As for any other phenomenon in the world, the existence of paradigmatic semantic relations among words calls for some kind of explanation – or perhaps several kinds of explanation. Are these relations among words, or among the things the words represent? Are the relations arbitrary or rule based? Language specific or universal? A product of linguistic or general cognition? These questions are the focus of this book. First, however, we must ask what these questions mean, and why we might care to trouble ourselves with them.

As linguistic theories have progressed in modeling human language ability, the lexicon has become more central to those theories. With this new or renewed attention to the mental lexicon, two problems become evident. Firstly, there is no generally accepted theory of how the lexicon is internally structured and how lexical information is represented in it.

The previous chapters have reviewed modern thinking on paradigmatic semantic relations and proposed that such relations are metalinguistic rather than linguistic in nature. Whereas many theorists have assumed that the lexicon is organized according to relations like synonymy and antonymy, the evidence is against semantic organization of the lexicon, since meaning itself is not wholly resident in the mental linguistic faculty. This chapter returns to the assumptions put forth in chapter 1. Three questions are raised and briefly discussed: Is semantic organization of the lexicon necessary? What is a plausible model of the lexicon? And finally, if paradigmatic relations are metalinguistic in nature, what business do linguists have in studying them?

Is semantic organization of the lexicon necessary?

Throughout this book, I have argued that paradigmatic semantic relations are derived and stored metalexically, rather than intralexically. This argument has rested on the assumptions that linguistic knowledge and processes belong to a mental faculty that is separate from general cognition and that the lexicon should contain all and only idiosyncratic information that contributes to the construction of well-formed expressions in a particular language.

Similarity of meaning is “the most important lexical relation” in the WordNet model (Miller and Fellbaum 1991: 202), and, in philosophy, Quine (1961: 22) has identified synonymy (along with analyticity) as “the primary business of the theory of meaning.” This relation raises issues of similarity that remain relevant in the following chapters. In addition, it presents an interesting challenge for the metalexical approach: Can synonymy, a relation among words with similar meanings, be defined in terms of contrast?

Section 4.1 starts the chapter with a description of synonymy in the metalexical approach. This involves a restatement of Relation by Contrast in order to apply it specifically to word sets that differ in form but are otherwise similar enough to be synonyms. Here the ideas presented in chapter 2 are further developed with reference to the notions of ‘similarity’ and ‘difference.’ The next two sections cover the key concepts in synonym studies, so that they can be reinterpreted from the metalexical perspective. Subtypes and degrees of synonymy are examined in 4.2. While some approaches attempt to reduce the number of relations that can count as synonymy, the approach taken here is inclusive. Section 4.3 reviews alleged (logical or traditional) properties of synonymy, such as transitivity and interchangeability, and some evidence against them.

The relations in this chapter are allotted fewer pages than synonymy and antonymy because they have only a tenuous claim to the label lexical relation. While inclusion and part-whole relations in particular are acknowledged in lexical semantics texts, they are rarely relations among words and almost always relations among concepts or things. Since approaches to meaning and semantic relations differ in the amount and nature of semantic content represented in the lexicon, relations like hyponymy and meronymy have varying levels of relevance to different theorists. The assumption here, however, is that the lexicon includes no paradigmatic relational information. In spite of this, synonymy and antonymy are lexicologically interesting as metalexical relations – relations among word-concepts. There is little evidence, however, that hyponymy, meronymy, or other semantic relations are relations among word-concepts rather than relations among the things (or concepts) that those words denote. This chapter, then, concerns why these relations are sometimes treated as (intra)lexical relations and demonstrates that they usually are not relations among words. Relation by Contrast (RC) is still relevant to these relations, since it can contrast any concepts, not just lexical concepts, but this book is concerned with its lexical-semantic applications, and the relations in this chapter are less clearly lexical-semantic relations.

Following the assumptions of chapter 1, only information that is both arbitrary and relevant to linguistic competence is contained in lexical entries. If relations among words are not arbitrary or not entirely dependent on the linguistic properties of the words, then these relations are not represented in the lexicon. This chapter demonstrates that semantic relations among words indeed are not (and cannot be) represented in the lexicon for three reasons: (a) They are not relevant to linguistic competence; (b) they depend upon the contexts in which they occur; and (c) they are predictable by means of a single relational principle. Semantic relations among words thus consist of conceptual knowledge about words, rather than lexical knowledge of words (Murphy 2000). In essence, this chapter shows that a metalexical treatment of semantic relations is forced by the divisions of labor among lexicon, grammar, and conceptual knowledge, as well as by facts about relational phenomena.

This metalexical treatment provides the means to account for all lexical relations – not just semantic ones and not just the ones that have been acknowledged with ᾽nym names. So while Cruse (1986) goes to some pains to distinguish between “interesting” and “uninteresting” semantic relations, this treatment considers any paradigm-based relation interesting and explainable, no matter the basis of the paradigm.

Antonymy (referring to binary opposition generally) is arguably the archetypical lexical semantic relation. Unlike synonymy, everyone agrees that antonymy exists, and it is robustly evident in natural language. Unlike hyponymy and meronymy, it can be as much a relation among words as it is a relation among concepts or denotata. Being such a good example of a lexical semantic relation, antonymy is also the focus of much of the debate as to whether semantic relations are stored or derived and whether they are intralexically or metalexically represented. This chapter is focused on that debate and on why contrast involving semantic incompatibility is so central a semantic relation. Section 5.1 presents the metalexical approach to antonymy and incompatible contrast sets, in which antonyms may be metalexically stored, but only count as antonyms if they are also derivable by Relation by Contrast. In 5.2 properties of antonymy that have been proposed in the linguistic literature, including symmetry, markedness, and binarity, are reviewed, then 5.3 covers antonym subtypes and how RC accounts for them. Section 5.4 concerns the roles of antonymy and non-binary semantic contrast in natural language in discourse, semantic or lexical change, language acquisition, and across cultures. Section 5.5 provides a summary.

In the last chapter, I proposed that paradigmatic relations among words relate conceptual representations of words, rather than linguistic representations in a modular lexicon. That approach is founded upon the assumptions that (a) relations among words can be studied as cognitive phenomena, (b) relations must be interpreted with respect to their linguistic use, (c) nonlinguistic context is relevant to these relations, and (d) definitional and encyclopedic aspects of meaning cannot be neatly separated. This chapter surveys other approaches to paradigmatic semantic relations, starting (in 3.1) with a historical survey of the role of these relations in five disciplines: philosophy, anthropology, linguistics, psychology, and computer science. The approaches to semantic relations in these disciplines frequently overlap, and so three (cross-disciplinary) categories of approaches are critically discussed in the following sections. Section 3.2 concerns those approaches that treat lexical meaning as composed of features or primitives. In these theories, semantic relations arise from the similarities and differences among words᾽ internal semantic structures. In 3.4 we look at approaches in which word meanings are not defined. In these cases, semantic relations must be explicitly stated in the lexicon or semantic memory. Between these two extremes, the approaches in 3.3 have it both ways: defining words in the lexicon and explicitly representing semantic relations among words.