engleski

Quantitative data and the structure of lexical categories

The present paper investigates the extent to which linguistic frequencies, as extracted from computer corpora, can be justifiably used in cognitive linguistic analyses as indicators of cognitive phenomena underlying language use. It has been shown that quantitative data, i.e. corpus frequencies and frequency-based statistical calculations and techniques can be used to make claims about the internal structure of linguistic categories. In case of lexical meanings, this refers not only to determining and distinguishing between several senses in a lexical network, but also to establishing the most prototypical sense and positing relationships among senses, in terms of their similarity (for example by means of cluster analyses), as well as their position with respect to the center-periphery distinction within the category. While such analyses and the resulting findings may be plausible in the case of some syntactic constructions and their prototypical instances, word senses exhibit internal structures that may not be amenable to this kind of analysis. More specifically, successful application of a technique which uses frequencies to make claims about the conceptual structure of a lexical category does not imply its applicability to other categories, because of the differing degrees of similarity between senses within polysemous networks, i.e. different types of polysemous categories. Some studies claim that the most frequent sense, as the most important among several factors, indicates prototypicality. While this may hold for categories with a characteristic prototypical core, there are various other cases where the correlation between frequency and prototypicality depends on the way the senses of a category are interrelated. Using as an example the English noun order, I examine how different frequency data extracted from the BNC can (or cannot) be used to point in the direction of the semantic structure of the lexical category in question. The most frequent senses of order are 1) arrangement (My files are in alphabetical order), 2) request for food, drink or goods (The waiter took our orders), and 3) instruction, command (When the order was repeated, he refused to obey). Although related, they are quite distinct and are not readily subsumed under a common schema (in cognitive grammar terms). Rather, they are more likely to have the status of entrenched units, which means that there is no clear prototypical core with a corresponding schema that holds the category together, regardless of their frequencies in the corpus that may seem to point to the salience of one sense, that of instruction, command. The analysis will thus show that the nature of sense relations of a word is not always captured by their frequency counts. While frequency-based techniques may be helpful in assigning corpus examples to different senses and pinning down contextual (semantic and syntactic) features of individual senses of order, their cognitive salience and the overall internal semantic structure of a word is left out of reach if we rely only on frequency data. In other words, quantitative data can validate conclusions made within a relevant theoretical framework, but we should avoid the pitfalls of turning them into a goal in itself.

cognitive semantics; corpus frequencies; lexical categories; word senses

nije evidentirano