losing streak. [El Pas Corpus]sition is actually coerced into an eventuality, that is, understood as an eventuality,the Joventu cut his se ies of defeats

(15) El Joventu corto su racha de derrotast(14) [. . .] el silencio se cortaba a veces por una tosthe silence SE cut.IMP to times by a cough

The silence was sometimes interrupted by a cough. [El Pas Corpus]

When states are the affected themes of a cutting event the verb conveys the ceasing of a state which was susceptible tocontinue. Yet another related use of cortar appears in (15), where racha de derrotas refers to a sequence of timecharacterised by losses.

(21) Juan/ el hacha/ el huracn/ el peso de los libros/ la explosion rompio la mesa.Juan/ the axe/ the hurricane/ the weight of the books/ the explosion broke the tableJuan/ the axe/ the hurricane/ the weight of the books/ the explosion broke the table. (Mendikoetxea, 1999:1589)

law/

n eaking events take causes that can range from agents, natural forces,

per-events, however, is not an across the board restriction, since there areow for a whole range of causes, as example (23) illustrates.

omico de Espana se rompio.omic of Spain REFL broketh stopped.

a The same explanationl S a

A.A. Spalek / Lingua 157 (2015) 36--53 41inventory of arguments collected in REDES were judged too fine-grained to provide sufficiently broad generalisations, such that the combinatorialcapacity of change of state verbs could be restricted to a very limited group of combinatorial restrictions.Spanish combinatorial ictionary REDES, which incl des he verb cortar cut Bosque, 004) Nevertheless, the notional groupings 4 It is worth mentioning that the present study of the combinatorial capacity of verbs has been preceded by a previous close exploration of thed u t ( 2 . of thecortar is regularly at ested in corpus, though t is is only the case for c rtain semantic classes of themes. Concretely, t

trary to what has been c aimed in the literature abo t cut and cortar, the caus tive anticausati e alternationt h e he

Yet, con l u a - v ofThe cloth

the cloth SE cut

cut.(25) a. El sastre corto la tela.the tailor cut the clothThe tailor cut the cloth.

b. #La tela se corto.of agencould bt-oriene appted mied feaninor the g components,panish equiva has led lent cortreser, asrchers to claim the following that this veexample illusrb istrate agentive.s.dged lack of the anticausative form or cut (Haspelmath 1993:93), w ich is usually aThe acknowle f , h ssociated with the lackb. #The cloth cut.

The tailor cut the c oth.(24) a. lWhat becomes obvious with this example is that the difference in argument realisation possibilities is not strictly correlatedwith the difference between physical and abstract theme arguments, contrary to a claim put forward by Pin on (2001).Rather, it is due to the fact that norms and limits belong to the world of human relations and social agreements. In addition,a broader generalisation that emerges from the contrast between (21) and (22) is that it is not the verb alone that decidesthe thematic nature of the subject.

Turning now to the case of cortar and how the difference in object arguments affects argument realisation possibilities,we also find some argument alternation facts that have received almost no attention in the literature.4 English cut, forinstance, has been claimed to lack the anticausative variant (Guerssel et al., 1985; Levin, 1993; Haspelmath, 1993;Bohnemeyer, 2007), as illustrated in example (24-b), taken from Haspelmath (1993:93).the development econSpains economic growb. El desarrollo econ(23) a. La crisis inmobiliaria rompio el desarrollo economico de Espana.the crisis property.ADJ broke the development economic of SpainThe housing crisis interrupted the economic development of Spain. [El Pas Corpus]respectactuallyto the many caus abse argumtract breent of akingsabstr thatact rom do allnd stative eventuali ies t eveninstrum nts a

shows that while physical br

t oe ts of all kinds, abstract breaking events require agents. This restriction with

The co trast #the wheel/ #the accident/ #the width of the swimming pool/ #the explosion broke the law/ the norm.Juan

the norm/(22) Juan/ #la rueda/ #el accidente/ #la anchura de la piscina/ #la explosion rompio laJuan/ the wheel/ the accident/ the width of the swimming pool/ the explosion broke theley/ la norma.

ya que se corto laalready that REFL cut the

i

va accion terroristas u action terrorist

them, as in example (29).

ies of four defeats consecutive

s

i c

A.A. Spalek / Lingua 157 (2015) 36--5342All these anticausative variants describe events that can arise from within the entity denoted by the unique argument. Thatis, these corpus data show that cortar events are not necessarily agentive, which explains why they allow for theanticausative variant.The rope rok while he was bungee jumping. [We ]

REFL to him cut the rope while doing.3.SG bungeeb e b(33) Se le corto la cuerda cuando haca bungee.also po

sative-anticausative variant can even appear with other physical object themes, a fact rarely acknowledged, asinted out by Rappaport Hovav and Levin (2010).

uThe milk sours 48 hours after leaving the dairy. [El Pas Corpus]The cathe milk REFL cuts past 48 hours s nce its exit of the entral

(32) [. . .] la leche se corta pasadas 48 horas d sde su salida de la entrale c(31) [. . .] se comprobar como se corta la mahonesaREFL verify.FUT how REFL cut the mayonnaise

You will see how the mayonnaise separates. [El Pas Corpus]similar ubst nce, as illustrated in t e follo ing examples.

Finally, food-denoting nouns also naturally combine with cortar in the anticausative, particularly if the food is a liquid or

a h w

The bad spate of four consecutive defeats has ended. [Web]

REFL cut the bad ser(30) Se corto la mala racha de cuatro derrotas consecutivas.The phone got cut off. [El Pas Corpus]

Furthermore, cortar in composition with themes that denote temporal sequences of events can also appear in thecausative-anticausative alternation.REFL cut the phone

(29) Se corto el telfono.(28) [. . .] a las 23.30 horas se corto el gasa the 23.30 hours REFL cut the gas

At 23.30 the gas went off. [El Pas Corpus]physical objects that are associated with a service that is supplied through

The alternation is also possible in cases where cortar combines with physicaThe process of dialogue would be cut off as soon as a new terrorist action took place. [El Pas Corpus]l substances that are supplied, as in (28), orthe process of dialogue REFL cut.IMP a soon as REFL prod ce a new

(27) [. . .] el proceso de dilogo se cor ara e cuan o se prod jera una nuet n t ucommunication at onceEtxegarai did not have time to contribute any more clarifications, since the communication was immediately cutoff. [El Pas Corpus]comun cacion enseguida.

Etxegarai no have time of ontribute more clar fications (26) Etxegarai no tuvo tiempo de aportar ms precisiones, c icausative-anticausative alternation is most frequent for cortar when it combines with process-denoting nouns, as inexamples (26) and (27).

s enoting nouns, as the following (b)-examples illustrate,w

f g o El Pas Corpus]

i elf alone

Cr p o

thematic ro the theme, powerfully illustrate that syntactic realisations

e t e with distinct theme arguments, Spanish romper ande ibe events of separation or division of physical entities,

e m o p h ey affect entities as well as eventualities which can bet e s t idea of connectedness of an entity has been associated

A.A. Spalek / Lingua 157 (2015) 36--53 43with a continuous path between every part of the object (Viberg, 2007) and can equally well be applied to the domain ofeventualities. This is why both romper and cortar seem so similar when they affect objects and events. When physicalobjects are affected by one of these predicates, the event describes a bringing about of some state of disruption ofboth prjudged dico bates a conncross thected ineir co their nbinatormal rial tateossibor devilities is telopmenat th. The events f cessation when applie to evenwhile they r

to have closely related meanings in that they descrefer to o d tuality-denoting themes. One semantic element that underliescortar s

ing emto th semantic con ent of the verbs as they combinTurn

the semantics of the them argumentdepend on

les for the subject in dependence on the semantics of

e .the bridges REFL cut by self aloneThe bridges cut off by themselves.

All these uses of cortar with action-denoting nouns or nouns that are understood as events require an agent to make thecutting event interrupt a particular kind of conscious action, such as the huida flight in example (35), the entrega delivery,in example (36), or even some kind of event associated with a particular location such as puente bridge in (37).

Putting all the data together, namely the distinct argument alternation possibilities as well as the restrictions for certainb. #

TheLos police cut o puentes sff thee bridgecortas. [Eon l Pas or s orps

us]los.the police cut the bridges

(37) a. La polica corto los puentes.erstood as events.which are und

ubgroup of themes that precludes the anticausative alternatiThe extradition of ETA members stops by itself.Yet another s on is represented by location-denoting nounsthe del very of ETA members REFL cuts by s

b. #La ent ega de etarras se corta por sFrance stops the extradition of ETA members. [El Pas Corpus]r sola.France cuts the delivery of ETA members

(36) a. Francia corta la entrega de etarras.b. #La huida de Hernando se corto por s sola.the flight of Hernando REFL cut by self aloneThe flight of Hernando cut off by itself.A fan cut of the fli ht of Hernand . [

a fan cut the fligh

Un aficionado [. . .] corto la huida de Hernando.

t of Hernando

(35) a. one of hich is repeated from abov (11).

The ame agent-restricton arises for cortar-events with action-d

eThe anticausative variant of cortar events, however, is precluded in certain cases, which, in fact, subcategorise foragents, as illustrated in the following example.

(34) a. [. . .] Sergio Ruiz corto la tarta nupcial [. . .]Sergio Ruiz cut the cake nuptial

Sergio Ruiz cut the wedding cake. [El Pas Corpus]b. #La tarta nupcial se corto por s sola.

the cake nuptial REFL cut by self aloneThe wedding cake cut by itself.

This example illustrates that food cutting events, though not necessarily involving one particular instrument or a particularmanner, as defended by Bohnemeyer (2007:159), are often associated with handling a specific type of instrument, whichin turn naturally triggers an agent. That is, these kinds of themes lack any internal properties that could initiate a cuttingevent.

level chang

events

f r i a p a l r w a e

a a e , as I will lay out in section 4.p d e es by drawing on the critique

f e rn to the Sense Enumeration Lexicon

A.A. Spalek / Lingua 157 (2015) 36--5344(henceforth SEL). According to this model, all possible meanings of a lexical item are listed as part of the lexical entry. Inthe case of the verb romper the proposal could be to postulate distinct lexical entries with differing combinatorialrestrictions, such that we could imagine a romper1 that selects for physical objects and means physical destruction, as inexample (38), a romper2 that selects for processes, as in (39), and means interrupt, and a romper3 that selects for states,as in (40), and conveys the meaning stop existing.

(38) Juan rompio la ventana/ un juguete/ el papel/ la camisa.Juan broke the window/ a toy/ the paper/ the shirtJuan broke the window/ the toy/ tore the paper/ the shirt.

(39) La crisis inmobiliaria rompio el desarrollo economico de Espan a.the crisis property.ADJ broke the development economic of SpainThe housing crisis interrupted the economic development of Spain. [El Pas Corpus]

(40) El cortejo llego hasta el cementerio y en ningn momento se rompio el silencio.the procession arrived until the cemetery and in no moment REFL broke the silenceThe procession reached the cemetery and at no time was the silence broken. [El Pas Corpus]

Formally, an SEL can be defined as follows (Pustejovsky, 1995:34):

A lexicon L is a Sense Enumeration Lexicon if and only if for every word w in L, having multiple senses s1, . . ., snassociated with that word, then the lexical entries expressing these senses are stored as fws1; . . . ; wsng.

A dictionary entry represents a typical example of an SEL. Within a sense enumeration model two ways of storinglexical entries are conceivable. When the senses are contrastive, such as in the case of bank as an institution and bank asa piece of furniture, the senses are stored independently. When, however, there is a direct relation between the sensessuch as in the case of lamb as count noun or lamb as mass noun the senses are often understood to be stored ascomplementary: wfs1; . . . ; sng. In an SEL composition is understood to take place with disambiguated lexical units, whichallows the process of composition to stay simple. Lexical entailments of predicates thus simply amount to stating that averb such as romper denotes the event of breaking. Concerning the distinct syntactic patterns a verb can appear in, asOne way of accounting or the meaning shifts I have just present d is to tu3. Possible approaches to meaning shiftsprovided by Asher (2011).

Before rovi ing my account, however, I will survey the most important alternativ approach

possibilities to illustrate that langu ges encode sophisticated system of s mantic types

system is required. Together with previous research (Asher, 2011; Luo, 2011a), I take these patterns and combinatorialof an external cause or if a possible cause is restricted to a particular thematic role. These patterns of verbal alternations

and the combinatorial variety should thus be accounted for by a theory of predication. To do this a much more flexible typesubject to very fine-grained semantic crite ia, which in turn determines hether particular vent can occur independently

semantic dif erences and the diffe ences n argument realis tion ossibilities in fact illustr te that argument rea isation isOve

, convey more manner than result meaning.rall it can be said that the theme contributes to establishing a concrete meaning for the verb phrase. These fineincrementality, while a less predictable location of impact, as in the case of romper, is associated with a less clearlyincremental event. I associate this predictable incrementality with the fact that cutting events, in contrast to breakingrequires an instrument and thus results in the description of a more controlled action that makes the trajectory ofe predictable. The predictability of the trajectory of change, in turn, seems to evoke a clearer notion ofphysical integrity. When eventualities such as states and processes are affected, the disconnection occurs at the level oftemporal connectedness in the expected development. The contrasting semantic element is that romper tends to beregarded as an event that brings about an uncontrolled kind of disconnection. Cortar, in contrast, describes a morecontrolled event of disconnection, as has been extensively argued by Majid et al. (2008). In the physical domain, the factthat the event of disconnection comes about in a controlled manner is often associated with the use of an instrument andthe presupposition of an agent. When cortar describes events with prolonged duration, the event has a more clearlypredictable trajectory than do romper events. The reason is that cutting is relatively smooth and straight and at a physical

the fact that a semantic theory based on sense enumeration is a perfectly licitr an independent analysis of syntax and semantic interpretation, it represents a

s ing collapses. As a consequence they defend that enrichment is necessaryr to what the meaning is. Contrary to the more radical contextualists Searles

in as far as he acknowledges the importance of the arguments of the verb,

A.A. Spalek / Lingua 157 (2015) 36--53 45while he still defends a basic lexical meaning. In the following example Searle (1980:222--223) basically argues that inorder to determine the exact way of cutting we need to know what is being cut, an argument that goes in the line withthe observed corpus data.to deteview semine ems tthe conditions with respect o be much more moderateddefend that the minimal bottom-up meanvery poor model of lexical description. Each lexical item simply has a unique selective behaviour and a particularsyntactic pattern. No broader relations are established between the kind of themes a verb selects for and the syntacticpatterns these combinations allow for or which they systematically preclude. Analysing the fact that the same verb canshow very different selection restrictions or may preclude certain syntactic realisation patterns depending on the typeof theme it combines with as a matter of merely correctly matching the functor and the argument from a list of lexicalentries represents a very simplistic view of the meaning of words and their interaction in composition. In other words,considering the lexicon a pure plug-in module basically gives the impression that there is not much to say about wordmeaning. This is the reason why researchers like Pustejovsky (1995) and Asher (2011) have argued that basing asemantic theory on a sense enumeration model amounts to saying almost nothing about how meanings combine in theprocess of predication.

Yet another argument employed in the literature against a sense enumeration model is based on the observation thatan SEL greatly exaggerates the discreteness of fixed word senses. This understanding of meaning as atomic units hasbeen argued to completely miss out that meanings overlap in diverse ways. The fact that in the case of romper the numberof listed meanings across dictionaries varies between 20 and 30 senses precisely illustrates the difficulty of distinguishingdiscrete word senses. And the fact that there is effectively a lot of overlapping in the core semantic component shows thatromper does not seem to be 20--30 ways ambiguous. All these facts render an SEL a poor model for natural language.Thus, in order to obviate the need of multiple listings of words as an answer to meaning shifts in composition, it isnecessary to turn to another theoretical approach.

Another way of providing an analysis for the observed data is to take meaning fluctuation in context as evidence for acontextualist approach. Contextualists (Recanati, 2005) maintain that the content of the sentence is not fixed only bylinguistic rules and lexical content, but also by the context. They argue for the consideration of pragmatic rules in thedetermination of truth conditions, which furnish a set of maps from one lexical meaning to another. Specifically,the contextualist approach argues that there are many pragmatically controlled processes, which are not triggered by thelinguistic sign, but rather by the context. Thus, for example, the process of transfer as illustrated in example (42), takenfrom Nunberg (1995:115), is typically considered as pragmatic and is only understood in the context of a bar where thewaiter refers to a client in terms of the food the client ordered.

(42) The ham sandwich is at table 7.

In general, a crucial process argued for by the contextualist approach is free enrichment, which, as its name says,allows to enrich freely the literal content of an utterance. Though triggered by a linguistic expression, free enrichmenttakes place to make the interpretation of an utterance more specific than its literal interpretation. It is considered a top-down, pragmatically controlled process that affects the truth conditions of utterances. The contextualist approach goesas far as to claim that content is only expressed in the context of a determinate speech act and thus fundamentallydenies that the interpretation of propositions in a bottom up way makes any sense without contextual modulation.Semantic content is thus assigned to simple expressions in context (Recanati, 2005:178). The contextualist viewjust ranmodel odom but are limited. Thus, despite f the lexicon, which has allowed foillustrated in example (41), Montague Grammar has postulated that each grammatical use of a verb should be registeredin a separate lexical entry.

(41) a. John wants to have a car.b. John wants a car.

According to this model the relation between the uses of a given verb is captured via meaning postulates, such that there isno deep semantic relation among them. In distinct ways, to date this is still the predominant model in the linguistic literature(Dowty, 1979; Levin, 1993; Alonso Ramos, 2011) as well as in philosophy (Fodor and Lepore, 1998) and it is crucially themodel of the lexicon that goes together with the simple type system envisaged by Montague Grammar and the standardcompositional model.

And yet, as soon as one consistently analyzes corpus data, it becomes evident that the senses of romper or want aresystematically associated with the internal complements of the verbs and the semantic classes of possible themes are not

and though the word [cut] is not ambiguous, it determines different sets of truth conditions for the differentsentences. The sort of thing that constitutes cutting the grass is quite different from, e.g., the sort of thing thatconstitutes cutting a cake. One way to see this is to imagine what constitutes obeying the order to cut something. Ifsomeone tells me to cut the grass and I rush out and stab it with a knife, or if I am ordered to cut the cake and I runover it with a lawnmower, in each case I will have failed to obey the order. That is not what the speaker meant by hisliteral and serious utterance of the sentence.

More in the line of Searle (1980) I defend that, in order to determine the verb phrase meaning, we do not need to appealto background assumptions and world knowledge, as suggested by the contextualists, but actually do well enoughproviding an underspecified verb meaning, which is fully specified in company of a specific argument.

Underspecification is a formal device for handling multiplicity of meanings in lexical semantics. In this model,lexical representations, though bringing in a rich internal semantic structure, are not fully specified with respect to theirmeaning until they enter the process of composition. Rather, lexical meaning is treated more like a potential that onlyfully fleshes out in composition, with both predicates and arguments being able to interact in complex ways withspecific aspects of the internal semantic structure. This approach was most prominently taken in the GenerativeLexicon Theory (hereafter GL) developed by Pustejovsky (1995). In contrast to the sense enumeration model, whichfavours a simple and straightforward compositional model at the expense of a simplistic view of the lexicon and incontrast to the contextualist model that minimises lexical meaning in exchange for a complex compositional process,GL takes an intermediate route. In GL lexical entries are slightly underspecified but articulated enough to allow for

as in example (43-b).

A.A. Spalek / Lingua 157 (2015) 36--5346Fig. 2. GL representation of cake; pag. 123.b. John baked the cake.

For the concrete analysis GL proposes that the complement carries information that acts on the governing verb, while it isassumed that bake has one underspecified sense, as represented in Fig. 1. Crucial here is that bake carries an agentivequale, which initially is specified to a state change lexical conceptual paradigm (lcp). That is to say, by itself bake is notsupposed to lead to a creation reading.

To understand how the creation reading for bake a cake arises, it is necessary to turn to the semantic structure of thetheme cake, illustrated in Fig. 2. Cake also includes an AGENTIVE feature in its qualia structure; here this feature conveysinformation about how cakes are brought into existence.

Fig. 1. GL representation of bake; pag. 123.(43) a. John baked the potato.more complex, but still restricted, compositional processes. In addition to the considerably more fine-graineddistinctions in the semantic content of lexical units Pustejovsky (1995:45) employs some generative devices, such asco-composition. Co-composition permits for a stronger involvement of the object as to determine the preferredreading of a particular verb phrase, which allows GL to account for certain meaning shifts of verbs Pustejovsky(1995:122). To give an example, Pustejovsky (1995:122) argues that the verb bake allows for two readings, a changeof state reading and a creation reading, depending on the semantics of the theme. Thus potato combined with bake,as in example (43-a), receives a change of state reading, while cake combined with bake leads to a creation reading,

G ning that results from the operation of combining the AGENTIVE quale of

A.A. Spalek / Lingua 157 (2015) 36--53 47

bake a cake; pag. 125.the noun with the AGENTIVE quale of the verb. What is important to emphasise here again, is that GL allows explaining howthe verb bake, which is initially specified with a change of state lexical conceptual paradigm, can convey a creationmeaning, as in the case of bake a cake. With respect to the process of composition this operation represents a morecomplex process than function application, namely what Pustejovsky calls co-composition. Bake applies to itscomplement, yet this operation requires the argument to co-specify the meaning of the verb phrase. Thus, according tothis system, when we unify the typed feature structure of bake with the one of cake the qualia structures compose andderive the sense of the verb phrase returning a new typed feature structure where the common arguments of the two typedfeature structures are replaced with what represents the greatest lower bound in the type hierarchy. In this case theunification of the qualia leads to a creation lexical conceptual paradigm, as illustrated in Fig. 3.

Potato, in contrast, being a natural entity, lacks the AGENTIVE quale that is characteristic of artefacts and all man-createdobjects. As a consequence combining bake with potato does not derive a creation sense, but rather a pure change of statemeaning. This is how both uses of bake can be derived by involving the semantics of the noun phrase into the compositionprocess.

This way GL represents an interesting approach to the process of meaning composition by selecting components oflexically specified meanings. That is, the exploitation of the qualia structure focuses on certain components of lexicalmeaning of a noun and how they interact with the predicate. Most importantly for the present approach, however, thequalia system is insufficient to account for the fact that verbs like romper can combine with nouns of totally distinct kinds,such as physical objects and events. In other words, GL composition process is too rigid and consequently breaks downwhen faced with data like those at issue here. In addition, it is totally unclear how the Generative Lexicon Theory wouldexplain the fact that the argument selection of the object affects the selection of the agent of the event as well as theargument alternation possibilities, as the contrast between (44) and (45), repeated from (6) and (22), illustrates.

(44) a. Juan rompio la ventana.Juan broke the windowJuan broke the window.argume(45)

Summinreasonsnt. In b.

a.

b. #

g up, to preL bakLa vthe wThe w(Repe

Juan Juan Juan La lethe laThe la

the sufer thee a cakeentana indow indow bated from

rompiobroke broke they se w REFLw broke

rvey of undersp getsse REFLroke

abo

la the

lawrom

bro by it

alternecific a creationrompio broke by itself.ve (6))

ley.law.pio por ke by self.

ative appration to th mea

In composition the qualia are unified and give as a result a particular reading of the verb that is co-specified by thepor by

s self

oache sens self

solaalon

es tose enFig. 3. GL represen ation of tsola.alone

.e

the treatment of meaning shifts at verb phrase level providesumeration as well as to the contextualism approaches. Crucially

the sense enumeration approach represents a very poor lexical theory, while the contextualist approaches leave themeaning shifting possibilities too unconstrained as to be explanatory. However, the best known underspecificationapproach, namely GL, provides lexical entries that are too rigid and thus is not adequate to accounting for the dataexplored here.

e hurricane/ the weight of the books/ the explosion broke the table

w h

g predication, then this is an argument for the need for a richer such as human, physical-object, etc. In addition, more recent

f an argument to, in a sense, influence the interpretation of the

A.A. Spalek / Lingua 157 (2015) 36--5348predicate it combines with. Consider, for example, the proposals in Luo (2011b), Asher (2011), Chatzikyriakidis and Luo(2013) and Asher and Pustejovsky (2013). Such type theories, now common in computer science, allow for making fine-grained distinctions between sorts of individuals (Chatzikyriakidis and Luo, 2013). Here I follow Luo (2011a) in callingthese rich type systems Modern Type Theories (hereafter, MTT). MTTs (Martin-Lf, 1984; Ranta, 1994; Chatzikyriakidisand Luo, 2013) allow fine-grained distinctions to be made within a given semantic type. In these theories (Martin-Lf,1984; Ranta, 1994; Chatzikyriakidis and Luo, 2013), propositions are conceived of as types, not as sets of worlds.Sentences are thus interpreted as propositions of type Prop. MTTs are many-sorted in the sense that individual verbs andcommon nouns are each assigned distinct types (e.g. vmanb and vwomanb are distinct types)5, rather than all belonging,for example, to the type he, ti. That is, common nouns are interpreted as types.6 For example, the noun man is notinterpreted as a set of entities, but rather as a type vmanb. Moreover, these types can stand in subtype relations to eachother. For example, vmanb can be defined as a subtype of vhumanb. The granularity of the many-sorted type systemsdepends on each theory, ranging from one type per lexical item on the one hand to complex type hierarchies on the other.

5 In these theories, the denotation brackets (vb) indicate types rather than objects in a model theory.6 For a detailed discussion on common nouns as types see Luo (2011a).logics have developed techniques for allowing the type o

system of types to guide predication, including subtypes

their ar uments and meaning adjustments appear with Juan/ the wheel/ the accident/ the width of the swimming pool/ the explosion broke the law/ the norm.

What emerges from the contrast between (47) and (48) is that the semantics of the themes that romper combines with hasa strong influence on the possible subjects and the type system presented in (46) does not account for the fact that anargument can influence the semantics of the resulting predication. If, however, predicates are sensitive to the semantics ofthe la / t e norm

la le / la norma.

Juan/ the wheel/ the accident/ the width of the swimming pool/ the explosion broke

y(48) Juan/ #la rueda/ #el accidente/ #la anchura de la piscina/ #la explosion rompioJuan/ the axe/ the urricane/ the weight f the books/ he e plosion broke th tab e.h o t x e l

Juan/ the ax / the(47) Juan/ el ha ha/ el c huracn/ el peso de los libros/ la explosion rompio la mesa.The approach I suggest to take to account for the systematic variation observed in the combinatorial behaviour of verbsand the variation in meaning that comes along with it involves an elaboration on the system of types. As I will explore inmore detail in the following section, a quite natural response has become the use of richer type theories inspired in thework of Martin-Lf (1984), where the intuitive and the formal meaning of the word conflate more clearly.

4. Restricted predication as an account of meaning shifts

With the data reported in section 2, together with the criticism put forward for an SEL, I have demonstrated that the widecombinatorial capacity of change of state verbs represents a clear problem that is not straightforwardly addressed by asimple type system. For example, in Montague semantics, the system of types was defined as follows:

(46) a. e is a typeb. t is a typec. for all a, b that are types, ha, bi is a type

This simple type system is not capable of ruling out any anomalies. Even more importantly, however, the simple typesystem has no transparent means to account for the fact that with different types of themes, the verb phrase can imposedifferent kinds of conditions on its subject, as illustrated in the following examples, repeated from above (21) and (22).

Importantly, each predicate places type presuppositions on its arguments, which makes the predication much morerestricted than in the normal Montagovian setting and which allows non-consistent semantic type applications to be ruledout without assuming any meaning postulates. Verbs, for example, are not interpreted over one single domain of entitiesbut rather can be functions over any variety of subdomains. Thus a verb can be represented as a predicate of type A !Prop, A being the domain of objects that the verb can meaningfully apply to (Luo, 2011a). Defining a domain of application

e ir arguments. n to serve as a

11). of the crucial

s ssibilities of romper and cortar, I provide a proposal for a fine-grained type for bothf n of Chatzikyriakidis and Luo (2013).

A.A. Spalek / Lingua 157 (2015) 36--53 494.1. Composing romper verb phrases with physical objects and events

For the causative pattern of romper, corpus evidence shows that romper requires its theme to be either a physicalobject, an atelic eventuality, namely a state or a process, or a norm. For physical objects, romper describes an event inwhich the theme argument, which starts out in a state of being whole, ends up in a state of no longer being whole. Whenromper applies to the class of atelic eventualities, which includes states and processes, it describes a ceasing of theseeventualities.8 Finally, applied to all kinds of norm-denoting nouns, it describes the transgression of a particular norm by ahuman. That is, the meaning of the verb phrase clearly depends on the type of its internal argument to denote one kind ofchange or another. The type of vromperb can thus be represented as a dependent type whose specific interpretations aredetermined by the type of entity that functions as the theme, as represented in the following:

(50) a. ROMPER: PA:entity.B:cause.ROMPER(B,A), which corresponds to a family of functions:b. x: solid-phys-obj.y:cause.ROMPER(y,x)c. x: atelic-eventuality.y:cause.ROMPER(y,x)d. x: norm.y:agent-cause.ROMPER(y,x)

7 Notation A B indicates that A is a subtype of B.8 An anonymous reviewer raises the possibility of having the aspectual restrictions follow from general conceptual considerations. However,

determining whether general conceptual considerations can account for the restriction all by themselves is not trivial, and must be left for ongoingresearch. For now, I consider the identification of aspectual generalisations as an important first step worth reporting.verbs, or wh ch I extend on the notatio

pattern and the argument realisation po

inotion of dependent type. Dependent types are types whose values depend on supplying a value from another type,namely the value of its argument or the type of its argument. The latter case, known as a dependent P-type, is what I usefor my analysis. A dependent P-type corresponds to a family of functions of the form A1!B, A2! B, A3! B, etc., where allAn are subtypes of the same supertype A. This is written PAn: A.An! B. This is the way the dependence of the meaning ofthe verb on the meaning of its argument can be encoded.

To give an example, dependent P-types are used by Chatzikyriakidis and Luo (2013) to interpret subsectiveadjectival modification. vSkilfulb, for instance, must apply to common nouns of type vhumanb or a subtype of it,e.g. vviolinistb vhumanb.7 The effect of the modification on e.g. vviolinistb, however, is not exactly the same as that onvhumanb, and we cannot infer that if one is a skilful violinist, s/he is also a skilful person. To allow for the different effects ofvskilfulb on different types of arguments, it is treated as a family of functions, namely the functions that result fromcombining it with different subtypes of the type of entity. In this way we get distinct propositional functions:

(49) a. vskilfulbb.vhumanb! Propb. vskilfulb.vviolinistb! Prop

The resulting typings in (49) apply to a restricted domain of arguments. A in this case is the domain of human beings thatvskilfulb can meaningfully apply to. In this way selectional restrictions are naturally incorporated through the type system.

Resorting to a more sophisticated type system thus allows us to make use of many subtypes of the type of entities,such as physical objects, humans, eventualities, etc. The differences in theme arguments will be accompanied by adifference in predication. What this means is that romper and cortar can be assigned a type that is A-indexed and whichdepends on the type of their themes respectively.

The use of a sophisticated type theory and the specification of how these items interact in composition provide a way ofconstraining the lexical semantics to guarantee semantically well-formed predications at the same time as they allow toformulate fine-grained semantic restrictions. In the following two sections, I provide a proposal based on the compositionalMTTs further allow the argument of a functor to specify several distinct aspects of its content by the use

mechanism for including more semantic features than the Montagovian system could handle (Asher, 20

Such a sema tics lies within the tradition of Montague Grammar, but it allows for powerful type structures

for a pr dicate represents a way of making explicit the selectional restrictions that predicates impose on the

In (50), the type ROMPER depends on three distinct types, SOLID-PHYS-OBJ, ATELIC-EVENTUALITY and NORM, for which somemore explanation is relevant. For the domain of physical objects specification to solid physical objects, as in the type SOLID-PHYS-OBJ, is needed to account for the fact that liquids are not found in combination with romper, as illustrated in (51).9

(51) #La piedra rompio el agua.the stone broke the waterThe stone broke the water

This fa ility for romper to combine with non-solids, as

atelic eventualities that combine with romper are representedt combine with romper are crucially defined by being gradually

of the intellectuality of the countryintellectuals of the country. [El Pas Corpus]

)

a r t t a c

Pas Corpus]

encia masculina.

A.A. Spalek / Lingua 157 (2015) 36--5350

that applies to liquid, as in La proa rompe el agua The bow breaks the water. In this case,, e of water, rather than the liquid itself. With this in mind, the example under discussion is very

w p solid physical objects. Consequently, I do not consider La proa rompe el agua to be a

counterexample to the generalisation that the physical object must be solid.10 An anonymous reviewer raises the question whether this kind of restriction is not a conceptual limitation, which can be inferred from the coremeaning of the verb, rather than being a lexical restriction. Evidence from cross-linguistic comparison, such as the fact that physical entities suchas fabric undergo the event of breaking in Spanish, as I illustrate in example (7), while they cannot do so in English, seems to justify theassumption that such fine semantic restrictions are part of the lexicon of each language.howeversimilar to agua the metonymscenario ically referhen roms to ther see surfaclects for9 There is a strongly lexicalised se o romper u fMarie Curie broke the idea of a science masculineMarie Curie ruptured the idea of a masculine science. [Web](57) Marie Curie rompio la idea de una ciUganda broke the agreement with Israel. [Elwith romper can be very diverse, as the following examples illustrate.

(56) Uganda rompio el acuerdo con Israel.Uganda broke the agreement with IsraelStates, in turn, are basic lly elated on some proper y present in a ime interv l so that the types of states that ombineThe rain broke he arrival of t e train at the station.

the rain broke the arrival of the train at the station

t h(55) #La lluvia rompio la llegada del tren a la estacion.as illustrated in (55 .

The restriction to atelic eventualities directly explains why achievement-denoting arguments cannot combine with romper,(54) Telefonica rompe la mejora de la Bolsa.Telefonica breaks the improvement of the stock marketTelefonica disrupts the improvement of the stock market. [El Pas Corpus]The Octob r Revolution nterrupted the organic dev lopment of the

the revolution of October broke the development organic

e i eincremental, as is the case with desarrollo development in (53) or mejora improvement in (54).

(53) La revolucion de octubre rompio el desarrollo orgnico de la intelectualidad del pas.Going dby botheeper into processe the exps and stalanattes. ion of the typThe kind of ping inroces(50), theses thaThe wind blow broke the sand.

the stroke of wind broke the sand(52) #El golpe de viento rompio la arena.d in (52).illustrat

ct seems to be due to a broader generalisation, namely the impossibe 10

chematised the

t

TAR(y,x)

e the eventuality type arguments, represented in (61-c) also needs a closer look. I

A.A. Spalek / Lingua 157 (2015) 36--53 51c. x: location.y: agent.CORTAR(y,x)d. x: artefact.y: agent.CORTAR(y,x)e. x: human.y: agent.CORTAR(y,x)f. x: continuous stream.y:agent.CORTAR(y,x)b. x: action-nouns: agent. CORTAR(y,x)take states, processes, time intervals and events to be direct instantiations of eventuality-denoting nouns. Many of thenouns included in the event-type class, however, restrict the subject thematic role for agents. Consequently, a restrictionhas to be introduced for the typing when cortar combines with these kind of themes.

(63) a. CORTAR: PA:event.B:agent.CORTAR(B,A), which correspond to a family of functions:The family of functions that li s behind d. x: phys-obj.y: cause.COR(62) a. CORTAR: PA:phys-obj.B:agent.CORTAR(B,A), which correspond to a family of functions:b. x: food.y: agent.CORTAR(y,x)c. x: liquid.y: internal.cause.CORTAR(y,x)The typing in (61) indicates that CORTAR has a dependent type that basically selects for physical objects andeventualities. The type of physical objects include mainly body parts, objects and food items, while the eventuality typecomprises actions and processes denoting nouns as well as temporal sequences. However, some further explanation isnecessary here to account for the data outlined in section 2.

As direct instantiations of the case of physical objects of cutting, there are many food items as well as body parts thatundergo the event of physical separation. As mentioned above, a food cutting event generally subcategorises for agents,which will require a special typing restriction. This restriction, however, does not apply to liquids and body parts as well asother physical entities, and thus the typing restriction in (61-b), which further elaborates on (62), can be maintained inthese cases.c. x: eventuality.y:cause.CORTAR(y,x)

b. x: phys-obj.y:cause.CORTAR(y,x)(61) a. CORTAR: PA:entity.B:cause.CORTAR(B,A), which corresponds to a family of functions:(60) vROMPERb NORM ! AGENT: Proposition about the state that holds between an agent and a norm.

The typing restrictions presented here illustrate how an articulated type system allows us to express that a transitive verbpasses its type presuppositions to its object, while a verb phrase passes its type presuppositions to its subject.

4.2. Composing cortar verb phrases with physical objects and events

As in the case of romper, the combinatorial diversity of cortar can be captured by positing two major groups of typicallyselected arguments, namely physical objects and eventualities. The dependent type that I propose for cortar is thefollowing:Seman ically the proposition triggered also conveys a distinct overall meaning:The wind broke the lawIt should be underscored further that, given the restriction on the thematic role of the cause, there is a different typingrestriction for the case when romper combines with norm-denoting nouns, as explicitly stated in (50-d). This typingrestriction renders predications such as (59) undefined.

(59) #El viento rompio la ley.the wind broke the law(58) vROMPERb EVENTUALITY ! CAUSE: Proposition about ceasing of a eventuality.following way:

The propositions triggered by romper verb phrases in combination with eventualities thus can be s

Science

References

A.A. Spalek / Lingua 157 (2015) 36--5352Achard, M., 2006. Breaking verbs in French: two intransitive constructions. In: Talk on the 8th Conceptual Structure in Discourse and LanguageConference.

Alonso Ramos, M., 2011. Sobre los usos figurados - extensiones de una nica definicion? In: Escandell Vidal, M.V., Leonetti Jungl, M., SnchezLopez, C. (Eds.), 60 Problemas de Gramtica Dedicados a Ignacio Bosque. AKAL, Tres Cantos, pp. 340--346.

Asher, N., 2011. Lexical Meaning in Context. Cambridge University Press, Cambridge.s, as well as by an ICREA Acadmia award to Louise McNally.Turning to the propositions that cortar triggers, similarly to romper, the propositions vary in each case depending on thetype of the argument taken. When cortar applies to a physical object it describes an action of incision into an object or theaction of losing homogeneity. When, in turn, it combines with an eventuality it denotes the interruption of an ongoing actionor process. We thus can get two distinct propositions with cortar:

(64) a. vCORTARb PHYSICAL OBJECT ! AGENT/ INSTRUMENT: Proposition about the target state of aphysical object.

b. vCORTARb EVENTUALITY ! CAUSE: Proposition about the interruption of an ongoing eventuality.

In both cases the dependent type represents a means for incorporating a richer notion of lexical semantics withincompositional semantics.

5. Conclusions

Throughout this paper I have explored the meaning shifts of two verbs, romper break and cortar cut. To guide theanalysis, I have addressed two aspects that are strictly related to these meaning shifts: their argument selection andrealisation possibilities.

Concerning argument selection, I have illustrated that the abstract entities selected as themes of romper and cortareither directly refer to eventualities or else are coerced into eventualities. Roughly, it can be said that romper within thegroup of physical objects selects for all kinds of physical entities except liquids, while the set of eventualities it selects for isstates, norms and processes. Cortar, in turn, combines in the physical domain with body parts, food items andsubstances. In the eventuality domain it selects for actions, processes and temporal sequences.

Among the findings several observations about the argument alternation facts have been made. As I have illustrated,argument alternation possibilities do not depend exclusively on the semantics of the verb, but rather are sensitive tothe semantics of the theme argument. In this respect I have shown that romper usually appears in the causative-anticausative alternation, although this is not always the case. The exception here lies in combinations with argumentsthat denote norms. In these cases romper verb phrases refer to situations of human-controlled interaction and thereforerestrict the choice of subject argument to humans and human-related action. Cortar, in turn, was shown to appear in thecausative-anticausative alternation, a fact that has not received much attention in the previous literature. This is regularlythe case for Spanish cortar when it appears with themes that denote processes, and supplied substances, such as gas orwater, and food-denoting substances, such as mayonnaise. Cortar does not appear in the anticausative for events andentities that are coerced into events such as locations.

All these facts about argument selection possibilities and how they influence argument realisation possibilities clearlyillustrate that the semantics of a verb phrase heavily depends on the semantics of the theme, which calls for a mechanismthat allows the theme to influence the meaning of the verb phrase in various respects. To account for these facts I haveelaborated on the types for romper and cortar, following recent suggestions in work on Modern Type Theories. Theseenriched type systems provide a way of specifying a restricted domain of application that guides predication.

Turning to the semantic content of the verbs this study has illustrated that both verbs romper and cortar can beassociated with a highly underspecified meaning, which I have related to the concept of disconnection, and both verbsdescribe change both in the domain of physical objects as well as in the domain of abstract events. Based on a corpusstudy I have postulated what kind of themes these underspecified verbs can be combined with and what kind of verbphrase meaning emerges in each case.

Acknowledgements

This research has been supported by a predoctoral FPU grant from the Spanish Ministry of Education, Culture andSports (MECD) included in the FFI2012-37654 research project and the UPF Department of Translation and Language

Asher, N., Pustejovsky, J., 2013. A type composition logic for generative lexicon. In: Pustejovsky, J., Bouillon, P., Isahara, H., Kanzaki, K., Lee, C.(Eds.), Advances in Generative Lexicon Theory. Vol. 46 of Text, Speech and Language Technology. Springer, Netherlands, pp. 39--66. http://dx.doi.org/10.1007/978-94-007-5189-7_3

Bohnemeyer, J., 2007. Morpholexical transparency and the argument structure of verbs of cutting and breaking. Cogn. Linguist. 18, 153--177.Bosque, I., 2004. REDES Diccionario combinatorio del Espan ol Contemporneo. Ediciones SM, Madrid.hatzikyriakidis, S., Luo, Z., 2013. Adjectives in a modern type-theoretical setting. Lect. Notes Comput. Sci. 8036, 159--174.Chierchia, G., 2004. A semantics of unaccusatives and its syntactic consequences. In: Alexiadou, A., Anagnostopoulou, E., Everaert, M. (Eds.),

The Unaccusative Puzzle. Oxford University Press, Oxford, pp. 22--59.Dowty, D., 1979. Word Meaning and Montague Grammar: The Semantics of Verbs and Times in Generative Semantics and in Montagues PTQ.

Reidel, Dodrecht.Fillmore, C., 1967. The grammar of hitting and breaking. In: Roderick, J., Rosenbaum, P. (Eds.), Readings in English Transformational Grammar.

A.A. Spalek / Lingua 157 (2015) 36--53 53Waltham, MA, Ginn, pp. 120--133.Fodor, J., Lepore, E., 1998. The emptiness of the lexicon: Critical reflections on J. Pustejovskys Generative Lexicon. Linguist. Inq. 29, 269--288.Guerssel, M., Hale, K., Laughren, M., Levin, B., White Eagle, J., 1985. A cross-linguistic study of transitivity alternations. In: Papers from the

Parasession on Causatives and Agentivity. Chicago Linguistic Society, pp. 48--63.Hale, K., Keyser, S.J., 1986. A view from the middle. In: Lexicon Project Working Papers 10.Hanks, P., Jezek, E., 2008. Shimmering lexical sets. In: Bernal, E., DeCesaris, J. (Eds.), Proceedings of the XIII EURALEX International

Conference. Institut Universitari de Lingustica Aplicada. Universitat Pompeu Fabra, Barcelona, pp. 391--402.Haspelmath, M., 1993. More on the typology of inchoative-causative verb alternations. In: Comrie, B., Polinsky, M. (Eds.), Causatives and

Transitivity. John Benjamins, Amsterdam, pp. 87--120.Jespersen, O., 1927. A Modern English Grammar on Historical Principles. Part 3. Heidelberg, Carl Winter.Lakoff, G., Johnson, M., 1980. Metaphors We Live By. University of Chicago Press, Chicago.Levin, B., 1993. English Verb Classes and Alternations: A Preliminary Investigation. University of Chicago Press, Chicago.Levin, B., Rappaport Hovav, M., 1995. Unaccusativity at the Syntax-Lexical Semantics Interface. MIT Press, Cambridge, MA.Luo, Z., 2011a. Contextual analysis of word meanings in type-theoretical semantics. In: Pogodalla, S., Prost, J.-P. (Eds.), Logical Aspects of

Computational Linguistics, Vol. 6736 of Lecture Notes in Computer Science. Springer, pp. 159--174. http://dx.doi.org/10.1007/978-3-642-22221-4_11

Luo, Z., 2011b. Type-theoretical semantics with coercive subtyping. In: Li, N., Lutz, D. (Eds.), Proceedings of Semantics and Linguistic Theory(SALT) 20. eLanguage, pp. 38--56.

Majid, A., Boster, J.B., Bowerman, M., 2008. The cross-linguistic categorization of everyday events. A study of cutting and breaking. Cognition109, 235--250.

Marantz, A., 1984. On the Nature of Grammatical Relations. MIT Press, Cambridge, MA.Martin-Lf, P., 1984. Intuitionisitic Type Theory, vol. 17. Bibliopolis, Naples.Mendikoetxea, A., 1999. Construcciones inacusativas y pasivas. In: Bosque, I., Demonte, V. (Eds.), Gramtica Descriptiva de la Lengua

Espan ola. Espasa Calpe, Madrid, pp. 1575--1629.Nunberg, G., 1995. Transfer of meaning. J. Semant. 27, 109--132.Pin on, C., 2001. A finer look at the causative-inchoative alternation. In: Ashton, N., et al. (Eds.), Proceedings of Semantics and Linguistic Theory

11. CLC Publications, Ithaca, NY, pp. 346--364.Pustejovsky, J., 1995. The Generative Lexicon. MIT Press, Cambridge, MA.Ranta, A., 1994. Type Theoretical Grammar. Oxford University Press, Oxford.Rappaport Hovav, M., Levin, B., 2005. Change of state verbs: Implications for theories of argument projection. In: Erteschik-Shir, N., Rapoport, T.

(Eds.), The Syntax of Aspect. Oxford University Press, Oxford, pp. 274--286.Rappaport Hovav, M., Levin, B., 2010. Reflections on manner/ result complementarity. In: Rappaport Hovav, M., Doron, E., Sichel, I. (Eds.),

Lexical Semantics, Syntax, and Event Structure. Oxford University Press, Oxford, pp. 21--38.Recanati, F., 2005. Literalism and contextualism: some varieties. In: Preyer, G., Peter, G. (Eds.), Contextualism in Philosophy. MIT Press,

Cambridge, MA, pp. 171--196.Reinhart, T., 1996. Syntactic Effects of Lexical Operations: Reflexives and Unaccusatives. Utrecht Institute of Linguistics OTS working papers in

Linguistics, Utrecht.Searle, J., 1980. The background of meaning. In: Searle, J., Kiefer, F., Bierwisch, M. (Eds.), Speech Act Theory and Pragmatics. D. Reidel

Publishing, Dordrecht, pp. 221--233.Viberg, ., 2007. Whole and broken. Breaking and cutting in Swedish from a crosslinguistic perspective. In: Ahlsn, A., et al. (Eds.),

Communication -- Action -- Meaning. A Festschrift to Jens Allwood. Department of Linguistics, Gteborg University, pp. 17--42.

Spanish change of state verbs in composition with atypical theme arguments: Clarifying the meaning shifts1 Introduction2 Empirical generalisations3 Possible approaches to meaning shifts4 Restricted predication as an account of meaning shifts4.1 Composing romper verb phrases with physical objects and events4.2 Composing cortar verb phrases with physical objects and events

5 ConclusionsAcknowledgementsReferences