1.1 Polysemy and vagueness

Establishing which meanings a word has is arguably the basic step in lexical semantic research. Polysemy is the common term for the situation in which a lexical item has more than one meaning, such as when late can mean ‘after the usual, expected, or agreed time’ (I am late again), ‘advanced in day or night’ (a late dinner), or ‘no longer alive’ (my late aunt Polly). Terminologically speaking, polysemy needs to be contrasted with homonymy and, more importantly, vagueness. When two (or more) words have the same shape, such as bank (‘slope, elevation in sea or river bed’) and bank (‘financial institution’), they are homonyms; whereas polysemy refers to multiplicity of meaning within a single word, the multiplicity is distributed over various words in the case of homonymy. As such, making a distinction between polysemy and homonymy comes down to determining whether we are dealing with one and the same word or with two different ones. The distinction between vagueness and polysemy involves the question of whether a particular piece of semantic information is part of the underlying semantic structure of the item or is the result of a contextual (and hence pragmatic) specification. For instance, neighbor is not polysemous between the readings ‘male dweller next door’ and ‘female dweller next door,’ in the sense that the utterance my neighbor is a civil servant will not be recognized as requiring disambiguation in the way that she is smart might (Do you mean ‘bright’ or ‘stylish’?). The semantic information that is associated with the item neighbor in the lexicon does not, in other words, contain a specification regarding sex; neighbor is vague (or general, or unspecified) as to the dimension of gender.

To decide between polysemy and vagueness, a number of tests can be invoked. The three main ones are the following. First, from a truth-theoretical point of view, a lexical item is polysemous if it can simultaneously be clearly true and clearly false of the same referent. Considering the readings ‘harbor’ and ‘fortified sweet wine from Portugal’ of port, the polysemy of that item is established by sentences such as Sandeman is a port (in a bottle), but not a port (with ships). This criterion basically captures a semantic intuition: are two interpretations of a given expression intuitively sufficiently dissimilar so that one may be said to apply and the other not?

Second, linguistic tests involve syntactic rather than semantic intuitions. Specifically, they are based on acceptability judgments about sentences that contain two related occurrences of the item under consideration (one of which may be implicit). If the grammatical relationship between both occurrences requires their semantic identity, the resulting sentence may be an indication for the polysemy of the item. For instance, the so-called identity test involves ‘identity-of-sense anaphora.’ Thus, at midnight the ship passed the port, and so did the bartender is awkward if the two lexical meanings of port are at stake. Disregarding puns, it can only mean that the ship and the bartender alike passed the harbor, or conversely that both moved a particular kind of wine from one place to another. A mixed reading, in which the first occurrence of port refers to the harbor and the second to wine, is normally excluded. By contrast, the fact that the notions ‘vintage sweet wine from Portugal’ and ‘blended sweet wine from Portugal’ can be combined in Vintage Noval is a port, and so is blended Sandeman indicates that port is vague rather than polysemous with regard to the distinction between blended and vintage wines.

Third, the definitional criterion specifies that an item has more than one lexical meaning if there is no minimally specific definition covering the extension of the item as a whole, and that it has no more lexical meanings than there are maximally general definitions necessary to describe its extension. Definitions of lexical items should be maximally general in the sense that they should cover as large a subset of the extension of an item as possible. Thus, separate definitions for ‘blended sweet fortified wine from Portugal’ and ‘vintage sweet fortified wine from Portugal’ could not be considered definitions of lexical meanings, because they can be brought together under the definition ‘sweet fortified wine from Portugal.’ On the other hand, definitions should be minimally specific in the sense that they should be sufficient to distinguish the item from other nonsynonymous items. A maximally general definition covering both port ‘harbor’ and port ‘kind of wine’ under the definition ‘thing, entity’ is excluded because it does not capture the specificity of port as distinct from other words.

The distinction between polysemy and vagueness is not unproblematic, methodologically speaking. An examination of different basic criteria for distinguishing between polysemy and vagueness reveals, first, that those criteria may be in mutual conflict (in the sense that they need not lead to the same conclusion in the same circumstances) and, second, that each of them taken separately need not lead to a stable distinction between polysemy and vagueness (in the sense that what is a distinct meaning according to one of the tests in one context may be reduced to a case of vagueness according to the same test in another context). Without going into detail (for a full treatment, see Geeraerts, 1993), let us illustrate the first type of problem. In the case of autohyponymous words, for instance, the definitional approach does not reveal an ambiguity, whereas the truth-theoretical criterion does. Dog is autohyponymous between the readings ‘Canis familiaris,’ contrasting with cat or wolf, and ‘male Canis familiaris,’ contrasting with bitch. A definition of dog as ‘male Canis familiaris,’ however, does not conform to the definitional criterion of maximal coverage, because it defines a proper subset of the ‘Canis familiaris’ reading. On the other hand, the sentence Lady is a dog, but not a dog, which exemplifies the logical criterion, cannot be ruled out as ungrammatical.

1.2 Semantic Relations

Once senses are identified (and assuming they can be identified with a reasonable degree of confidence), the type of relationship that exists between them needs to be established. The most common classification of semantic relations emerges from the tradition of historical semantics, that is, the vocabulary used to describe synchronic relations between word meanings is essentially the same as the vocabulary used to describe diachronic changes of meaning. In the simplest case, if sense a is synchronically related to sense b by metonymy, then a process of metonymy has acted diachronically to extend sense a to sense b: diachronic mechanisms of semasiological change reappear synchronically as semantic relations among word meanings.

The four basic types are specialization, generalization, metaphor, and metonymy (described here, from a diachronic perspective, as mechanisms rather than synchronic relations). In the case of semantic specialization, the new meaning is a restriction of the old meaning: the new meaning is a subcase of the old. In the case of semantic generalization, the reverse holds: the old meaning is a subcase of the new. Classical examples of specialization are corn (originally a cover-term for all kinds of grain, now specialized to ‘wheat’ in England, to ‘oats’ in Scotland, and to ‘maize’ in the United States), starve (moving from ‘to die’ to ‘to die of hunger’), and queen (originally ‘wife, woman,’ now restricted to ‘king’s wife, or female sovereign’). Examples of generalization are moon (primarily the earth’s satellite, but extended to any planet’s satellite), and French arriver (which originally meant ‘to reach the river’s shore, to embank,’ but which now signifies ‘to reach a destination’ in general). There is a lot of terminological variation in connection with specialization and generalization. ‘Restriction’ and ‘narrowing’ of meaning equal ‘specialization,’ while ‘extension,’ ‘schematization,’ and ‘broadening’ of meaning equal ‘generalization.’ Also, the meanings involved can be said to entertain relations of taxonomical subordination or superordination: in a taxonomy (a tree-like hierarchical classification) of concepts, the specialized meaning is subordinate with regard to the original one, whereas the generalized meaning is superordinate with regard to the original.

Like specialization and generalization, it is convenient and customary to introduce metaphor and metonymy together, even though the relationship is not as close as with the former pair. (More on metaphor and metonymy follows in section 1.6, “Conceptual Metaphor and Metonymy.”) Metaphor is then said to be based on a relationship of similarity between the old and the new reading, and metonymy on a relationship of contiguity. Current computer terminology yields examples of both types. The desktop of your computer screen, for instance, is not the same as the desktop of your office desk—except that in both cases, it is the space (a literal space in one case, a virtual one in the other) where you position a number of items that you regularly use or that urgently need attention. The computer desktop, in other words, is not literally a desktop in the original sense, but it has a functional similarity with the original: the computer reading is a metaphorical extension of the original office furniture reading. Functional similarities also underlie metaphorical expressions like bookmark, clipboard, file, folder, cut, and paste. Mouse, on the other hand, is also metaphorically motivated, but here, the metaphorical similarity involves shape rather than function. But now consider a statement to the effect that your desktop will keep you busy for the next two weeks, or that you ask aloud where your mouse has gone when you are trying to locate the pointer on the screen. In such cases, desktop and mouse are used metonymically. In the former case, it’s not the virtual space as such that is relevant, but the items that are stored there. In the latter case, it’s not the mouse as such (the thing that you hold in your hand) that you refer to, but the pointer on the screen that is operated by the mouse. The desktop and the stored items, or the mouse and the pointer, have a relationship of real-world connectedness that is usually captured by the notion of ‘contiguity.’ When, for instance, one drinks a whole bottle, it is not the bottle but merely its contents that are consumed: bottle can be used to refer to a certain type of container, and the (spatially contiguous) contents of that container. When lexical semanticians state that metonymical changes are based on contiguity, contiguity should not be understood in a narrow sense as referring to spatial proximity only, but more broadly as a general term for various associations in the spatial, temporal, or causal domain.

1.3 Lexical Fields and Componential Analysis

A lexical field is a set of semantically related lexical items whose meanings are mutually interdependent. The single most influential study in the history of lexical field theory is Trier’s (1931) monograph, in which he presents a theoretical formulation of the field approach and investigates how the terminology for mental properties evolves from Old High German up to the beginning of the 13th century. Theoretically, Trier emphasizes that only a mutual demarcation of the words under consideration can provide a decisive answer regarding their exact value. Words should not be considered in isolation, but in their relationship to semantically related words: demarcation is always a demarcation relative to other words.

While different conceptions of the notion ‘lexical field’ were suggested after Trier’s initial formulation, the most important development is the emergence of componential analysis as a technique for formalizing the semantic relationships between the items in a field: once a lexical field has been demarcated, the internal relations within the field will have to be described in more detail. It is not sufficient to say that the items in the field are in mutual opposition—these oppositions will have to be identified and defined. Componential analysis is a method for describing such oppositions that takes its inspiration from structuralist phonology: just like phonemes are described structurally by their position on a set of contrastive dimensions, words may be characterized on the basis of the dimensions that structure a lexical field. Componential analysis provides a descriptive model for semantic content, based on the assumption that meanings can be described on the basis of a restricted set of conceptual building blocks—the semantic ‘components’ or ‘features.’

A brief illustration of the principles of componential analysis is given by Pottier (1964), who provides an example of a componential semantic analysis in his description of a field consisting of, among others, the terms siège, pouf, tabouret, chaise, fauteuil, and canapé (a subfield of the field of furniture terms in French). The word which acts as a superordinate to the field under consideration is siège, ‘seating equipment with legs.’ If we use the dimensions s1 ‘for seating,’ s2 ‘for one person,’ s3 ‘with legs,’ s4 ‘with back,’ s5 ‘with armrests,’ s6 ‘of rigid material,’ then chaise ‘chair’ can be componentially defined as [+ s1, + s2, + s3, + s4, − s5, + s6], and canapé ‘sofa’ as [+ s1, − s2, + s3, + s4, + s5, + s6], and so on.

While componential forms of description are common in formal types of semantic description (see the historical overview in section 2, “The Theoretical Evolution of Lexical Semantics,” specifically section 2.3, “Neostructuralist Semantics”), the most important theoretical development after the introduction of componential analysis is probably Wierzbicka’s (1996) attempt to identify a restricted set of some 60 universally valid, innate components. The Natural Semantic Metalanguage aims at defining cross-linguistically transparent definitions by means of those allegedly universal building-blocks.

1.4 Lexical Relations

Like componential analysis, relational semantics, as introduced by Lyons (1963), develops the idea of describing the structural relations among related words. It, however, restricts the theoretical vocabulary to be used in such a description. In a componential analysis, the features are essentially of a ‘real world’ kind: as in Pottier’s example, they name properties of the things referred to, rather than properties of the meanings as such. But if linguistics is interested in the structure of the language rather than the structure of the world, it may want to use a descriptive apparatus that is more purely linguistic. Relational semantics looks for such an apparatus in the form of sense relations like synonymy (identity of meaning) and antonymy (oppositeness of meaning): the fact that aunt and uncle refer to the same genealogical generation is a fact about the world, but the fact that black and white are opposites is a fact about words and language. Instead of deriving statements about the synonymy or antonymy of a word (and in general, statements about the meaning relations it entertains) from a separate and independent description of the word’s meaning, the meaning of the word could be defined as the total set of meaning relations in which it participates. A traditional approach to synonymy would for instance describe the meaning of both quickly and speedily as ‘in a fast way, not taking up much time,’ and then conclude to the synonymy of both terms on the basis of their definitional identity. Lyons by contrast deliberately eschews such content descriptions, and equates the meaning of a word like quickly with the synonymy relation it has with speedily, plus any other relations of that kind.

In the actual practice of relational semantics, ‘relations of that kind’ specifically include—next to synonymy and antonymy—relations of hyponymy (or subordination) and hyperonymy (or superordination), which are both based on taxonomical inclusion. The major research line in relational semantics involves the refinement and extension of this initial set of relations. The most prominent contribution to this endeavor after Lyons is found in Cruse (1986). Murphy (2003) is a thoroughly documented critical overview of the relational research tradition.

1.5 Distributional Relations

Given a Saussurean distinction between paradigmatic and syntagmatic relations, lexical fields as originally conceived are based on paradigmatic relations of similarity. One extension of the field approach, then, consists of taking a syntagmatic point of view. Words may in fact have specific combinatorial features which it would be natural to include in a field analysis. A verb like to comb, for instance, selects direct objects that refer to hair, or hair-like things, or objects covered with hair. Describing that selectional preference should be part of the semantic description of to comb. For a considerable period, these syntagmatic affinities received less attention than the paradigmatic relations, but in the 1950s and 1960s, the idea surfaced under different names. Firth (1957) for instance introduced the (now widely used) term collocation.

The distributional approach can be more radical than the mere incorporation of lexical combinatorics into the description of words: if the environments in which a word occurs could be used to establish its meaning, lexical semantics could receive a firm methodological basis. The general approach of a distributionalist method is summarized by Firth’s dictum: ‘You shall know a word by the company it keeps,’ that is, words that occur in the same contexts tend to have similar meanings. In the final decades of the 20th century, major advances in the distributional approach to semantics were achieved by applying a distributional way of meaning analysis to large text corpora. Sinclair, a pioneer of the approach, developed his ideas (see Sinclair, 1991) through his work on the Collins Cobuild English Language Dictionary, for which a 20-million-word corpus of contemporary English was compiled. In Sinclair’s original conception, a collocational analysis is basically a heuristic device to support the lexicographer’s manual work. A further step in the development of the distributional approach was taken through the application of statistics as a method for establishing the relevance of a collocation and, more broadly, for analyzing the distributional co-occurrence patterns of words (see Glynn & Robinson, 2014, for a state-of-the-art overview of quantitative corpus semantics).

1.6 Conceptual Metaphor and Metonymy

Metaphorical relations of the kind mentioned in section 1.2 (“Semantic Relations”) do not only exist between the readings of a given word: several words may exhibit similar metaphorical patterns. Conceptual metaphor theory, the approach introduced by Lakoff and Johnson (1980), includes two basic ideas: first, the view that metaphor is a cognitive phenomenon, rather than a purely lexical one; second, the view that metaphor should be analyzed as a mapping between two domains. To illustrate the first point, metaphor comes in patterns that transcend the individual lexical item. A typical example (Lakoff & Johnson, 1980, pp. 44–45) is the following.

love is a journey

Look how far we’ve come.

We are at a crossroads. We’ll just have to go our separate ways. We cannot turn back now. We are stuck. This relationship is a dead-end street. I don’t think this relationship is going anywhere. It’s been a long, bumpy road. We have gotten off the track.

The second pillar of conceptual metaphor theory is the analysis of the mappings inherent in metaphorical patterns. Metaphors conceptualize a target domain in terms of the source domain, and such a mapping takes the form of an alignment between aspects of the source and target. For love is a journey, for instance, the following correspondences hold (compare Lakoff & Johnson, 1999, p. 64).

Metonymies too can be systematic in the sense that they form patterns that apply to more than just an individual lexical item. Thus, the bottle example mentioned in section 1.2 (“Semantic Relations”) exhibits the name of a container (source) being used for its contents (target), a pattern that can be abbreviated as container for contents. Making use of this abbreviated notation, other common types of metonymy are the following: a spatial location for what is located there (the whole theater was in tears); a period of time for what happens in that period, for the people who live then, or for what is produced during that period (the 19th century had a nationalist approach to politics); a material for the product made from it (a cork); the origin for what originates from it (astrakhan, champagne, emmental); an activity or event for its consequences (when the blow you have received hurts, it is not the activity of your adversary that is painful, but the physical effects that it has on your body); an attribute for the entity that possesses the attribute’ (majesty does not only refer to ‘royal dignity or status,’ but also to the sovereign himself); and of course part for whole (a hired hand). The relations can often work in the other direction as well. To fill up the car, for instance, illustrates a type whole for part: it’s obviously only a part of the car that gets filled. For the current state of affairs in metonymy research from a cognitive semantic point of view, see Benczes, Barcelona, and Ruiz de Mendoza Ibáñez (2011).

1.7 Frames

Yet another approach to semantic structure in the lexicon focuses on the way our knowledge of the world is organized in larger ‘chunks of knowledge’ and how these interact with language. The most articulate model in this respect is Fillmore’s frame theory (Fillmore & Atkins, 1992; and see Ruppenhofer, Ellsworth, Petruck, Johnson, & Scheffczyk, 2006, for the large-scale application of frame theory in the FrameNet project). Frame theory is specifically interested in the way in which language may be used to perspectivize an underlying conceptualization of the world: it’s not just that we see the world in terms of conceptual models, but those models may be verbalized in different ways. Each different way of bringing a conceptual model to expression so to speak adds another layer of meaning: the models themselves are meaningful ways of thinking about the world, but the way we express the models while talking adds perspective. This overall starting point of Fillmorean frame theory leads to a description on two levels. On the one hand, a description of the referential situation or event consists of an identification of the relevant elements and entities, and the conceptual role they play in the situation or event. On the other hand, the more purely linguistic part of the analysis indicates how certain expressions and grammatical patterns highlight aspects of that situation or event.

An illustration comes from the standard example of frame theory, the commercial transaction frame. The commercial transaction frame involves words like buy and sell. The commercial transaction frame can be characterized informally by a scenario in which one person gets control or possession of something from a second person, as a result of a mutual agreement through which the first person gives the second person a sum of money. Background knowledge involved in this scenario includes an understanding of ownership relations, a money economy, and commercial contracts. The categories that are needed for describing the lexical meanings of the verbs linked to the commercial transaction scene include Buyer, Seller, Goods, and Money as basic categories. Verbs like buy and sell then each encode a certain perspective on the commercial transaction scene by highlighting specific elements of the scene. In the case of buy, for instance, the buyer appears in the participant role of the agent, for instance as the subject of the (active) sentence. In active sentences, the goods then appear as the direct object; the seller and the money appear in prepositional phrases: Paloma bought a book from Teresa for €30. In the case of sell, on the other hand, it is the seller that appears in the participant role of the agent: Teresa sold a book to Paloma for €30.

1.8 Prototype Effects and Radial Sets

The prototype-based conception of categorization originated in the mid-1970s with Rosch’s psycholinguistic research into the internal structure of categories (see, among others, Rosch, 1975). Rosch concluded that the tendency to define categories in a rigid way clashes with the actual psychological situation. Linguistic categories do not have sharply delimited borderlines. Instead of clear demarcations between equally important conceptual areas, one finds marginal areas between categories that are unambiguously defined only in their focal points. This observation was taken over and elaborated in linguistic lexical semantics (see Hanks, 2013; Taylor, 2003). Specifically, it was applied not just to the internal structure of a single word meaning, but also to the structure of polysemous words, that is, to the relationship between the various meanings of a word. Four characteristics, then, are frequently mentioned in the linguistic literature as typical of prototypicality.

By way of example, consider fruit as referring to a type of food. If you ask people to list kinds of fruit, some types come to mind more easily than others. For American and European subjects (there is clear cultural variation on this point), oranges, apples, and bananas are the most typical fruits, while pineapples, watermelons, and pomegranates receive low typicality ratings. This illustrates the third characteristic mentioned above. But now, consider coconuts and olives. Is a coconut or an olive a fruit in the ordinary everyday sense of that word? For many people, the answer is not immediately obvious, which illustrates the fourth characteristic: if we zoom in on the least typical exemplars of a category, membership in the category may become fuzzy. A category like fruit should be considered not only with regard to the exemplars that belong to it, but also with regard to the features that these category members share and that together define the category. Types of fruit do not, however, share a single set of definitional features that sufficiently distinguishes fruit from, say, vegetables and other natural foodstuffs. All are edible seed-bearing parts of plants, but most other features that we think of as typical for fruit are not general: while most are sweet, some are not, like lemons; while most are juicy, some are not, like bananas; while most grow on trees and tree-like plants, some grow on bushes, like strawberries; and so on. This absence of a neat definition illustrates the first characteristic. Instead of such a single definition, what seems to hold together the category are overlapping clusters of representative features. Whereas the most typical kinds of fruit are the sweet and juicy ones that grow on trees, other kinds may lack one or even more of these features. This then illustrates the second characteristic mentioned above.

The four characteristics are systematically related along two dimensions. On the one hand, the third and the fourth characteristics take into account the referential, extensional structure of a category. In particular, they consider the members of a category; they observe, respectively, that not all referents of a category are equal in representativeness for that category and that the denotational boundaries of a category are not always determinate. On the other hand, these two aspects (centrality and nonrigidity) recur on the intensional level, where the definitional rather than the referential structure of a category is envisaged. For one thing, nonrigidity shows up in the fact that there is no single necessary and sufficient definition for a prototypical concept. For another, family resemblances imply overlapping of the subsets of a category; consequently, meanings exhibiting a greater degree of overlapping will have more structural weight than meanings that cover only peripheral members of the category. As such, the clustering of meanings that is typical of family resemblances implies that not every meaning is structurally equally important (and a similar observation can be made with regard to the components into which those meanings may be analyzed).

The four characteristics are not coextensive; that is, they do not necessarily occur together. In that sense, some words may exhibit more prototypicality effects than others. In the practice of linguistics, the second feature in particular has attracted the attention, and the radial set model (which graphically represents the way in which less central meanings branch out from the prototypical, core reading) is a popular representational format in lexical semantics; see Tyler and Evans (2001) for an example.

1.9 Basic Levels and Onomasiological Salience

Possibly the major innovation of the prototype model of categorization is to give salience a place in the description of semasiological structure: next to the qualitative relations among the elements in a semasiological structure (like metaphor and metonymy), a quantifiable center-periphery relationship is introduced as part of the architecture. But the concept of salience can also be applied to the onomasiological domain.

The initial step in the introduction of onomasiological salience is the basic-level hypothesis. The hypothesis is based on the ethnolinguistic observation that folk classifications of biological domains usually conform to a general organizational principle, in the sense that they consist of five or six taxonomical levels (Berlin, 1978). The basic-level hypothesis embodies a notion of onomasiological salience, because it is a hypothesis about alternative categorizations of referents: if a particular referent (a particular piece of clothing) can be alternatively categorized as a garment, a skirt, or a wrap-around skirt, the choice will be preferentially made for the basic-level category ‘skirt.’ But differences of onomasiological preference also occur among categories on the same level in a taxonomical hierarchy. If a particular referent can be alternatively categorized as a wrap-around skirt or a miniskirt, there could just as well be a preferential choice: when you encounter something that is both a wrap-around skirt and a miniskirt, the most natural way of naming that referent in a neutral context would probably be ‘miniskirt.’ If, then, we have to reckon with intra-level differences of salience next to inter-level differences, the concept of onomasiological salience has to be generalized in such a way that it relates to individual categories at any level of the hierarchy.

This notion of generalized onomasiological salience was first introduced in Geeraerts, Grondelaers, and Bakema (1994). Using corpus materials, this study established that the choice for one lexical item rather than the other as the name for a given referent is determined by the semasiological salience of the referent (i.e., the degree of prototypicality of the referent with regard to the semasiological structure of the category), by the overall onomasiological salience of the category represented by the expression, and by contextual features of a classical sociolinguistic and geographical nature, involving the competition between different language varieties. By zooming in on the last type of factor, a further refinement of the notion of onomasiological salience is introduced, in the form the distinction between conceptual and formal onomasiological variation. Whereas conceptual onomasiological variation involves the choice of different conceptual categories for a referent (like the examples presented so far), formal onomasiological variation merely involves the use of different synonymous names for the same conceptual category. The names jeans and trousers for denim leisure-wear trousers constitute an instance of conceptual variation, for they represent categories at different taxonomical levels. Jeans and denims, however, represent no more than different (but synonymous) names for the same denotational category.

2.1 Prestucturalist Historical Semantics

The prestructuralist period (ranging from the middle of the 19th century up to the 1930s) was the heyday of historical semantics, in the sense that the study of meaning change reigned supreme within semantics. The main theoretical achievement of prestructuralist historical semantics consists of various classifications of types of semantic change, coupled with considerable attention to psychological processes as the explanatory background of changes: the general mechanisms of change included in the classifications were generally considered to be based on the associative patterns of thought of the human mind. Important figures (among many others) are Hermann Paul, Michel Bréal, and Gustaf Stern (see Ullmann, 1962, for an introductory overview). With the shift toward a structuralist approach that occurred round 1930, lexical semantics switched from a preference for diachronic studies to a preference for synchronic studies. However, the poststructuralist cognitive approach provides a new impetus for historical lexical semantics.

2.2 Structuralist Semantics

Inspired by the Saussurean conception of language, structural semantics originated as a reaction against prestructural historical semantics. The origins of structural semantics are customarily attributed to Trier (1931), but while Trier’s monograph may indeed be the first major descriptive work in structural semantics, the first theoretical and methodological definition of the new approach is to be found in Weisgerber (1927), a polemical article that criticized historical linguistics on three points. First and foremost, because the vocabulary of a language is not simply an unstructured set of separate items, and because the meaning of a linguistic sign is determined by its position in the linguistic structures in which it takes part, the proper subject matter of semantics is not the atomistic changes of word meanings that historical semantics had concentrated on, but the semantic structure of the language that demarcates the meanings of individual words with regard to each other. Second, because that structure is a linguistic rather than a psychological phenomenon, linguistic meanings should not be studied from a psychological perspective, but from a purely linguistic one. And third, because semantic change has to be redefined as change in semantic structures, synchronic semantics methodologically precedes diachronic semantics: the synchronic structures have to be studied before their changes can be considered. The realization of this attempt to develop a synchronic, nonpsychological, structural theory of semantics depends on the way in which the notion of semantic structure is conceived. In actual practice, there are mainly three distinct definitions of semantic structure that have been employed by structuralist semanticians. More particularly, three distinct kinds of structural relations among lexical items have been singled out as the proper methodological basis of lexical semantics. First, there is the relationship of semantic similarity that lies at the basis of semantic field analysis and componential analysis: see section 1.3, “Lexical Fields and Componential Analysis.” Second, there are unanalyzed lexical relations such as synonymy, antonymy, and hyponymy: see section 1.4, “Lexical Relations.” Third, syntagmatic lexical relations lie at the basis of a distributional approach to semantics: see section 1.5, “Distributional Relations.”

2.3 Neostructuralist Semantics

While componential analysis was developed in the second half of the 1950s and the beginning of the 1960s by European as well as American structural linguists, its major impact came from its incorporation into generative grammar: the publication of Katz and Fodor (1963) marked a theoretical migration of lexical semantics from a structuralist to a generativist framework. As a model for lexical semantics, Katzian semantics combined an essentially structuralist approach with two novel characteristics: the explicit inclusion of lexical description in a generative grammar and, accordingly (given that the grammar is a formal one), an interest in the formalization of lexical descriptions. Although Katzian semantics as such has long been abandoned, both features continue to play a role in this ‘neostructuralist’ tradition (the label is not an established one, but it will do for lack of a more conventional one). On the one hand, the integration of the lexicon into the grammar informs the continuing debate about the interface of lexicon and syntax; see Wechsler (2015) for an overview. On the other hand, a number of models for the formalization of word meaning have been developed, the most prominent of which is Pustejovsky’s ‘generative lexicon’ approach (1995).

2.4 Cognitive Semantics

Compared to prestructuralist semantics, structuralism constitutes a move toward a more purely ‘linguistic’ type of lexical semantics, focusing on the linguistic system rather than the psychological background or the contextual flexibility of meaning. With the poststructuralist emergence of cognitive semantics, the pendulum swings back to a position in which the distinction between semantics and pragmatics is not a major issue, in which language is seen in the context of cognition at large, and in which language use is as much a focus of enquiry as the language system. Cognitive lexical semantics emerged in the 1980s as part of cognitive linguistics, a loosely structured theoretical movement that opposed the autonomy of grammar and the marginal position of semantics in the generativist theory of language. Important contributions to lexical semantics include prototype theory (see section 1.8, “Prototype Effects and Radial Sets”), conceptual metaphor theory (see section 1.6, “Conceptual Metaphor and Metonymy”), frame semantics (see section 1.8), and the emergence of usage-based onomasiology (see section 1.9, “Basic Levels and Onomasiological Salience”).

From a theoretical perspective, the various traditions are to some extent at odds with each other (as may be expected). Specifically, structuralist (and to a large extent neostructuralist) theories tend to look at word meaning primarily as a property of the language, that is the linguistic system as an entity in its own right. Prestructuralist historical semantics and cognitive semantics, on the other hand, tend to emphasize the way in which word meanings are embedded in or interact with phenomena that lie outside language in a narrow sense, like general cognitive principles, or the cultural, social, historical experience of the language user. They then also take a more ‘pragmatic’ perspective: if the emphasis moves away from the linguistic system as a more or less stable, more or less autonomous repository of possibilities, there will be more attention to language use as the actualization of those possibilities.

Descriptively speaking, however, each of the major theoretical frameworks has contributed to the expansion of lexical semantics, that is they have drawn attention to specific phenomena and they have proposed terms, classifications, and representational formats for analyzing those phenomena. Focusing on the major topics, these contributions successively include the links between the various senses of words in prestructuralist historical semantics, the semantic relationships within the vocabulary in the structuralist era, and the importance of semasiological and onomasiological salience effects in cognitive semantics. Regardless of the theoretical oppositions, these phenomena all belong to the descriptive scope of current lexical semantics: the emergence of new points of attention has not made the older topics irrelevant.

A summary of the contribution of the major theoretical approaches is given in Table 2. If one keeps in mind the chronology of the various theories, it will be clear that regardless of the theoretical differences, lexical semantics has witnessed an outspoken descriptive expansion, from a semasiological starting point to various forms of onomasiological structure, and from a focus on elements and structures alone to the relevance of salience effects on the semasiological and onomasiological architecture of meaning.