Introduction

One of the fundamental problems of lexical semantics is the fact that what C. Ruhl (1989) calls the ‘perceived meaning’ of a word can vary so greatly from one context to another. In this chapter I want to survey the ways in which the contribution the same grammatical word makes to the meaning of a larger unit may differ in different contexts. There are two main sources of explanatory hypotheses for contextual variations in word meaning: lexical semantics and pragmatics. While there are probably no contexts where each of these is not involved in some way, their relative contributions can vary. For instance, in the following examples the difference between 1 and 2 in respect of the interpretation of the word teacher (i.e.,“male teacher” and “female teacher”, respectively) can be accounted for entirely by differential contextual enrichment of a single lexical meaning for teacher (in other words, pragmatically):

1. 1. The teacher stroked his beard.

2. 2. Our maths teacher is on maternity leave.

The only involvement of lexical semantics here is that the specification of the meaning of teacher must somehow make it clear that although it is unspecified for sex, it is, unlike, say, chair, specifiable for sex. Examples 3 and 4 exemplify a slightly different type of contextual enrichment, in that the extra specificity in context is of a meronymous rather than a hyponymous type:

1. 3. John washed the car.

2. 4. The mechanic lubricated the car.

The different actions performed on the car allow us to infer that the agents were occupied with different parts of the car in each case.

Introduction

This chapter builds on the title and theme of Apresjan's 1974 paper, Regular Polysemy. Apresjan was concerned merely to define the phenomenon and identify where it occurred. Here, we shall explore how it can be exploited.

Regular polysemy occurs where two or more words each have two senses, and all the words exhibit the same relationship between the two senses. The phenomenon is also called ‘sense extension’ (Copestake & Briscoe, 1991), ‘semantic transfer rules’ (Leech, 1981), ‘lexical implication rules’ (Ostler & Atkins, 1991), or simply ‘lexical rules’. An example, taken direct from a dictionary (Longman Dictionary of Contemporary English, hereafter LDOCE) is:

In each case, two senses are referred to, one with the ‘bracketed optional part’ included in the definition and the other with it omitted; the relation between the two is the same in both cases.

Recent work on lexical description has stressed the need for the structure of a lexical knowledge base (LKB) to reflect the structure of the lexicon (Atkins & Levin, 1991) and for the LKB to incorporate productive rules, so the rulebound ways in which words may be used are captured without the lexicon needing to list all options for all words (Boguraev & Levin, 1990). These arguments suggest that generalizations regarding regular polysemy should be expressed in the LKB, and that the formalism in which the LKB is written should be such that, once the generalization is stated, the specific cases follow as consequences of the inference rules of the formalism.

Introduction

We are interested in how a verb's description changes as it evolves into different usages. In order to examine this issue we are attempting to combine a LEXICAL ANALYSIS of a particular verb with a CONCEPTUAL ANALYSIS. LEXICAL ANALYSIS is aimed at producing a LEXICAL DEFINITION which functions as a linguistic paraphrase. This should ideally be a lexical decomposition composed of terms with simpler, “more primitive” meanings than the term being defined (Mel'čuk 1988; Wierzbicka 1984). In contrast, a conceptual analysis produces a CONCEPTUAL DESCRIPTION of the lexical item.

Criteria for building a conceptual description are linked to arguments borrowed from “real-world” considerations or expressed in terms of denotations.

Introduction

A major motivation for the introduction of default inheritance mechanisms into theories of lexical organization has been to account for the prevalence of the family of phenomena variously described as blocking (Aronoff, 1976:43), the elsewhere condition (Kiparsky, 1973), or preemption by synonymy (Clark & Clark, 1979:798). In Copestake and Briscoe (1991) we argued that productive processes of sense extension also undergo the same process, suggesting that an integrated account of lexical semantic and morphological processes must allow for blocking. In this chapter, we review extant accounts which follow from theories of lexical organization based on default inheritance, such as Paradigmatic Morphology (Calder, 1989), DATR (Evans & Gazdar, 1989), ELU (Russell et al., 1991, in press), Word Grammar (Hudson, 1990; Fraser & Hudson, 1992), or the LKB (Copestake 1992; this volume; Copestake et al., in press). We argue that these theories fail to capture the full complexity of even the simplest cases of blocking and sketch a more adequate framework, based on a non-monotonic logic that incorporates more powerful mechanisms for resolving conflict among defeasible knowledge resources (Common-sense Entailment, Asher & Morreau, 1991). Finally, we explore the similarities and differences between various phenomena which have been intuitively felt to be cases of blocking within this formal framework, and discuss the manner in which such processes might interact with more general interpretative strategies during language comprehension. Our presentation is necessarily brief and rather informal; we are primarily concerned to point out the potential advantages using a more expressive default logic for remedying some of the inadequacies of current theories of lexical description.

Introduction

The work reported here is part of research on the ACQUILEX project which is aimed at the eventual development of a theoretically motivated, but comprehensive and computationally tractable, multilingual lexical knowledge base (LKB) usable for natural language processing, lexicography and other applications. One of the goals of the ACQUILEX project was to demonstrate the feasibility of building an LKB by acquiring a substantial portion of the information semi-automatically from machine readable dictionaries (MRDs). We have paid particular attention to lexical semantic information. Our work therefore attempts to integrate several strands of research:

1. • Linguistic theories of the lexicon and lexical semantics. In this chapter we will concentrate on the lexical semantics of nominals where our treatment is broadly based on that of Pustejovsky (1991), and in particular on his concepts of the generative lexicon and of qualia structure.

2. • Knowledge representation techniques. The formal lexical representation language (LRL) used in the ACQUILEX LKB system is based on typed features structures similar to those of Carpenter (1990, 1992), augmented with default inheritance and lexical rules. Our lexicons can thus be highly structured, hierarchical and generative.

3. • Lexicography and computational lexicography. The work reported here makes extensive use of the Longman Dictionary of Contemporary English (LDOCE; Procter, 1978). MRDs do not just provide data about individual lexical items; our theories of the lexicon have been developed and refined by considering the implicit organization of dictionaries and the insights of lexicographers.

In this chapter we will show how these strands can be combined in developing an appropriate representation for group nouns in the LRL, and in extracting the requisite information automatically from MRDs.

Lexical semantics is still in a rather early stage of development. This explains the reason why there are relatively few elaborated systems for representing its various aspects. A number of semantic aspects can be straightforwardly represented by means of feature-value pairs and by means of typed feature structures. Others, such as the Qualia Structure or the lexical semantics relations are more difficult to represent. The difficulty is twofold: (1) there is first the need to define appropriate models to represent the various levels of semantic information, including their associated possible inference systems and their properties (e.g., transitivity, monotonicity, etc.) and (2) there is the need to develop complex algorithms that allow for as efficient as possible treatments from these models. The first chapter of this section tackles the first point and the second one addresses some algorithmic problems.

“Introducing Lexlog” by J. Jayez, is a set of specifications for constructing explicit representations for lexical objects in restricted domains. Lexlog offers two types of functions: control functions to formalize representations and updatings on these representations in a controlled way, and expression functions to express different semantic operators and to tailor these operators with syntactic operators, for example, the trees of the Tree Adjoining Grammar framework. The discussion ends with a detailed presentation of the implementation in Prolog.

The last chapter, “Constraint propagation techniques for lexical semantics descriptions,” by Patrick Saint-Dizier, addresses the problem of the propagation in parse trees of large feature structures. The motivation is basically to avoid computations of intermediate results which later turn out to be useless.

Another current major issue in lexical semantics is the definition and the construction of real-size lexical databases that will be used by parsers and generators in conjunction with a grammatical system. Word meaning, terminological knowledge representation and extraction of knowledge in machine readable dictionaries are the main topics addressed. They really represent the backbone of a lexical semantics knowledge base construction.

The first chapter, “Lexical semantics and terminological knowledge representation” by Gerrit Burkert, shows the practical and formal inadequacies of semantic networks for representing knowledge, and the advantages of using a term subsumption language. In a first stage, this document shows how several aspects of word meaning can be adequately described using a term subsumption language. Then, some extensions are proposed that make the system more suitable for lexical semantics. Formal aspects are strongly motivated by several examples borrowed from an in-depth study of terminological knowledge extraction, which is a rather challenging area for lexical semantics.

“Word meaning between lexical and conceptual structure”, by Peter Gerstl, presents a method and a system to introduce world-knowledge or domain-dependent knowledge in a lexicon. The meaning of a word is derived from general lexical information on the one hand and from ontological knowledge on the other hand. The notion of semantic scope is explored on an empirical basis by analyzing in a systematic way the influences involved in natural language expressions. This component has been integrated into the Lilog system developed at IBM Stuttgart.

Introduction

One of the most difficult areas for research in machine translation (MT) is the representation of meanings in the lexicon. The lexicon plays a central role in any MT system, regardless of the theoretical foundations upon which the system is based. However, it is only recently that MT researchers have begun to focus more specifically on issues that concern the lexicon, e.g., cross-linguistic variations that arise during the mapping between lexical items in the source and target languages.

The traditional approach to constructing dictionaries for MT has been to massage on-line dictionaries that are primarily intended for human consumption. Given that most natural language applications have focused primarily on syntactic information that can be extracted from the lexicon, these methods have constituted a reasonable first-pass approach to the problem. However, it is now widely accepted that MT requires language-independent conceptual information in order to successfully process a wide range of phenomena in more than one language. Thus, the task of constructing lexical entries has become a much more difficult problem as researchers endeavor to extend the concept base to support more phenomena and additional languages.

This chapter describes how parameterization of the lexicon allows an MT system to account for a number of cross-linguistic variations, called divergences, during translation. There are many cases in which the natural translation of one language into another results in a very different form than that of the original. These divergences make the straightforward transfer from source structures into target structures impractical.

This volume on computational lexical semantics emerged from a workshop on lexical semantics issues organized in Toulouse, France, in January 1992. The chapters presented here are extended versions of the original texts.

Lexical semantics is now becoming a major research area in computational linguistics and it is playing more of a central role in various types of applications involving natural language parsers as well as generators.

Lexical semantics covers a wide spectrum of problematics from different disciplines, from psycholinguistics to knowledge representation and to computer architecture, which makes this field relatively difficult to perceive as a whole. The goal of this volume is to present the state of the art in lexical semantics from a computational linguistics point of view and from a range of perspectives: psycholinguistics, linguistics (formal and applied), computational linguistics, and application development. The following points are particularly developed in this volume:

• psycholinguistics: mental lexicons, access to lexical items, form of lexical items, links between concepts and words, and lexicalizing operations;

• linguistics and formal aspects of lexical semantics: lexical semantics relations, prototypes, conceptual representations, event structure, argument structure, and lexical redundancy;

• knowledge representation: systems of rules, treatment of type coercion, aspects of inheritance, and relations between linguistics and world knowledge;

• applications: creation and maintenance of large-size lexicons, the role of the lexicon in parsing and generation, lexical knowledge bases, and acquisition of lexical data;

• operational aspects: processing models and architecture of lexical systems.

Lexical semantics offers a large variety of uses in natural language processing and it obviously allows for more refined treatments. One of the main problems is to identify exactly the lexical semantic resources that one needs to solve a particular problem. Another main difficulty is to know how best to organize this knowledge in order to keep the system reasonably efficient and maintainable; this is particularly crucial for a number of large-scale applications. This volume contains two chapters that explore application of lexical semantics in the area of natural language generation and in the area of machine translation with an interlingua representation.

The first chapter of this section, “Lexical functions of the Explanatory Combinatorial Dictionary for lexicalization in text generation”, by Margarita Alonso Ramos et al., applies Mel'čuk's framework to natural language generation. It shows that the problem of lexicalization, i.e., the relation between a concept (or a combination of concepts) and its linguistic realization, cannot really be correctly carried out without making reference to a lexicon that takes into account the diversity of the lexico-semantics relations. This approach views lexicalization both as a local process (lexicalization is solved within a restricted phrase) and a more global one, taking into account the ‘contextual effects’ of a certain lexicalization with respect to the others in a sentence or in a text. Paradigmatic lexical functions are shown to be well adapted to treat lexicalization in the context of a text, whereas syntagmatic ones operate at the sentence or proposition levels.

Language comprehension and interpretation can be set within a general cognitive science perspective that encompasses both human minds and intelligent machine systems. This comprehensive view, however, must necessarily entail the search for compatibility between various types of concepts or models subsuming classes of facts obtained through specific methods, and hence belonging to various scientific domains, from artificial intelligence to cognitive psychology, and ranging over linguistics, logics, neurosciences, and others.

No sub-domain of cognitive science is more fascinating for the exploration of such functional identities between artificial and natural processing or representation and none deserves more to be comparatively worked out than language comprehension or interpretation. The purpose of this chapter is to highlight some of these identities, but also some differences, in particular as concerns lexical semantics.

A large number of concepts are obviously shared by computational semantics and cognitive psychology. I will approach them in this chapter mainly in the form of relational properties, belonging to lexical-semantic, or lexical-conceptual, units, and will classify them according to whether they can be attributed, or not, to both mental and machine representations. For the sake of simplicity I will often restrict this comparison to consideration of lexical units that are expressed by nouns in most natural languages and which, as a rule, denote classes of objects or individuals. But other kinds of lexical-conceptual units, expressed in these natural languages by verbs, adjectivals, prepositions, function words, etc., and which denote events or actions, properties, various sorts of relations, etc., could also be submitted to this type of analysis and comparison.

The issue of lexical content for NLU

The basic elements of most recent Natural Language Understanding (NLU) systems are a syntactic parser which is used to determine sentence structure, and a semantic lexicon whose purpose is to access the system's factual knowledge from the natural language input. In this regard, the semantic lexicon plays the key role of relating words to world knowledge. But the semantic lexicon is also used in solving some specifically linguistic issues when recovering sentence structure, and should contain linguistic knowledge. In this chapter we discuss the issue of lexical content in terms of linguistic and world knowledge, through the so-called dictionary–encyclopedia controversy. To illustrate the discussion we will describe the lexical semantics approach adopted in our NLU program processing sentences from medical records (Zweigenbaum and Cavazza, 1990). This program is a small-scale but fully implemented prototype adopting a broad view to NLU, from syntactic analysis to complex domain inferences through model-based reasoning (Grishman and Ksiezyk, 1990). The dictionary–encyclopedia controversy opposes two extreme conceptions of word definitions: according to the dictionary approach a word is described in terms of linguistic elements only, without recourse to world knowledge, whereas an encyclopedic definition includes

This point has been discussed by many authors including Katz and Fodor (1963), Eco (1984), Wierzbicka (1985) and Taylor (1989).

Introduction

Recent works in Computational Linguistics show the central role played by the lexicon in language processing, and in particular by the lexical semantics component. Lexicons tend no longer to be a mere enumeration of feature-value pairs but tend to have an intelligent behavior. This is the case, for example, for generative lexicons (Pustejovsky, 1991) which contain, besides feature structures, a number of rules to create new (partial) definitions of word-senses such as rules for conflation and type coercion. As a result, the size of lexical entries describing word-senses has substantially increased. These lexical entries become very hard to be used directly by a natural language parser or generator because their size and complexity allow a priori little flexibility.

Most natural language systems consider a lexical entry as an indivisible whole which is percolated up in the parse/generation tree. Access to features and feature values at grammar rule level is realized by more or less complex procedures (Shieber, 1986; Johnson, 1990; Günthner, 1988). The complexity of real natural language processing systems makes such an approach very inefficient and not necessarily linguistically adequate. In this document, we propose a dynamic treatment of features in grammars, embedded within a Constraint Logic Programming framework (noted as CLP hereafter) which permits us to access a feature-value pair associated to a certain word-sense directly into the lexicon, and only when this feature is explicitly required by the grammar, for example, to make a check. More precisely, this dynamic treatment of features will make use of both constraint propagation techniques embedded within CLP and CLP resolution mechanisms.

Lexical semantics knowledge often requires various kinds of inferences to be made in order to make more precise and more explicit the meaning of a word with respect to the context in which it is uttered. The treatment of inheritance is particularly crucial from this point of view since linguistic as well as conceptual hierarchies allow us to relate a word with more generic terms that contribute, in a certain sense, to the definition of the meaning of that word. Because hierarchies are complex, inheritance often involves non-monotonic forms of reasoning. Another relation with artificial intelligence is the development of models and techniques for the automatic acquisition of linguistic knowledge.

In “Blocking”, Ted Briscoe et al. elaborate on the introduction of default inheritance mechanisms into theories of lexical organization. Blocking is a kind of inference rule that realizes a choice when two kinds of inheritances can be drawn from resources occurring at different levels in the lexical organization. For example, a general morphological rule adds -ed at the end of verbs for the past construction. A subclass of verbs consists of irregular verbs, which receive a particular mark for the past. A default rule states that the exceptional mark is preferred to the general case. However there exist verbs such as dream, which accept both flections (dreamt, dreamed). Then the default rule is no longer adequate and a more flexible solution is required. In this chapter, a framework based on a non-monotonic logic that incorporates more powerful mechanisms for resolving conflicts of various kinds is introduced.

Introduction

Some approaches to semantics

Various theories are nowadays used in linguistic analysis, particularly in syntax, and it does not seem reasonable to expect a reduction of their number in the near future. Nor does it seem reasonable to expect that one theory will cover them, as a kind of meta-theory. Nevertheless, all these theories have in common the need for a lexicon which would include the necessary and sufficient information for combining lexical items and extracting a representation of the meaning of such combinations.

If it is not possible to propose a canonical theory to organize the storage of the lexical information, it is necessary to adopt a “polytheoretic” conception of the lexicon. (Hellwig, Minkwitz, and Koch, 1991). A point shared by different theories concerns the need for some semantic information even for a syntactic parsing of a sentence (and a fortiori, beyond this level, for the parsing of a text).

During a first period, the semantic information required was above all concentrated on the thematic roles (Gruber, 1965; Jackendoff, 1972). These thematic roles were introduced because the grammatical functions were insufficient for discriminating between various interpretations and for describing the similarities of sentences. For instance, in:

1. The door opened

2. Charlie opened the door

the door is considered as having the same semantic function but not the same grammatical function – subject for (1) and object for (2). Gruber (1965) gives to the door in the two sentences the same thematic role of theme. A similar objective was assigned to the case-theory (Fillmore, 1968).

Introduction

LexLog is not a lexical theory, but a package of logical specifications for constructing explicit representations for lexical items in restricted information domains. We usually call a domain “restricted” in two main cases.

1. The domain is notionally bounded. It is sufficiently simple to be decomposed into an organized finite set of notions, which might be enriched, if necessary, by applying a finite number of explicit combination rules. Metaphorically, one could say that the domain has been “axiomatized”. A clear example of this situation can be found in Palmer (1990).

2. The domain is lexically bounded. The representations to be constructed apply to a finite set of lexical items, and not to the whole of a language, or to an indefinite set of terms.

While they are often associated, neither of these properties strictly implies the other. For example, although the use of a statistical tool like PATHFINDER (Schvaneveldt, 1990) allows the construction of limited lexical clusters (case 2), their notional homogeneity is not warranted by the clustering procedure. E.g., in a study of the word marriage based on the definitions in the Longman Dictionary of Contemporary English (Wilks et al., 1989), the words marriage and muslim co-occur. Although the religious link is obvious, making precise the amount of knowledge about muslim which would be relevant to marriage is not easy. LexLog is designed only for situations which exhibit the TWO forms of boundedness, that is, it can be of some help only when the domain allows the use of a finite notional and a finite lexical basis.

Introduction

Knowledge representation and reasoning are central to all fields of Artificial Intelligence research. It includes the development of formalisms for the representation of given subject matters as well as the development of inference procedures to reason about the represented knowledge. Before developing a knowledge representation formalism, one must determine what type of knowledge has to be modeled with the formalism. Since a lot of our knowledge of the world can easily be described using natural language, it is an interesting task to examine to what extent the contents of natural language utterances can be formalized and represented with a given representation formalism. Every approach to represent natural language utterances must include a method to formalize aspects of the meaning of single lexical units.

An early attempt in this direction was Quillian's Semantic Memory (Quillian, 1968), an associational model of human memory. A semantic memory consists of nodes corresponding to English words and different associative links connecting the nodes. Based on that approach, various knowledge representation systems have been developed which can be subsumed under the term semantic network. Common to all these systems is that knowledge is represented by a network of nodes and links. The nodes usually represent concepts or meanings whereas the links represent relations between concepts. In most semantic network formalisms, a special kind of link between more specific and more general concepts exists. This link, often called IS-A or AKO (a kind of), organizes the concepts into a hierarchy in which information can be inherited from more general to more specific concepts.