Pustejovskian semantics
The problem of word-sense disambiguation was the stimulus for an important approach to the lexicon in computational linguistics developed by Pustejovsky (1991, 1995; Pustejovsky and Boguraev 1994). This approach, dubbed the generative lexicon approach, claims to solve many of the types of problem seen in the previous section by adopting an entirely different picture of lexical information. Pustejovsky criticizes the standard view of the lexicon, on which each lexeme is associated with a fixed number of senses. Lexical databases like WordNet are exactly like traditional dictionaries, in that they simply enumerate a static number of fixed senses for every word. Pustejovsky and Boguraev describe the generative lexicon alternative as follows:
In contrast, rather than taking a ‘snapshot’ of language at any moment of time and freezing it into lists of word-sense specifications, the model of the lexicon proposed here does not preclude extendability; it is open ended in nature and accounts for the novel, creative uses of words in a variety of contexts by positing procedures for generating semantic expressions for words on the basis of particular contexts. (Pustejovsky and Boguraev 1994: 297)
This can be exemplified with the adjective fast, which traditional descriptions (including WordNet) need to credit with a large number of different senses. For instance, fast seems to have rather different senses in the three phrases in (25):
These uses of fast seem to involve at least three senses: ‘performing an act quickly’ (25a), ‘moving quickly’ (25b) and ‘involving a fast tempo’ (25c). Any computer needs to keep these senses separate, recognizing, for example, that a ‘fast typist’ is not one who moves quickly in the way that a fast car does. This is usually achieved simply by listing a number of distinct senses for fast. Pustejovsky notes a problem with this strategy: it will not account for ‘creative applications’ of the adjective in English, like fast motorway or fast garage. Neither of these fits any of the senses in (25): the former refers to a motorway on which vehicles can travel fast, the latter to a garage which services cars quickly. Pustejovsky claims that the standard sense-listing approach to the lexicon will always fail to cover all the possible meanings of an adjective like fast; furthermore, it is ‘unable to capture interesting generalizations concerning relationships between “senses” of the same word’ (Pustejovsky and Boguraev 1994: 302).
In order to avoid these problems, Pustejovsky proposes that lexical analysis needs to recognize different levels or perspectives of lexical meaning. Two of these levels, argument structure and lexical inheritance, are already familiar. Argument structure is the level of lexical representation for verbs which specifies the number and type of noun complements (see Chapter 10). Lexical inheritance structure refers to the conceptual relations between words in the lexicon, as discussed in 8.2.1. In addition, Pustejovsky identifies event structure (see 10.2), and qualia structure as crucial levels of word meaning. We will concentrate here on qualia structure.
Qualia structure is a system of relations that has a similar importance for the meaning of nouns as argument structure has for verbs. It reflects those aspects of the referent of a noun which ‘have long been considered crucial for our common-sense understanding of how things interact in the world’ (Pustejovsky and Boguraev 1994: 305). A word’s qualia structure has four aspects, its Constitutive Role, Formal Role, Telic Role and Agentive Role. These roles constitute the framework for the word’s meaning. Pustejovsky describes the type of information expressed by these roles as follows (1995: 426):
Constitutive Role: the relation between an object and its constituents, or proper parts.
• Material
• Weight
• Parts and component elements
Formal Role: that which distinguishes the object within a larger domain.
• Orientation
• Magnitude
• Shape
• Dimensionality
• Colour
• Position
Telic Role: purpose and function of the object.
• Purpose that an agent has in performing an act
• Built-in function or aim that specifi es certain activities
Agentive Role: factors involved in the origin or ‘bringing about’ of an object.
• Creator
• Artefact •
Natural Kind
• Causal Chain
How does this relate to the problem of determining the right meaning of fast in the examples above? The essence of Pustejovsky’s theory is that the lexical entry for fast specifies that it only ever applies to the telic role of nominals. The telic role specifies the purpose and function of the noun. Thus the specification of the telic role of motorway tells us that a motor way’s purpose is to bring about a certain event, road travel: this is what fast is referring to. Similarly, typist has a telic role determining the function of performing a different type of event: typing, and it is this which fast qualifies. Similarly, a fast waltz is fast with respect to the telic role of that noun, which will refer to dancing. By introducing an additional level of structure into the description of nominals, this approach succeeds in retaining a single meaning for fast, which will be defined as something like ‘at a rapid rate’; the contextual meanings it takes on in the noun phrases above are a result of the differing telic roles to which this single meaning applies.
Easy and hard might be two other adjectives which are specified as refer ring to a noun’s telic role. Thus, easy books, easy loads, easy (ski) slopes and easy software are easy with respect to their telic roles: being read, being lifted, being skied down and being used. Similar remarks apply to hard. We are not likely to interpret an easy book as a book that is easy to lift, since lifting is not part of the event described in the telic role for book.
Another application of qualia structure occurs in the analysis of sentences like (26a) and (26b):
The sentences’ interpretations are fundamentally different: the potato exists before Sam baked it, whereas the cake is only created by the act of baking. Pustejovsky (1995: 421–423) accounts for this contrast through differing agentive roles in the qualia structure of the two nouns. Cake is classified as an artefact; as a result, it is part of our specific lexical knowledge about it that there is an event associated with its coming into being. The verb bake is thus interpreted as referring to this aspect of the noun’s qualia structure. Potato, on the other hand, is classified as a natural kind. It cannot therefore be brought into being by any artificial process. The predicate bake is thus understood as denoting an action which takes place on the potato, but which does not bring the potato into being.
The introduction of qualia structure allows us to avoid postulating a large number of polysemous senses for a single lexical item. Systematic ambiguity in words like fast, easy and bake is not explained by postulating multiple senses for these words, but by recognizing that the underlying structure of their associated nominals is more complex than was originally thought. The increased semantic complexity of lexical entries for nominals is ultimately a saving, since the uniformity of qualia structure allows both a systematic account of polysemy and a reduction in the over all number of senses postulated in the lexicon.
