Please use this identifier to cite or link to this item: http://hdl.handle.net/11707/4457
Title: Syntax and Semantics of Adjectives in Portuguese. Analysis and Modelling.
Authors: Mendes, Sara
Keywords: Indo-European Languages
Romance Languages
Portuguese
Issue Date: 2009
Publisher: Universidade de Lisboa
Abstract: In this dissertation we approach the syntax and semantics of adjectives in languages like Portuguese. More than any other POS, adjectives can take different meanings depending on their linguistic context. Our approach involves an integrated perspective of syntax and semantics, and, even if its main motivation is linguistic, the research is developed under a computational perspective, aiming at the modelling of the linguistic properties of adjectives, in view of their computation in large scale lexica and grammars. Thus, the research presented here is a deep study of linguistic phenomena, modelled in an operatively efficient way, and aiming to meet the following goals: (i) determining adjectives prototypical features, while establishing the relations holding between this and other POS; (ii) characterising the syntactic and semantic behaviour of representative Portuguese adjective classes; (iii) defining the set of appropriate relations and role features for the representation of adjectives in linguistic knowledge databases; (iv) identifying the appropriate modelling structures in a framework of maximal generalisation and minimal complexity, able to account for complex linguistic phenomena. With regard to (i), in this dissertation we provide strong additional evidence supporting and strengthening the controversial thesis that adjectives constitute an independent category. The idea that adjectives are an “independent” lexical category is thus strengthened by the data, particularly through the contrastive description of the linguistic behaviour of adjectives, nouns and verbs. We determine an operative definition of what is an adjective, showing that there is a set of common features which generally hold for all adjectives: (a) a particular and precise function (adjectives are words that apply to words that denote entities to ascribe a property or a set of properties to them); (b) specific semantic properties (adjectives denote states and only exceptionally are transitive); and (c) a set of ‘typical’ features (independence from the object, ability to ascribe properties to objects and gradability). However, ascribing the status of lexical category to adjectives does not entail arguing that all adjectives display exactly the same characteristics. We deal with this fact very naturally, arguing that it is precisely the presence or absence of some of these features, in combination with other properties, that is on the basis of the definition of adjective classes. The research depicted in this dissertation is developed under the scope of Computational Lexical Semantics, within a feature structure framework: the Generative Lexicon (Pustejovsky, 1995). We also adopt the EuroWordNet framework (Vossen, 1998), for a representation of adjectives in linguistic knowledge databases, specifically relational models of the lexicon – WordNet.PT (Marrafa, 2001; 2002). Having established and characterised relevant adjective classes in languages like Portuguese, we delineate a linguistically motivated approach for modelling their syntactic and semantic properties. Combining syntactic and semantic criteria, we argue that the semantics of adjectives can be appropriately captured in wordnet-like lexica by means of the implementation of a small set of relations – particularly cross-POS relations. We test the adequacy of this set of relations by implementing it in WordNet.PT, a task which makes apparent that these relations preserve the coherence of the wordnet model, and have an important impact in increasing its adequacy. The strategies used for WordNet.PT adjective representation constitute a linguistically motivated approach for encoding adjectives in computational relational lexica in a principled and integrated way. Hence, the modelling strategies delineated are mostly concerned with adjective definitional properties. Realising that some of the syntactic contrasts involving adjectives remain unaccounted for, as well as the different ways the meaning of compound expressions is built, we argue that wordnets should include information on event and argument structures. Discussing relevant data in detail, we conclude for the need for fine-grained, rich and structured lexical representations, in order to enable a principled account of the different ways the meaning of compound expressions is built. In this context, we put forth a homogeneous and economic approach for representing all adjective classes in the lexicon, in the Generative Lexicon framework. We make apparent that putting a small set of economic generative mechanisms to work – unification, underspecification and information sharing between structures – allows us to account for complex linguistic phenomena such as relativity to a comparison class, selection restrictions, construction of meaning in context, and sense change. This way, we account for the most characteristic and general syntactic and semantic aspects of adjective behaviour. Given the important role played by adjective relative position in the NP with regard to the construction of NP meaning, we put forth an explicative principle for the distribution restrictions evinced by adjective classes. Along with many authors in the literature, we argue for the markedness of prenominal position, underlining its emphatic role and the contrast between prenominal and postnominal adjectives in terms of the relation they establish with the modified noun. We assume a dependence between nominal items with a set reference and prenominal adjectives, and we provide a unified analysis for all adjectives occurring in prenominal position, as well as a unified treatment of verbal and adjective alternations Moreover, we make our crucial assumptions apparent, by modelling full NPs in GL. Representing NPs with prenominal and postnominal adjectives enables us to straightforwardly make adjective semantic contribution self-evident, as well as the meaning contribution coming from the structure in which they occur – prenominal or postnominal position in the NP –, in a linguistically motivated way and without introducing any important changes to the adjective lexical entries proposed. Finally, we address specific phenomena related to event modification by adjectives. We analyse event modifying adjectives which show adverbial readings whose scope seems to go beyond the NP in which they occur, being extended over the whole sentence. Our approach underlines the role played by events associated to the lexical items involved in these structures, providing a unified treatment of adjectives and adverbs. In fact, modelling this semantic similarity makes apparent the number of semantic features shared by these POS. This is particularly relevant if we consider that these are the two POS that play the role of modifiers in language. Although event modifying adjectives make up a very specific group of contexts displaying apparently exceptional semantic behaviours, our analysis shows that understanding them better allows for more accurately delineating the fine line distinguishing different POS in language, and, particularly, various types of modifiers. Moreover, our approach makes the expressiveness of GL apparent: having thoroughly provided modelling strategies for members of all adjective classes, we are able to straightforwardly deal with apparently exceptional data, in a linguistically motivated way and without the need to introduce any changes in adjective lexical entries. Being so, this dissertation contributes to a better understanding of adjectives as a word class and to an accurate and economic modelling of this POS in the lexicon, in constant dialogue with other related POS and in an integrated approach to syntax and semantics. Also, we make apparent that the economic and underspecified modelling strategies we put forth are suitable for the computation of this POS in large scale lexica and grammars. In fact, as shown in this dissertation, the development of increasingly robust linguistic resources benefits from the incorporation in linguistic models of mechanisms such as unification, inheritance and recursivity, which allow for the representation and computation of the relation holding between form and meaning with economy of means and efficiency of results, also being more adapted to the caption of linguistic properties such as discreteness, compositionality, incrementality and productivity, among others.
URI: http://hdl.handle.net/11707/4457
Appears in Collections:Dissertations (restricted access)

Files in This Item:
File Description SizeFormat 
PortugueseAdjectives.pdf2.27 MBAdobe PDFView/Open    Request a copy


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.