The fuzzy boundaries of operator verb and support verb constructions

Loading...

The fuzzy boundaries of operator verb and support verb constructions with dar “give” and ter “have” in Brazilian Portuguese Amanda Rassi1,2,3 , Cristina Santos-Turati1,2,3 , Jorge Baptista2,3 , Nuno Mamede3 , Oto Vale1 {aprassi,mcturati,jbaptis}@ualg.pt, [email protected], [email protected] 1 UFSCar, Rodovia Washington Lu´ıs, km 235 - SP-310 S˜ao Carlos-SP, Brazil 2 UAlg/CECL, Campus de Gambelas, 8005-139 Faro, Portugal 3 INESC-ID Lisboa/L2F, Rua Alves Redol, n.o 9, 1000-029 Lisboa, Portugal Abstract This paper describes the fuzzy boundaries between support verb constructions (SVC) with ter “have” and dar “give” and causative operator verb (VopC) constructions involving these same verbs, in Brazilian Portuguese (BP), which form a complex set of relations: (i) both verbs are the support verb of the same noun (SVC); (ii) dar is the standard (active-like) SVC while ter is a converse (passive-like) SVC; and (iii) dar is a VopC, operating on a ter SVC. In this paper we have systematically studied these complex relations involving SVC and VopC for BP, which constitute a challenge to Natural Language Processing (NLP) systems, and have been often ignored in related work. The paper proposes a lexically-based strategy to implement SVC in a fully-fledged, rule-based parsing system, yielding an adequate semantic structure of the events (predicates) denoted by predicative nouns in SVC.

1

Introduction: basic concepts and a little history

The notion of support verb has been in use for a long time, under many different theoretical perspectives and various terminologies. In this paper, we adopt the Zellig S. Harris’s (1991) transformational operator grammar framework. As early as 1964, Harris (1964, p.216-7) proposed the concept and named this particular type of construction as “U operator” nominalizations, linking sentences such as He studies eclipes = He makes studies of eclipses. It was, however, M. Gross (1981) who first provided the definition of support verb we will rely upon here. The support verb make (in the example above) can be seen as a sort of an auxiliary of the predicative noun studies, in charge of carrying the grammatical values of tense and person-number agreement that the noun is morphologically unable to express. In many cases, support verbs are practically devoid of meaning. For lack of space, we cannot detail further the properties of SVC, and only the briefest outline is provided here; a good overview can be found in (Gross, 1996; Gross, 1998; Lamiroy, 1998). One of the most important theoretical contribution of the notion of support verb came from the fact that it provides a natural framework to adequately include in the kernel sentences of the language the large number of ‘abstract’ nouns, which do not have neither a verbal nor an adjectival counterpart; that is, they are isolated or autonomous nouns, lacking any nominalizations (in a synchronic perspective, at least). This phenomenon is particularly evident in Romance languages (French, Italian, Portuguese, Romanian and Spanish): FR: Jean a fait gr`eve “Jean did strike”; IT: Giovanni ha fatto sciopero “id.”; PT: O Jo˜ao fez greve “id.”; RU: Ioan a f˘acut grev˘a “id.”; SP: Joan hizo huelga “id.”; cp. EN: *John did strike, John was on strike). Finally, nominal constructions are unlike any other predicative part-of-speech by the fact that predicative nouns can present more than one construction with different support verbs, while still expressing the same semantic predicate. Hence, for example, greve “strike” can have a SVC with both fazer “to make” (as above) and estar em “to be in”: O Jo˜ao est´a em greve “Jo˜ao is on strike” (Ranchhod, 1990). Each SVC has its own specific properties, e.g. only SVC with fazer can undergo passive, while the general predicate remains the same. This work is licenced under a Creative Commons Attribution 4.0 International License. Page numbers and proceedings footer are added by the organizers. License details: http://creativecommons.org/licenses/by/4.0/

92 Proceedings of the Workshop on Lexical and Grammatical Resources for Language Processing, pages 92–101, Coling 2014, Dublin, Ireland, August 24 2014.

In this paper, we also consider the concept of operator verb (VopC), introduced in the same paper (Gross, 1981, p. 23-39); two relatively clear situations were distinguished: • a causative operator verb (VopC), which adds a new element to an elementary sentence; this element has an easily identifiable meaning: CAUSE; distributionally, this element suffers very loose constraints (and we define this as a distributionally non constraint position (Nnr)); if the base sentence under the operator is a support verb construction 1, the VopC may “absorb” the support verb and it may also introduce some formal changes in that sentence 1; (1) Isso d´a # Max tem (fome + medo + sede).

1

“This gives # Max has (hungry + fear + thirst).”

(2) Isso d´a (fome + medo + sede) em Max. “This gives Max (hungry + fear + thirst).” In (2), the support verb ter is absorbed under the operator dar and its subject becomes a dative, indirect complement, though the semantic roles of subject of dar (CAUSE) and of the subject of the predicative noun (EXPERIENCER), after this restructuring, remain the same. • a linking operator-verb (VopL), which hardly modifies the meaning of the underlying sentence; it also adds an argument to the base sentence 1, but this is not a new one since it is bounded linked to a noun complement of the base sentence 1 (Ranchhod, 1990). (3) Max tem # Ana est´a sob o controle do Max. “Max has # Ana is under Max’s control.” (4) = Maxi tem Ana sob o (seui + *meu + *teu) controle. “Maxi has Ana under (hisi + *my + *your) control.” This paper reports an ongoing research to systematically classify the predicative nouns built with the support verbs dar and ter in Brazilian Portuguese (Rassi and Vale, 2013; Santos-Turati, 2012). Similar work has already been developed for the European variety (Vaza, 1988; Ranchhod, 1990; Baptista, 1997; Baptista, 2005). For many languages, including Portuguese, the studies on support verb constructions and causative constructions use a lexical approach, aiming at building dictionaries or lists of predicative nouns or at identifying those constructions (semi)automatically, e.g. for Portuguese (Hendrickx et al., 2010; Duran et al., 2011), for English (Grefenstette and Teufel, 1995), for German (Hanks et al., 2006; Storrer, 2007) and many other languages. As far as we could ascertain, no implementation of these SVC constructions has been made yet in NLP systems, particularly in parsers. Most systems considering these constructions just treat them as multiword expressions, ignoring their internal syntactic structure. In this paper, we will show the complex set of relations involved in these SVC, where these verbs can function not only as support but also as operator verbs, thus rendering their description remarkably complex, particularly in view of Natural Language Processing. We aim at capturing the syntactic dependencies involved in these expressions, not as multiword, fixed strings, but as analyzable syntactic structures. The paper is structured as follows: Next, Section 2 presents the current state of the collection and classification if these SVC in Brazilian Portuguese; Section 3 illustrates the syntactic-semantic relations between different constructions of ter and dar; Section 4 proposes a strategy for implementing the data in a rule-based parsing system; and, finally, Section 5 presents some concluding remarks and perspectives on future work.

2

Support verb constructions with ter “have” and dar “give”

The predicative nouns in this paper select the support verbs dar “give” and ter “have”, and were retrieved from previous lists of predicative nouns in European Portuguese (Vaza, 1988; Baptista, 1997) and from the PLN.BR Full corpus (Bruckschein et al., 2008). This corpus contains 103,080 texts, with 29 million tokens, consisting of news pieces from Folha de S˜ao Paulo, a Brazilian newspaper (from 1994 to 2005). All these constructions were validated in real data, and in some cases also ressourcing to the web. 1 In the examples, elements between brackets and separated by the ‘+’ sign can all appear in that given syntactic slot. The symbol ‘#’ delimits clauses, while the ‘*’ mark signals the sentence as unacceptable. Correferent elements are linked by correference indexes i . For clarity, all support verbs will be shown without italics in the examples. An approximate translation of Portuguese examples is provided, but its acceptability is irrelevant for the paper.

93

2.1

Nominal predicates with support verb ter “have”

We adopted several criteria that allowed us to constitute lexical-syntactic, relatively homogeneous, classes. These criteria were inspired in those taken from previous classifications, developed in the Lexicon-grammar framework of Maurice Gross (1975; 1988; 1996), for both Portuguese and other languages. The main classification criteria can be summarized as follows: (i) the number of arguments, considering constructions with a subject and one or two essential complements as arguments; (ii) the possibility of a noun admitting a sentential construction (in subject or complement position); (iii) the distributional nature of the arguments: if they are obligatorily human or allow for non-human nouns; (iv) the property of symmetry 2 between the arguments. Following these criteria, we have so far classified around 1,000 nominal constructions from a list with 3,000 candidates of predicative nouns censed in the corpus (Santos-Turati, 2012). The already classified nominal predicates that select the support verb ter “have” in Brazilian Portuguese were divided into 9 classes (Table 1) 3 . Class TH1

Structure Nhum0 ter Npred

TNH1

N-hum0 ter Npred

TR1

N±hum0 ter Npred

TH2

Nhum0 ter Npred Prep Nhum1

TNH2

N-hum0 ter Npred Prep Nhum1

TR2

N±hum0 ter Npred Prep N-hum1

TS2 TQF1

Nhum0 ter Npred Prep Nhum1 (Simetry) QueF0 ter Npred Prep N1

TQF2

N0 ter Npred Prep QueF1

TOTAL

Example/Gloss Ana tem uma beleza impressionante “Ana has an amazing beauty” A tinta tem um tom escuro “The paint has a dark tone” (Ana + a m´usica) tem um ritmo contagiante “(Ana + the music) has a contagious rhythm” Ana tem respeito por Max “Ana has respect for Max” O bombom tem gosto de avel˜a “The bonbon has taste like hazelnut” (O carro + a cidade) tem um alto consumo de a´ gua “(The car + the city) has a high consumption of water” O patr˜ao tem um acordo com o empregado “The boss has an agreement with the employee” Esse fato tem uma grande importˆancia para Ana “This fact has a great importance for Ana” Ana tem medo de dirigir na estrada “Ana has fear to drive on the road”

Count 465 138 139 111 6 22 38 6 80 1,005

Table 1: SVC with support verb ter (Santos-Turati, 2012) 2.2

Nominal predicates with support verb dar “give”

The same criteria were also adopted for SVC with verb dar “give” (Rassi and Vale, 2013), though two differences were considered: (i) the constructions with a body-part noun (Npc) as argument were distinguished as a special class for their particular properties; and (ii), no symmetric constructions were found. We classified 900 support verb constructions with verb dar “give” in Brazilian Portuguese into 11 classes (Table 2).

3

Relations between ter “have” and dar “give”

First of all, it is necessary to distinguish three different kinds of relations established between verb dar and verb ter constructions. The first type of relation considers the verbs dar “give” and ter “have” as synonymous and classified as standard support verb constructions. The verb dar can replace the verb ter without any changes in the meaning of the sentence or in the selection restrictions of the arguments: 2 The notion of symmetry in verbal constructions was initially presented by Borillo (1971) for French verbs - Paul rencontre son fr`ere “Paul meets his brother” / Paul et son fr`ere se rencontrent “Paul and his brother meet”. In the case of the Portuguese nominal constructions, symmetry was presented in Ranchhod (1990) and Baptista (2005), who described the nominal predicates with the support verbs estar com and ser de, respectively. 3 In Table 1 and Table 2, the left column shows the conventional codes for designating each class; and the second column represents its syntactic structure, indicated as follows: Nhum and N-hum for human and non-human noun respectively; N±hum for both human or non-human noun; Npc for body-part noun; the indexes ‘0 ’ and ‘1 ’ indicate the subject and the complement position, respectively; Npred stand for the predicative noun; Prep for preposition; QueF for completive.

94

Class DH1

Structure Nhum0 dar Npred

DNH1

N-hum0 dar Npred

DR1

N± hum0 dar Npred

DH2

Nhum0 dar Npred Prep Nhum1

DNH2

Nhum dar Npred Prep N-hum1

DPC2

Nhum0 dar Npred Prep Npc1

DQF2

Nhum0 dar Npred Prep QueF1

DHR2

Nhum0 dar Npred Prep N± hum1

DRH2

N± hum0 dar Npred Prep Nhum1

DR2

N± hum0 dar Npred Prep N-hum1

D3

N0 dar Npred Prep N1 Prep N2

TOTAL

Example/Gloss Ana deu uma pirueta “Ana gave a pirouette” O bal˜ao deu um estouro “The baloon gave a burst” (Max + O clima) deu uma refrescada “(Max +The weather) gave a refreshed” Max deu um castigo para a Ana “Max gave a punishment to Ana” Ana deu uma cozida nos legumes “Ana gave a cooked in the vegetables” Max deu um tapa na cara da Ana “Max gave a slap in Ana‘s face” Max deu um jeito de consertar o carro “Max gave a way to fix the car” Ana deu destaque ao (Max + problema) “Ana gave emphasis to (Max + the problem)” (Ana + O telhado) deu protec¸a˜ o ao Max “(Ana + The roof) gave protection to Max” (Ana+A lei) deu embasamento a` teoria “(Ana+The law) gave basis to the theory” Ana deu um apelido de macaco ao Max “Ana gave the nickname monkey to Max”

Count 133 20 51 217 137 114 52 60 32 25 59 900

Table 2: SVC with support verb dar (Rassi and Vale, 2013) (5) Ana (deu + teve) um + um(a) (birra + chilique + pirepaque + tremelique + troc¸o). “Ana (gave + had) (a + an) (tantrum + hissy fit + outburst + shiver + thing).” The second type of relation concerns the transformation named Conversion by G. Gross (1982; 1989), in which the predicative noun is maintained and their arguments change their relative position, without, however, changing their semantic roles. In these constructions, the sentence with AGENT subject is called the standard construction, while its equivalent sentence with the reversed argument order is called the converse construction. Usually, the support verbs of the standard and the converse construction are different, as it is also the preposition introducing the converse complement: (6) Ana deu algum apoio ao Max. “Ana gave some support to Max.” [Conv.] = Max teve algum apoio da Ana. “Max had some support from Ana.” The third kind of relation linking the sentences with the verb ter and the verb dar is the causative operator construction (already mentioned in §1): (7) Isso deu # Ana tem coragem. “This gave # Ana has courage.” = Isso deu coragem a` Ana. “This gave courage to Ana.” These three types of relations are presented in the table below, with an example and the respective number of constructions in each type. From the intersection between the list of predicative nouns constructed with verb ter “have” and those with verb dar “give”, we found 693 predicative nouns, distributed as shown in Table 3. dar “give” SVCstandard

ter “have” SVCstandard

SVCstandard

SVCconverse

VopCausative (VopC)

SVCstandard

Example/Gloss Ana deu um chilique “Ana gave a hissy fit” Ana teve um chilique “Ana had a hissy fit” O policial deu uma multa ao Max “The officer gave Max a fine” Max teve uma multa “Max had a fine” A flor deu alergia a Ana “The flower gave allergy to Ana” Ana tem alergia a` flor “Ana has an allergy”

Table 3: Comparative table with syntactic relations

95

Count 35 72 586

3.1

Verbs dar and ter in standard SVC

Around 4.8% of the predicative nouns (35 constructions) accept both support verbs dar and ter in standard constructions, such as: (8) A empresa (d´a + tem) atenc¸a˜ o ao cliente. “The company (gives + has) attention to the client.” (9) O rem´edio (d´a + tem) um efeito positivo no organismo. “The medicine (gives + has) a positive effect on human body.” (10) O resultado (deu + teve) um impacto significativo para o time. “The result (gave + had) a significant impact to the team.” In Brazilian Portuguese, around 35 predicative nouns, such as febre “fever” and dengue “dengue”, besides having both dar and ter as their support verb also allow dar as a causative operator on them (examples taken from the web): [VopC]: [Sua lic¸a˜ o de casa:] a´ gua parada d´a dengue. “[...] still water gives (= causes) dengue.” [CVS dar]: Inclusive, a vizinha tamb´em deu dengue. “Inclusive, the neighbour gave (= had) dengue.” [CVS ter]: O meu esposo j´a teve dengue. “My husband already had dengue.” A few nouns (around 10), such as amor “love”, confianc¸a “trust” and respeito “respect”, besides admitting the two support verbs in their basic construction, also admit ter in a converse construction: (11) O filho d´a respeito a` m˜ae. “The son gives respect to the mother.” = O filho tem respeito pela m˜ae. “The son has respect for the mother.” [conv.] = A m˜ae tem o respeito do filho. “The mother has respect from her son.” 3.2

Verb dar as standard SVC and ter as converse SVC

Around 10.4% of the predicative nouns (72 constructions) admit the verb dar in the standard construction and the verb ter in a converse construction, but not ter as a standard support. In Brazilian Portuguese, predicative nouns constructed with the support verb dar in a standard construction accept other converse verbs beyond the verb ter “have”, such as receber “receive”, ganhar “gain”, levar “get” and tomar “take”4 . (12) Ana deu protec¸a˜ o ao Max. “Ana gave protection to Max.” = Max (teve + recebeu) a protec¸a˜ o da Ana. “Max (had + received) the protection from Ana.” (13) Ana deu uma ajuda ao Max. “Ana gave a help to Max.” = Max (teve + ganhou) uma ajuda da Ana. “Max (had + gained) a help from Ana.” (14) Ana deu uma resposta no Max. “Ana gave an answer to Max.” = Max (teve + levou) uma resposta da Ana. “Max (had + got) an answer from Ana.” (15) O policial deu uma multa ao Max. “The officer gave a fine to Max.” = Max (teve + tomou) uma multa do policial. “Max (had + took) a fine from the officer.” 3.3

Verb dar as VopC and ter as SVC

Around 84.8% (586 predicative nouns) of the elementary constructions with the support verb ter “have” also allow the causative operator verb dar “give”; some of these nouns constitute relatively homogenous semantic sets, e.g. the predicative nouns that express , , or those that indicate (this semantic classification is just approximative): 4 For European Portuguese equivalent converse constructions, see Baptista (1997); for a comparison between the two language variants, see Rassi et al. (2014).

96

(16) Ana tem alegria. “Ana has happiness.” (Z´e + A vinda do Z´e + O fato de Z´e ter voltado + Isso) deu alegria a` Ana. “(Z´e + Z´e’s coming + The fact of Z´e has came + That) gave happiness to Ana.” (17) Ana tem c´olica. “Ana has colic.” (O chocolate + O fato de ter comido chocolate + Isso) deu c´olica na Ana. “(The chocolat + The fact of she has eaten chocolat + That) gave a colic in Ana.” These predicative nouns allow a particular (impersonal?) construction with dar, in which the argument in subject position is not explicit, so the CAUSE element is also absent, and the sentence has the same overall meaning of the SVC with verb ter standard, but with an inchoative aspect; notice that the verb dar must be in the 3rd person singular, and it does not agree with the predicative noun: (18) (Deu + *Deram) (uma) (alegria + c´olica) na Ana. “Gives/gave (a) (hapiness + colic) in Ana.” = Ana teve (uma) (alegria + c´olica). “Ana had (a) (hapiness + colic).” (19) (Deu + ?*Deram) umas (palpitac¸o˜ es + c´olicas) na Ana. “Gives/gave some (palpitations + colics) in Ana.” = Ana teve umas (palpitac¸o˜ es + c´olicas). “Ana had some (palpitations + colics).” 3.4

Formalization into the Lexicon-Grammar

Because of the complex relations and the different syntactic status that the verbs dar and ter may show, these constructions are essentially determined by the lexicon, i.e., they depend on the specific predicative noun. It is only natural that a lexically-based approach be taken in order to describe this properties, particularly in view of the implementation of such type of expressions in NLP systems. The Lexicon-Grammar framework constitutes such a methodological setting, as it presupposes the extensive and systematical survey and formal representation of the lexicon properties. In the Lexicon-Grammar, a systematic description of linguistic phenomena is usually presented in the form of binary matrices: the lines contain the lexical entries while the columns represent syntacticsemantic properties of each entry. For example, for each predicative noun, distributional constraints on the arguments are represented; the elementary support verb and the main variants of this verb are encoded; the possibility of accepting conversion and the converse support verbs are explicitly provided; and all these syntactic-semantic informations are specified for each predicative noun. Besides its intrinsic linguistic interest, the main purpose for this formalization requirements is the application of the data in NLP. In the next section, we present a preliminary proposal for the implementation problems of these type of SVC in a rule-based parsing system of Portuguese.

4

Towards the implementation of SVC in a NLP system

Besides its linguistic interest, one of the goals of the formal representation of the lexical properties of predicative nouns and SVC into a Lexicon-Grammar such as described above (§3.4) is to allow for the implementation of these data in NLP systems. In this section an outline of the strategy adopted for its implementation specifically into a rule-based system, namely STRING (Mamede et al., 2012) 5 , is presented. This is still an on-going work, so in the next lines we briefly sketch the system’s architecture (§4.1.) and then (§4.2.) we present the strategy that we intend to implement for the adequate parsing of SVC with ter and dar, having in mind the complex structures and relations mentioned in §3. 4.1

STRING architecture

STRING is an NLP chain with a modular structure that executes all the basic processing tasks, namely: tokenization and text segmentation, part-of-speeh tagging, morphosyntactic disambiguation, shallow parsing (chuking) and deep parsing (dependency extraction). The parsing stage is performed by the rule-based parser XIP (Xerox Incremental Parser) (Mokhtar et al., 2002). XIP identifies the elementary 5

http://string.l2f.inesc-id.pt/

97

constituents of a sentence, such as noun phrases (NP) or prepositional phrases (PP), and then these are structured by binary dependencies between them, corresponding to the syntactic relations, such as subject (SUBJ), direct complement (CDIR) or modifier (MOD). STRING also extracts Named Entities, performs time expressions identification and normalization, Anaphora Resolution and some Word-Sense disambiguation (WSD). At the final stages of parsing, the system extracts the text events (or predicates) and their participants (arguments). The system currently extracts the EVENT structure for all full verbs and predicative nouns. In the case of verbs, it associates the events to their participants and circumstances, identifying their corresponding semantic roles (Talhadas, 2014), based on the sentence parse and the information available on the Portuguese full verbs Lexicon-Grammar (Baptista, 2012) 6 . Hence, for a sentence such as (20), the system parser extracts the event structure by way of the following dependencies: (20) Max costuma ler o jornal no caf´e a` s sextas-feiras. “Max uses to read the newspaper at the caff´ee on Fridays.” EVENT AGENT(ler,Max) EVENT OBJECT(ler,jornal) EVENT LOC-PLACE(ler,caf´ e) EVENT TIME-FREQUENCY(ler,a as sextas-feiras)

4.2

Strategy

In the case of a predicative noun in a SVC, one would want the predicative noun also to be captured as an EVENT, but not the support verb, since its role is basically that of an auxiliary of the noun. However, since the support verb conveys several important grammatical information, particularly the standard/converse orientation of the predicate7 , a SUPPORT dependency is first extracted, so in sentences such as in (21) one would get the dependency shown below: (21) Max deu um beijo na Ana. “Max gave a kiss in Ana.” SUPPORT STANDARD(beijo,deu)

To do so, one needs to provide the system with the information that dar is the (basic) standard support verb of the predicative noun beijo “kiss”. It is also necessary to know that in this construction, the predicative noun is the direct complement (CDIR) of the support verb and that the dative complement can be introduced, in Brazilian Portuguese, by preposition em “in/on”. The following rules illustrate (in a simplified way8 ) the functioning of the rule-based system: if (CDIR(#1[lemma:dar],#2[lemma:beijo]) & ˜SUPPORT(#2,#?)) SUPPORT[vsup-standard](#1,#2) if (SUPPORT(#1,?)) EVENT[OTHER=+](#1). if (SUPPORT[vsup-standard](#1[lemma:beijo],#2) & EVENT[other](#1) & SUBJ(#2,#3)) EVENT[agent-generic=+](#1,#3). if (SUPPORT[vsup-standard](#1[lemma:beijo],#2) & EVENT[other](#1) & ˆMOD(#2,#3) & PREPD(#3,?[lemma:em]) ) COMPL(#1,#3), EVENT[patient=+](#1,#3). 6

This semantic role information is still not available for the predicative nouns, but it is currently being encoded. The support verb can convey aspectual, modal and even stylistic values, which are encoded in the lexicon and remain available in the system’s output, even if not necessarily visible in the EVENT representation. 8 The rule system should also take into account the distributional constraints on the argument slots, but, for simplicity, we dismissed it in this paper. 7

98

if (SUPPORT[vsup-standard](#1[lemma:beijo],#2) & EVENT[other](#1) & (ˆMOD[dat](#2,#3) || ˆCLITIC(#2,#3[dat]) ) ) CINDIR[dat=˜](#1,#3), EVENT[patient=+](#1,#3).

The rules read as follows: First, a SUPPORT dependency with the feature VSUP-STANDARD is extracted when the noun beijo “kiss” is the direct complement of the verb dar “give” (and no other support verb was extracted yet for that noun); based on this dependency, an EVENT (unary) dependency is extracted for the predicative noun; then, the subject of the standard support verb is assigned the agent semantic role (agent-generic in STRING’s terminology); next, the prepositional phrase modifying (MOD) the support verb and introduced by preposition em “in” is converted into a complement (COMPL) of the predicative noun and assigned a semantic role of patient; a similar procedure is used for the dative complement, when reduced to a dative pronominal form, but in this case, instead of COMPL the CINDIR (indirect complement) dependency is used. All these rules are automatically produced for each predicative noun, from the information in the Lexicon-Grammar. The corresponding EVENT structure is represented below: SUPPORT VSUP-STANDARD(beijo,deu) EVENT AGENT(beijo,Max) EVENT PATIENT(beijo,Ana)

For the converse construction 4.2, while the EVENT structure remains the same, the SUPPORT dependency is: (22) Ana ganhou um beijo do Max. “Ana got a kiss from Max.” SUPPORT VSUP-CONVERSE(beijo,ganhou)

The converse construction entails the “swapping” of the arguments’ syntactic function, while keeping their respective semantic roles. The detection of the converse construction triggers a set of rules that also swap the semantic roles associated to the predicative noun’s syntactic slots. In the case where the same verb is both the standard and the converse support of a predicative noun, they are both extracted, at first, and then the presence of prepositional complements or the determiner of the noun can be used for disambiguation. This will be part of future work as, for the moment, whenever this happens, the converse construction is discarded. The assigning of semantic roles to the predicative noun’s arguments is then made only once, and by general rules, both in the standard and in the converse constructions. The situation is somewhat similar in the case of a causative-operator verb 4.2: (23) Essa not´ıcia deu estresse no Max. “This news gave stress in Max.” In this case, since the Lexicon-Grammar has encoded that the verb dar can be an operator on ter, and since the predicative noun estresse “stress” does not allow for dar to be its support, a general rule can apply, extracting the CAUSE relation expressed by the VopC, in a similar way as for the SUPPORT dependency. The EVENT structure is thus construed as shown below: VOPC(estresse,deu) EVENT(estresse,other) EVENT EXPERIENCER(estresse,Max) EVENT CAUSE(estresse,not´ ıcia)

However, when the same verb can be both a support and an operator verb, in the absence of tell-tale prepositional complements or other syntactic evidence, the detection of the adequate structure can not be done at this stage. We found only 35 predicative nouns which can be associated to the verb dar “give” with both categories, i.e. as a support and a VopC. It is also possible that both dependencies SUPPORT and CAUSE be extracted in order to disambiguate them at a later stage. 99

5

Conclusions and future work

In the near future, we intend to use the data encoded in the Lexicon-Grammar of these predicative nouns and build a SVC identification module for STRING. For the moment, the identification of all the syntactic phenomena, constituting as many different parsing cases as possible, is underway, in order to fully automatize the processing of converting the Lexicon-Grammar tables into the STRING, with XIPcompliant rules, in a similar way as it has already been done for the verbs (Baptista, 2012; Travanca, 2013; Talhadas, 2014). After implementing all the data in STRING we also intend to evaluate the system in order to check the extraction of the dependencies involving the support verbs and predicative nouns. An important task ahead is the systematic comparison of the structures and properties here described against those of European Portuguese. First of all, the set of nouns available in each variant is not exactly the same, even if the concepts are shared; for example, carona in BP corresponds to the EP boleia “ride”; in other cases, the choice of the nominalization suffixes differ: in BP one uses the term parada card´ıaca, while its equivalent in EP is paragem card´ıaca “cardiac arrest”. False-friends are also common: in BP, chamada “rebuke” is unrelated to EP chamada “phone call” (but, in this sense, it is also used in BP); the set of support verbs for each noun are different: as a synonym of rebuke we find the pair dar-levar (only in BP), while as equivalent to phone call the basic support verbs are fazer-receber (the same in BP and EP). Naturally, much in both variants is quite similar, though some patterns begin to emerge: the different choice of prepositions for the complement, mostly the alternation between em “in” in BP and a “to” in EP (both as dative complements); the choice of support verbs, with some being used for these predicative noun exclusively in BP (ganhar “gain” and tomar “take”) or in EP (pregar “throw” and apanhar “take”).

Acknowledgements The authors would like to acknowledge the financial supports from Capes/PDSE under Processes BEX 12751/13-8 and BEX 6561/13-6 and also national funds through FCT – Fundac¸a˜ o para a Ciˆencia e a Tecnologia, under project PEst-OE/EEI/LA0021/2013. We would like to thank the comments of the anonymous reviewers, which helped to improve this paper.

References Jorge Baptista. 1997. Serm˜ao, tareia e facada. Uma classificac¸a˜ o das construc¸o˜ es conversas dar-levar. In Semin´arios de Lingu´ıstica, volume 1, pages 5–37, Faro. Universidade do Algarve. Jorge Baptista. 2005. Sintaxe dos predicados nominais com SER DE. Fundac¸a˜ o Calouste Gulbenkian/Fundac¸a˜ o para a Ciˆencia e Tecnologia, Lisboa. Jorge Baptista. 2012. A Lexicon-grammar of European Portuguese Verbs. In Jan Radimsky, editor, Proceedings of the 31st International Conference on Lexis and Grammar, volume 31, pages 10–16, Czech Republic, September. Universit`a degli Studi di Salerno and University of South Bohemia in Nov´e Hrady. Andr´ee Borillo. 1971. Remarques sur les verbes sym´etriques. Langue Franc¸aise, (11):17–31. M´ırian Bruckschein, Fernando Muniz, Jos´e Guilherme Camargo Souza, Juliana Thiesen Fuchs, Kleber Infante, Marcelo Muniz, Patr´ıcia Nunes Gonc¸alvez, Renata Vieira, and Sandra Maria Aluisio. 2008. Anotac¸a˜ o lingu´ıstica em XML do corpus PLN-BR. S´erie de relat´orios do NILC, NILC- ICMC - USP. Magali Sanches Duran, Carlos Ramisch, Sandra Maria Alu´ısio, and Aline Villavicencio. 2011. Identifying and analyzing Brazilian Portuguese complex predicates. In Proceedings of the Workshop on Multiword Expressions: from Parsing and Generation to the Real World, pages 74–82, Portland, USA. Gregory Grefenstette and Simone Teufel. 1995. Corpus-based Method for Automatic Identification of Support Verbs for Nominalizations. CoRR, cmp-lg/9503010. Maurice Gross. 1975. M´ethodes en syntaxe. Hermann, Paris. Maurice Gross. 1981. Les bases empiriques de la notion de pr´edicat s´emantique. Langages, 63(3):7–52. Gaston Gross. 1982. Un cas des constructions inverses: donner et recevoir. Lingvisticae Investigationes, 2:1–44.

100

Maurice Gross. 1988. Methods and tactics in the construction of a Lexicon-grammar. In The linguistic Society of Korea, editor, Linguistics in the Morning Calm 2. Selected papers from SICOL-1986, pages 177–197, Seoul. Hanshin Publishing Company. Gaston Gross. 1989. Les constructions converses du franc¸ais. Droz, Genebra. Maurice Gross. 1996. Lexicon grammar. In K. Brown and J. Miller, editors, Concise Encyclopedia of Syntactic Theories, pages 244–259. Pergamon, Cambridge. Maurice Gross. 1998. La fonction s´emantique des verbes supports. In B´eatrice Lamiroy, editor, Travaux de Linguistique, number 37, pages 25–46. Patrick Hanks, Anne Urbschat, and Elke Gehweiler. 2006. German light verb constructions in corpora and dictionaries. International Journal of Lexicography, 19(4):439–457. Zellig Harris. 1964. The Elementary Transformations. Transformations and Discourse Analysis Papers, (54):211– 235. Zellig Harris. 1991. A Theory of Language and Information: A Mathematical Approach. Clarendon Press, New York. Iris Hendrickx, Am´alia Mendes, S´ılvia Pereira, Anabela Gonc¸alves, and Inˆes Duarte. 2010. Complex predicates annotation in a corpus of Portuguese. in: Proceedings of the 4th ACL. In Proceedings of the 4th ACL Linguistic Annotation Workshop, pages 100–108, Uppsala, Sweden. B´eatrice Lamiroy. 1998. Le Lexique-grammaire: Essai de synth`ese. In B´eatrice Lamiroy, editor, Travaux de Linguistique, volume 37, pages 7–23. Nuno Mamede, Jorge Baptista, Vera Cabarr˜ao, and Cl´audio Diniz. 2012. STRING: An hybrid statistical and rule-based natural language processing chain for Portuguese. In International Conference on Computational Processing of Portuguese (Propor 2012), volume Demo Session, Coimbra, Portugal, April. Salah Ait Mokhtar, Jean-Pierre Chanod, and Claude Roux. 2002. Robustness beyond shalowness: Incremental dependency parsing. Natural Language Engineering, pages 121–144. Elisabete Ranchhod. 1990. Sintaxe dos predicados nominais com Estar. INIC - Instituto Nacional de Investigac¸a˜ o Cient´ıfica, Lisboa. Amanda Rassi and Oto Vale. 2013. Predicative Nouns Suffixation associated to the verb dar (give) in Brazilian Portuguese. In Jorge Baptista and Mario Monteleone, editors, Proceedings of the 32nd International Conference on Lexis and Grammar, volume 32, pages 151–158, Faro, September. UAlg. Amanda Rassi, Nathalia Perussi, Jorge Baptista, and Oto Vale. 2014. Estudo contrastivo sobre as construc¸o˜ es conversas em PB e PE. In Cristina Fargetti, Odair Silva, Clotilde Murakawa, and Anise Ferreira, editors, Anais do I CINELI - Congresso Internacional Estudos do L´exico e suas Interfaces, volume 1, page (no prelo), Araraquara-SP, Maio. Universidade Estadual Paulista - UNESP. Maria Cristina Andrade Santos-Turati. 2012. Descric¸a˜ o da estrutura argumental dos predicados nominais com o verbo-suporte ter. In Semin´ario do GEL, number 60, pages 20–21, S˜ao Paulo, Brasil. Grupo de Estudos Lingu´ısticos do Estado de S˜ao Paulo - GEL. Angelika Storrer. 2007. Corpus-based investigations on german support verb constructions. In Christiane Fellbaum, editor, Collocations and Idioms: Linguistic, Lexicographic, and Computational Aspects, pages 164–188. Continuum Press, London. Rui Talhadas. 2014. Automatic Semantic Role Labeling for European Portuguese. Master’s thesis, Universidade do Algarve, Faro. Tiago Travanca. 2013. Verb Sense Disambiguation. Master’s thesis, Instituto Superior T´ecnico, Universidade de Lisboa, June. Aldina Vaza. 1988. Estruturas com nomes predicativos e o verbo-suporte dar. Master’s thesis, Faculdade de Letras - Universidade de Lisboa.

101

Loading...

The fuzzy boundaries of operator verb and support verb constructions

The fuzzy boundaries of operator verb and support verb constructions with dar “give” and ter “have” in Brazilian Portuguese Amanda Rassi1,2,3 , Cristi...

161KB Sizes 1 Downloads 16 Views

Recommend Documents

THE VERB
I couldn't believe she (make) ................................. a phone call during the lecture. I asked what was going

Regular verb and Irregular verb | chelsea's doc
Jun 18, 2012 - Produk yang diperoleh dari reaksi antara logam alkali dengan oksigen yakni berupa oksida logam. Berikut r

Verb patterns: verb + infinitive or verb + - ing ? - gramática inglés en
Let, make. Let and make are followed by an infinitive without to in active voice sentences. They always have an object (

THE VERB TENSES
T he Present tense describes an action which is happening at the moment : Ex: I am listening = yo escucho.. It also desc

The Auxiliary Verb
Will, be = auxiliary verbs; growing = present participle completing the verb phrase. Soon, Alex will be eating organic c

Spanish verb conjugation -- Verbix verb conjugator
Conjugate Spanish verbs on-line. ... Spanish verbs are conjugated in four moods, four simple tenses, and in six persons.

Verb 'To Cut' - Irregular Verb Definition - UsingEnglish.com
Irregular verb definition for 'to Cut', including the base form, past simple, past participle, 3rd person singular, pres

VERB SHIFTS
VERB SHIFTS. PS-17. Revised Summer 2012. 1. Sentences often include more than one verb, and, if so, the verbs often refe

VERB TENSES
I (go) have been going to the circus ever since I was a child. 6. a. Right now, I (see) (can) see two elephants doing tr

Subject – Verb Agreement and VERB as Complements | fenichat
Mar 28, 2011 - Mereka berani menunjukkannya, dan pasti akan terlihat dalam perubahan mimik wajah atau gerak tubuh," tuka