AN INTERLINGUA REPRESENTATION BASED ON THE LEXICO-SEMANTIC INFORMATION Irene Castellón* Ana Fernández** M. Antònia Martí* Roser Morante* Glòria Vázquez*** Universitat de Barcelona(*) Universitat Autònoma de Barcelona(**) Universitat de Lleida (***) Abstract The objective of this paper is to present a proposal of representation for a lexical entry that provides the necessary information required to express the characteristics of its behavior in a particular language. In addition, the representation of an interlingua representation is proposed. For this representation, which will be common to all the treated languages, a logical form associated with each type of lexical entry will be used. 1. Introduction Our representation proposal of an interlingua is based on the relation established between the argument, the syntax and the event structure in the logical form. From this representation, each language generates different syntactic realizations. The representation proposal arises from the investigation that has been carried out in the framework of the Pirápides project. This project has as its main objective the creation of a multilingual lexical knowledge base (LKB) that is being implemented currently for the above mentioned languages. In this LKB, we collect the information that has been considered as relevant to account for the syntactic and semantic characteristics of the predicates. We consider this information to be that which a speaker of a language needs in order to understand and generate messages appropriately. The main objective of this work is to present the interlingua representation of the verbal lexical entries and the different syntactic realizations that are allowed. With this end in mind, a type hierarchy, a model of a lexical entry and a series of lexical rules that permit the generation of new entries have been developed. We have chosen as an example a verb that presents an ambiguity in the pronominal Spanish and Catalan forms. The two interpretations are the agentive one (reflexive diatheses) and the anticausative one (anticausative diatheses). In the English language every interpretation corresponds to two different lexical items. In this way, we can see how our representation system allows the treatment of the lexical mismatches that exist between these three languages. In the first place we will present the theoretical framework that has been developed within the Pirápides project [Fernández et al 98]. The representation that we propose for the lexical verb items is based on the specification of information that belongs to different linguistic areas: meaning components, diatheses alternations and the argument and event structure. The next section (3) is devoted to the linguistic analysis of the class to which the chosen lexical item belongs. To accomplish this, we will show both what this verb has in common with the rest of the predicates that belong to the semantic class as well as its individual characteristics. In the fourth section we detail the implementation of the model in the LKB and we will also see the different levels that form our theoretical framework. Notice will be paid to the way in which these different levels interact and also to the information that is shared or coindexed. In section 5, we will see the implementation of an entry. In the first place the entry considered as the basic one is detailed, and from it the anticausative and reflexive forms will be generated through the application of lexical rules. From this application new lexical entries and their logical forms will be obtained. These logical forms are what we use as our interlingua. Finally in section 6, a passage of the text will be represented to see how the logical information obtained from the entries can be used as the guideline for the generation of the logical form of the whole sentence. 2. Theoretical Framework From a theoretical point of view, the first problem addressed in the project has been that of how to define a model for the lexical entry, both formally and theoretically. As far as formal representation of the information is concerned, strict criteria of explicitness have been introduced so that the model can be used for computational purposes; that is to say, the information that is specified in each lexical entry is strictly codified and kept in a format retrievable for natural languages processing systems [Copestake 1992]. The theoretical foundations of the model are somewhat eclectic in character. We believe that in order to analyze verbal behavior appropriately it is necessary to keep the following three items in mind when studying predicates: meaning components, syntactic behavior (diatheses alternations) and argument and event structure. With this hypothesis as our starting point we have used these items as the criteria for a classification of the verbs. In addition, taking Pustejovsky's Generative Lexicon [Pustejovsky 1995] as a reference, we have used all this information and incorporated it into the lexical items to carry out the implementation. Both the theoretical framework adopted and the development of the multilingual computational lexicon imply that a definition of these modules and the way they relate to each other has to be made explicit as we will see below. 2.1 Meaning Components The definition of the meaning components in this framework is not based on determining the type of semantic relation that is established between the verb and its arguments. Rather, it is considered that the verb and its arguments form a conceptual unit that can be expressed using diverse linguistic resources (s. section 2.2). In this paper we are going to analyze predicates characterized as being events and for which we have defined two fundamental meaning components: the initiator and the entity. The former makes reference to the participant that initiates the event and can coincide with the entity. The implied entity, on the other hand, is the argument which the predication is about. Moreover, specific semantic classes can be distinguished according to the requirement of other meaning components such as trajectory or change. In section 3, in which the knowledge base is defined, this type of information is codified with the label M-COMP, defined in terms of the feature COMP, that has predefined as possible values the list of meaning components previously defined (initiator, entity, etc.) and an index that allows the coindexation of these components with one of the arguments described in the argument structure. 2.2 Diatheses The diatheses alternations are understood as the different possibilities of syntagmatic expression that a predicate presents according to the possibilities of focalization on the different semantic aspects it covers, either in the event structure or in the meaning components. Each possibility of focalization gives birth to different meaning oppositions, even though the same sense of the verb is always kept. In [Fernández and Martí 98] a first approximation is presented and in it four possible semantic oppositions based on focalization are considered. These semantic oppositions can be expressed through diverse procedures: at the lexical level, where the item allows both structures or two items are involved; at the morphological level, that usually implies adding a free morpheme to the verb; and at the syntactic level, when an auxiliary construction is required. The information about the diatheses is introduced in the LKB in the feature SINTAX, that holds the feature structures SUBCAT and DIAT. In the latter structure, there are expressed not only the diatheses alternations the item participates in but also the possible expressions (D-LEX, D-MORF, D-SINT) previously mentioned. 2.3 Event structure As has been observed in [Pustejovsky 91, 95] and [Rosen 96], differences in the event structure is a factor to be taken into account when describing the syntactic behavior of the verbs and, moreover, it provides semantic information. As a typology of predicates, a modified version of [Vendler 97] is considered. So, two basic types are established, the states and the events. The latter are further subdivided, following [Pustejovsky 91], into simple events (EEVENT-SIMPLE) and complex events (EEVENT-COMPLEX). Moreover, for this latter type of event the relations that can be established between them and the event that is functioning as the head are also declared. 3. The semantic class of the predicates of change The predicates of change are characterized by their sharing of a series of common features with regard to the three level described in the previous section: they have an event structure of a complex type, all of them participate in the diatheses alternation known as the anticausative, and finally all of them share the same meaning components: change, entity and initiator (fig. 1). Fig. 1: Meaning components of the predicates of change The meaning components can be expressed sintagmatically or else they can be incorporated. This latter case also includes the lexicalization. The greater part of the predicates that belong to this class present the meaning component `change' incorporated into the lexeme as is the case of ampliar (broaden). In the case of the predicate convertir (convert), this component is expressed sintagmatically through a prepositional component. This will be reflected in the representation with the definition of two different types of subcategorization for the predicates of change As far as event structure is concerned, these predicates are characterized for presenting a complex event structure that consists of an event limited by a state of the resulting type as is shown in figure 2. This behavior is homogenous in the entire group: Fig. 2: Event structure of the predicates of change The diatheses alternation characteristic of the predicates of change is the anticausative. This diatheses belongs to a more general group that has been called `change of focus with information loss'. Of the structures involved in the opposition, one contains more information that the other. These verbs present three possible patterns of behavior: convertir takes the intransitive through morphological mechanisms (attaching the se pronoun) (fig.3) whereas pasar, for instance, is inherently intransitive and participates in the transitive construction with a syntactic auxiliary hacer (make), and hervir (boil), on the other hand, requires neither an auxiliary nor a pronoun in either form. In the three cases there is argument switch. Fig. 3: Syntactic alternations of the predicates of change The predicates of this class can also participate in other alternations such as the passive (either one) and the reflexive. 4. Implementation of the application world Our representation proposal seeks to be a reusable system for the different applications of NLP (automatic translation, generation, analysis, etc.) The unit subject to study is the sentence, within which the predicate is taken as the head. For this reason, we have included in the verbal entry the information necessary to generate the logical form of the sentence. This information is extracted, on the one hand, from the lexical entry and, on the other, from the compositional analysis of the arguments that accompany the verb in the sentence. We define as arguments those complements that are compulsory in the sentence and that can be considered the sintagmatic realization of the meaning components. The optionality of an argument is given by a lexical rule application. The lexical knowledge base (LKB) used is composed of two basic levels [Copestake 93]: on the one hand we have defined a type system and on the other we have the lexical entries and the lexical rules. Every object of the LKB is represented using feature structures. In the first one of these two levels, all the objects necessary for describing the entries are declared. The entries are described in a hierarchy. The second one of these levels is composed of the lexical entries and the lexical rules always associated with a type on the first level. This structure allows the generalization of certain informative structures thus facilitating the construction of the lexicon. The lexical rules make possible the generation of new lexical entries from those already declared in the lexicon. This type of generalization has allowed us to create a representation system that is both economical and powerful in that it allows us to express the data through inheritance mechanisms. Therefore it is not an indispensable requirement to make all the information explicit for each entry. The system is also sufficiently open as to allow the user to especify peculiar features within each entry. We have declared two types of lexical entries according to two possible logical forms and to the event structure type. We differentiate two types of predicates: those that give rise to a simple semantic representation (fig. 4) and those that give rise to a complex one (fig. 5). Fig. 4: Simple lexical entry Both, the simple lexical entry as well as the complex lexical entry share the following feature structures: SENSE-ID, ORT, FON, MORF, SINTAX, EE, SEM and FORLOG. In this paper we will focus on the last four modules. As far as SINTAX, SEM (semantics) and EE (event structure), these structures hold the information from which we will obtain the logic representation (FORLOG), that forms our proposal of an interlingua. The relation between the information that is contained in these structures is reflected through the relations of reentrancy that are established between its components and its coindexation in FORLOG. Fig. 5: Complex lexical entry Starting from the idea that certain verbs present a similar behavior in determined linguistic levels (semantics, syntax and argument structure), certain feature structures have been defined with the aim of accounting for all the information that is shared and therefore susceptible of being generalized. In the first of these levels, two types of information are observed, the meaning components and the event structure; in the syntax we have created patterns for the alternation of diatheses as well as for different subcategorization possibilities; lastly, we have also proposed different grouping possibilities according to the argument structure. Next, we will present with greater detail the modules relevant for this work. Regarding SINTAX (fig. 6), this element holds the elements subcategorized by the verb and the diverse syntactic realizations in which this verb participates. The former information is expressed as the value for the descriptor SUBCAT, that has been defined as the recursive list where the basic element is COMPL. Each COMPL is defined as being formed by two components: CAT and SEMREF. In CAT the syntactic category of the complement is indicated (NP, PP, ...) and only argument complements are included. On the other hand, the value SEMFEF belongs to an ontological type (animated, object, ...) that has an index assigned that relates it with an element of the argument structure. DIAT consists of four elements that make reference to the four basic oppositions of a semantic nature around which the different syntactic realizations (diatheses alternations) are organized. Each one of these oppositions is defined in terms of four features. The first one of these features, DIAVAL, is a boolean that indicates the participation of the verb in that particular opposition. Fig. 6: The SINTAX module The other three, D-LEX, D-SINT and D-MORF, are used to activate or to block the application of the lexical rules that generate the alternations of diatheses. This information is fundamental for the generation of the output in the target language from the data contained in FORLOG. The module SEM (fig.7) contains the information about the argument structure (EST-ARGUM). It serves as a bridge between the syntax and the semantics. The latter is expressed in terms of the meaning components (COMP-SIGNIFICADO), among which we have differentiated between those that appear syntactically and those that do not. In this last group we refer to both those that are understood and those that are incorporated. As far as event structure (EE) is concerned, two types of predetermined structures are defined, the simple and the complex. Each one of them also presents different subtypes. In the case of complex event structures, the relation established among the diverse events that conform it is expressed through the feature RESTR, (three relations of this type have been established so far: precedence, parallelism and overlapping). Finally, with the feature HEAD we mark the event focussed on that particular structure. Fig. 7: The SEM module Finally, FORLOG has as its objective to relate the events with the predicate and the implied arguments. As we have already commented, we have defined two types of FORLOG, the simple type (fig. 4) and the complex type (fig. 5). The elements contained in both types are the following: PRE (predicate), whose value corresponds to the orthography of the entry; OPERATOR, that allows to express the operators of negation, possibility and necessity; EVE, the type of event coindexed with the event structure; and, finally, the arguments that are required (ARGLIST) coindexed with the semantic module and also the syntactic. The complex forms are a concatenation of simple forms joined with a value associated to CONECTIVA(and/or). 5. Implementation of an entry The sentence in which we have based our proposal for a representation is the following: A. En cinco años BancoSol se ha convertido en un banco comercial especializado en microfinanza. B. In five years BancoSol itself has converted into a bank commercial specialized in microfinance. C. In five years BancoSol has transformed itself into a commercial bank which specializes in microfinance. In Spanish this sentence (A) presents a predicate, convertirse, that is ambiguous. The possible readings of this form are: the anticausative construction and the reflexive construction. These constructions are different as for the irrelevance and the presence of the agentivity respectively. In English it is common to use the verb convert in the transitive non reflexive constructions (The witch converted the prince into a frog). On the other hand, become and transform are reserved for the anticausative and the reflexive constructions respectively. Then, in the mentioned context, the option B, with convert, is unusual in English, whereas the verb proposed in the text C, transform, is a good option. In this case the verb become would not be possible since the reflexive interpretation is more adequate in this context than the anticausative one. In order to relate those three verbs correctly and that the translation system could choose the adequate lexical form in every case, we propose as a basic entry the most explicit, convert. If we apply to it the corresponding lexical rules, the adequate lexical variants (become and transform) will be obtained. Those variants correspond with the pronominal forms in Spanish (convertirse), which will also be obtained applying the lexical rules to the verb convertir. We have built the entry convertir for the Spanish and the entry convert (see appendix) for the English. Both entries have been associated with an entry of the type `cambio', which contains the common characteristics of this kind of verbs in both languages. As we can observe, one of the differences between these entries is that whereas the means to form the diatheses in Spanish is morphological (D-MORF), a lexical mechanism (D-LEX) is used in English. Besides, we have developed two objects (called `regla-anticausativa' and `regla-reflexiva') in the LKB to which we have associated the lexical rules that are needed for this transformation. These two rule-types generalize all that information common for both languages in the input and the output entries. The information pertinent for a subgroup of predicates or for one of the languages is specified in the lexical rules that inherit from these two types. In this case the specified information would be the syntactic realizations. The lexical rules will be applied to the basic entries in order to generate new entries that will allow individual interpretations. The new entry will inherit from the basic one those informative structures that are not affected in that interpretation and, in addition, it will contain specific information provided by the lexical rule. {{entrada complex} {ORTH}{ind{0}} {transform} {SENSE-ID}{{{sense-id} {FS-ID}{transform} {LANGUAGE}{english} {WORD}{ind{0}} {N-SENTIDO}{1}}} {FON}{string} {MORF}{transfom} Fig. 8: Lexical entry of transform In the figure 8 we exemplify the application of the reflexive rule and in the figure 9 the anticausative one for the English language. Fig. 9: Lexical entry of become To finish this section, we would like to mention that the translations are linked by using a bilingual dictionary that allows the application of a specific translation rule for each case. This bilingual dictionary is composed of translation links [Copestake 93] between lexical entries from different languages. Every translation link includes the application of a rule for each language and that allows us to make the adequate transformation. 6. Analysis of a fragment In order to check the capacity of representation in a complex predicate of the structure provided by the module called FORLOG, we will analyze a complex sentence of the text to see how all the information required by the logical form can be retrieved from the information provided by the lexical entries. La finalidad del programa (bautizado PRODEM) era ofrecer a los trabajadores independientes más desfavorecidos posibilidades de crédito y de formación para ampliar sus perspectivas de empleo, alentar las inversiones en las microempresas y aumentar la rentabilidad del sector. The interlingua representation that we propose for this sentence is obtained from the information contained in the modules of the lexical entry: P[ser(e1,x,Q) and Q[ofrecer(e2,y,z,w,R)] and R[(ampliar(e3,z,v)) and (alentar(e4,z,t))and (aumentar,e5,z,u))]] The instantiation of the variables with the information from the text would be carried out in the following terms: sentence1[ser(estado,la_finalidad,oración2) and sentence2[ofrecer(evento-simple, posiblidades_de crédito_y_formación, el_programa, a_los_trabajadores(...), oración3)] and sentence3[(ampliar(evento-complex, posibilidades_de_crédito_y_formación, sus_perspectivas_de_empleo) and (alentar, posibilidades_de_crédito_y_formación, las_inversiones) and (aumentar, posibilidades_de_crédito_y_formación, la rentabilidad)]]. As can be observed, this sentence includes a central predicate P. It is composed of an event of a simple type (state-type) that has as its arguments a subject x and an attribute, that is another predicate which we have called Q. This information is provided at the subcategorization and syntactic levels. The predicate named Q is made up of the predicate ofrecer and its four arguments: a subject y, a direct complement z, an indirect complement w and a prepositional complement, the predicate R. The latter contains three infinitive sentences which are coordinated. All of them share the same subject that is instantiated with the direct complement z of the ofrecer predicate. Besides, these three predicates (ampliar, alentar and aumentar) each has as their argument a direct complement: v, t and u, respectively. The feature structure that we are using as interlingua is FORLOG. From this module defined for the Spanish we can access the representation of the Spanish predicates in the LKB. From this information and through the use of links to the English and Catalan entries we can generate the equivalent sentences in these two languages: The aim of the program (named PRODEM) was to offer the most disadvantaged independent workers the possibility of credit and of training to broaden their perspectives of employment, to encourage investment in microenterprises and to increase the profitability of the sector. La finalitat del programa (batejat PRODEM) era oferir als treballadors independents més desfavorits possibilitats de crèdit i de formació per tal d'ampliar llurs perspectives de treball, encoratjar les inversions a les microempreses i augmentar la rendibilitat del sector. Conclusion In this paper, a theoretical model for the representation of lexical entries in an LKB has been presented. This model integrates the linguistic information that we have considered relevant to account for the properties and behavior of the verbs: phonology, orthography, morphology, syntax and semantics, as well as multilingual relations. We consider that in the areas of syntax and semantics, the linguistic fields in which we have centered ourselves, the information about these three aspects must be expressed: meaning components, event structure and diatheses. The relation between them is made explicit in the logical form that is included in the proposed entry and that serves as the interlingua in the translation process. The level of specification allows one to account for the mismatches between the studied languages and those that are produced by the intrinsic characteristics of the different modules. These mismatches are dealt with in the module in which a divergence is located. In this way, two equivalent verbs of the different languages can participate in the same type of semantic opposition even though it is possible that they differ in the type of mechanism they use in order to express it (syntactic, morphological or lexical). This model of representation, based on feature structures and mechanisms of inheritance, allows us to express in detail those relevant aspects in the characterization of the verb lexical entries and therefore it allows us to express both the generalizations as well as the lexical peculiarities. This has permitted us to define different types of predicates at different levels. In some cases, the generalization can be made combining more than one module (syntax and semantics) and some other times the proposed pattern takes the values of only one component. With this procedure the representation of data is economized since each type will constitute a model of verbal behavior. In this manner all the verb that follow a model will inherit all the common features of that model and instantiate those that are their own. This mechanism will thus permit us to capture all the relevant generalizations from a linguistic point of view. REFERENCES [Copestake 93] Copestake, A. (1993) The compleat LKB, Acquilex II Deriverable [Fernández y Martí 98] Fernández, A. y M. A. Martí (1998) "Alternancias de diátesis: estudio comparativo entre el comportamiento de los verbos en castellano e inglés. Hacia una propuesta de tipología de desajustes verbales" (en preparación). [Pustejovsky 91] Pustejovsky, J. (1991) 'The syntax of event structure', Cognition págs. 47-81. [Pustejovsky 95] Pustejovsky, J. (1995) The generative lexicon, The MIT Press, Cambridge MA. [Rosen 96] Rosen, S.Th. (1996) 'Events and verb classification', Linguistics 34, p. 191-223. [Vendler 67] Vendler, Z. (1967) Linguistics in Philosophy, Cornell University Press, Ithaca. APPENDIX CONVERTIR_1 {{entrada-tipo-cambio3} {ORTH}{ind{0}} {convertir} {SENSE-ID}{{{sense-id} {FS-ID}{convertir_1} {LANGUAGE}{spanish} {WORD}{ind{0}} {N-SENTIDO}{string}}} {FON}{string} {MORF}{convert-} {SINTAX}{{{sintax} {SUBCAT}{{{subcat-3com2} {COMPL}{{{compl} {CAT}{nominal} {SEMREF}{ind{1}} {todo}}} {RESTCOMP}{{{subcat} {COMPL}{{{compl} {CAT}{nominal} {SEMREF}{ind{2}} {todo}}} {RESTCOMP}{{{subcat} {COMPL}{{{compl} {CAT}{{{sp} {PREP}{en}}} {SEMREF}{todo}}} {RESTCOMP}{nul}}}}}}} {DIAT}{{{tipo-dia-cambio2} {CAMBIO-ENFOQUE-GEN}{{{cambio-enf-gen} {REFLEXIVA}{true} {CAMBIO-PREP}{{{diat} {DIATVAL}{bool}}}}} {CAMBIO-ENFOQUE-PERD}{{{cambio-enf} {ANTICAUSATIVA}{{{diatrue} {DIATVAL}{true} {D-LEX}{string} {D-SINT}{string} {D-MORF}{true}}} {PASIVA}{{{diatrue} {DIATVAL}{true} {D-LEX}{string} {D-SINT}{true} {D-MORF}{string}}}}} {GENERALIZACION}{{{general} {SUJETO}{{{diat} {DIATVAL}{bool}}} {COMPLEMENTO}{{{tipo-diat-com} {ARGUM2}{{{diat} {DIATVAL}{bool}}} {ARGUM3}{{{diat} {DIATVAL}{bool}}} {ARGUM2-3}{{{diat} {DIATVAL}{bool}}}}}}} {CAMBIO-ASPECTUAL}{{{cambio-aspec} {CON-CRUCE}{{{diatrue} {DIATVAL}{true} {D-LEX}{string} {D-SINT}{string} {D-MORF}{true}}} {SIN-CRUCE}{{{diat} {DIATVAL}{bool}}}}}}}}} {SEM}{{{sem-tipo-cambio3} {EST-ARGUM}{{{est-argum-tipo-cambio3} {ARG1}{ind{1}} {RESTARG}{{{est-argumental} {ARG1}{ind{2}} {RESTARG}{{{est-argumental} {ARG1}{ind{3}} {todo} {RESTARG}{est-argum}}}}}}} {COMP-SIGNIFICADO}{{{mcomlist-tipo-cambio3} {MCOM}{{{mcom} {COMP}{iniciador} {SINTAGMATICO}{true} {INDEX}{ind{1}}}} {RESTCOM}{{{mcomlist} {MCOM}{{{mcom} {COMP}{entidad} {SINTAGMATICO}{true} {INDEX}{ind{2}}}} {RESTCOM}{{{mcomlist} {MCOM}{{{mcom} {COMP}{cambio} {SINTAGMATICO}{true} {INDEX}{ind{3}}}} {RESTCOM}{mcomlista}}}}}}}}} {EEVENT}{{{eevent-complex-cambio2} {E1}{ind{4}} {evento} {E2}{ind{5}} {estado} {REST}{precedencia} {HEAD}{e1}}} {FORLOG}{{{forlog-complex} {CONECTIVA}{and} {FOR1}{{{forlog-simple} {OPERADOR}{operador} {PRE}{ind{0}} {EVE}{ind{4}} {ARGLIST}{{{est-argumental} {ARG1}{ind{1}} {RESTARG}{{{est-argumental} {ARG1}{ind{2}} {RESTARG}{{{est-argumental} {ARG1}{ind{3}} {RESTARG}{est-argum}}}}}}}}} {FOR2}{{{forlog-simple} {OPERADOR}{operador} {PRE}{ind{0}} {EVE}{ind{5}} {ARGLIST}{{{est-argumental} {ARG1}{ind{2}} {RESTARG}{{{est-argumental} {ARG1}{ind{3}} {RESTARG}{nul}}}}}}}}}} CONVERT_1 {entrada-tipo-cambio3} {ORTH}{ind{0}} {convert} {SENSE-ID}{{sense-id} {FS-ID}{convert_1} {LANGUAGE}{english} {WORD}{ind{0}} {N-SENTIDO}{string}}} {FON}{string} {MORF}{convert/ed/ed} {SINTAX}{{sintax} {SUBCAT}{{subcat-3com2} {COMPL}{{compl} {CAT}{nominal} {SEMREF}{ind{1}} {todo}}} {RESTCOMP}{{subcat} {COMPL}{{compl} {CAT}{nominal} {SEMREF}{ind{2}} {todo}}} {RESTCOMP}{{subcat} {COMPL}{{compl} {CAT}{{sp} {PREP}{into}}} {SEMREF}{todo}}} {RESTCOMP}{nul}}}}}}} {DIAT}{{tipo-dia-cambio2} {CAMBIO-ENFOQUE-GEN}{{cambio-enf-gen} {REFLEXIVA}{transform} {CAMBIO-PREP}{{diat} {DIATVAL}{bool}}}}} {CAMBIO-ENFOQUE-PERD}{{cambio-enf} {ANTICAUSATIVA}{{diatrue} {DIATVAL}{true} {D-LEX}{become} {D-SINT}{string} {D-MORF}{true}}} {PASIVA}{{diatrue} {DIATVAL}{true} {D-LEX}{string} {D-SINT}{true} {D-MORF}{true}}}}} {GENERALIZACION}{{general} {SUJETO}{{diat} {DIATVAL}{bool}}} {COMPLEMENTO}{{tipo-diat-com} {ARGUM2}{{diat} {DIATVAL}{bool}}} {ARGUM3}{{diat} {DIATVAL}{bool}}} {ARGUM2-3}{{diat} {DIATVAL}{bool}}}}}}} {CAMBIO-ASPECTUAL}{{cambio-aspec} {CON-CRUCE}{{diatrue} {DIATVAL}{true} {D-LEX}{string} {D-SINT}{string} {D-MORF}{true}}} {SIN-CRUCE}{{diat} {DIATVAL}{bool}}}}}}}}} {SEM}{{sem-tipo-cambio3} {EST-ARGUM}{{est-argum-tipo-cambio3} {ARG1}{ind{1}} {RESTARG}{{est-argumental} {ARG1}{ind{2}} {RESTARG}{{est-argumental} {ARG1}{ind{3}} {todo} {RESTARG}{est-argum}}}}}}} {COMP-SIGNIFICADO}{{mcomlist-tipo-cambio3} {MCOM}{{mcom} {COMP}{iniciador} {SINTAGMATICO}{true} {INDEX}{ind{1}}}} {RESTCOM}{{mcomlist} {MCOM}{{mcom} {COMP}{entidad} {SINTAGMATICO}{true} {INDEX}{ind{2}}}} {RESTCOM}{{mcomlist} {MCOM}{{mcom} {COMP}{cambio} {SINTAGMATICO}{true} {INDEX}{ind{3}}}} {RESTCOM}{mcomlista}}}}}}}}} {EEVENT}{{eevent-complex-cambio2} {E1}{ind{4}} {evento} {E2}{ind{5}} {estado} {REST}{precedencia} {HEAD}{e1}}} {FORLOG}{{forlog-complex} {CONECTIVA}{and} {FOR1}{{forlog-simple} {OPERADOR}{operador} {PRE}{ind{0}} {EVE}{ind{4}} {ARGLIST}{{est-argumental} {ARG1}{ind{1}} {RESTARG}{{est-argumental} {ARG1}{ind{2}} {RESTARG}{{est-argumental} {ARG1}{ind{3}} {RESTARG}{est-argum}}}}}}}}} {FOR2}{{forlog-simple} {OPERADOR}{operador} {PRE}{ind{0}} {EVE}{ind{5}} {ARGLIST}{{est-argumental} {ARG1}{ind{2}} {RESTARG}{{est-argumental} {ARG1}{ind{3}} {RESTARG}{nul}}}}}}}}}} {SINTAX}{{{sintax} {SUBCAT}{{{subcat} {COMPL}{{{compl} {CAT}{nominal} {SEMREF}{ind{1}} {todo}}} {RESTCOMP}{{{subcat} {COMPL}{{{compl} {CAT}{pron} {SEMREF}{ind{1}}{todo}}} {RESTCOMP}{{{subcat} {COMPL}{{{compl} {CAT}{{{sp} {PREP}{into}}} {SEMREF}{ind{2}} {todo}}}}}}}}} {DIAT}{nul}}} {SEM}{{{sem} {EST-ARGUM}{{{est-argumental} {ARG1}{ind{1}} {RESTARG}{{{est-argumental} {ARG1}{ind{1}} {RESTARG}{{{est-argumental} {ARG1}{ind{2}} {todo} {RESTARG}{est-argum}}}}}}} {COMP-SIGNIFICADO}{{{mcomlist-tipo-cambio3} {MCOM}{{{mcom} {COMP}{iniciador} {SINTAGMATICO}{true} {INDEX}{ind{1}}}} {RESTCOM}{{{mcomlist} {MCOM}{{{mcom} {COMP}{entidad} {SINTAGMATICO}{true} {INDEX}{ind{1}} {todo}}} {RESTCOM}{{{mcomlist} {MCOM}{{{mcom} {COMP}{cambio} {SINTAGMATICO}{true} {INDEX}{ind{2}}}} {RESTCOM}{mcomlista}}}}}}}}} {EEVENT}{{{eevent-complex-cambio2} {E1}{ind{4}} {evento} {E2}{ind{5}} {estado} {REST}{precedencia} {HEAD}{e1}}} {FORLOG}{{{forlog-complex} {CONECTIVA}{and} {FOR1}{{{forlog-simple} {OPERADOR}{operador} {PRE}{ind{0}} {EVE}{ind{4}} {ARGLIST}{{{est-argumental} {ARG1}{ind{1}} {RESTARG}{{{est-argumental} {ARG1}{ind{2}} {RESTARG}{{{est-argumental} {ARG1}{ind{3}} {RESTARG}{est-argum}}}}}}}}} {FOR2}{{{forlog-simple} {OPERADOR}{operador} {PRE}{ind{0}} {EVE}{ind{5}} {ARGLIST}{{{est-argumental} {ARG1}{ind{2}} {RESTARG}{{{est-argumental} {ARG1}{ind{3}} {RESTARG}{nul}}}}}}}}}} {{entrada-simple} {ORTH}{ind{0}} {become} {SENSE-ID}{{{sense-id} {FS-ID}{become} {LANGUAGE}{english} {WORD}{ind{0}} {N-SENTIDO}{1}}} {FON}{string} {MORF}{become} {SINTAX}{{{sintax} {SUBCAT}{{{subcat} {COMPL}{{{compl} {CAT}{nominal} {SEMREF}{ind{1}} {todo}}} {RESTCOMP}{{{subcat} {COMPL}{{{compl} {CAT}{nominal} {SEMREF}{ind{3}} {todo}}} {RESTCOMP}{nul}}}}} {DIAT}{nul}}} {SEM}{{{sem-tipo-cambio3} {EST-ARGUM}{{{est-argum-tipo-cambio3} {ARG1}{ind{2}} {todo} {RESTARG}{{{est-argumental} {ARG1}{ind{1}} {RESTARG}{{{est-argumental} {ARG1}{ind{3}} {todo} {RESTARG}{est-argum}}}}}}} {COMP-SIGNIFICADO}{{{mcomlist-tipo-cambio3} {MCOM}{{{mcom} {COMP}{iniciador} {SINTAGMATICO}{false} {INDEX}{ind{2}}}} {RESTCOM}{{{mcomlist} {MCOM}{{{mcom} {COMP}{entidad} {SINTAGMATICO}{true} {INDEX}{ind{1}}}} {RESTCOM}{{{mcomlist} {MCOM}{{{mcom} {COMP}{cambio} {SINTAGMATICO}{true} {INDEX}{ind{3}}}} {RESTCOM}{mcomlista}}}}}}}}} {EEVENT}{{{eevent-simple} {E1}{ind{4}} {estado}}} {FORLOG}{{{forlog-simple} {OPERADOR}{operador} {PRE}{ind{0}} {EVE}{ind{4}} {ARGLIST}{{{est-argumental} {ARG1}{ind{1}} {RESTARG}{{{est-argumental} {ARG1}{ind{3}} {RESTARG}{nul}}}}}}}}