Thursday Feb 11th:

Adam Kilgarriff
ITRI
University of Brighton

"Automatic word sketching for corpus lexicography"

Abstract

Large text corpora provide lexicographers with untold riches.  Given
suitable software, the lexicographer can instantly see as many
contextualised examples of the word they are trying to define (the
nodeword) as they could wish for.  This presents new problems: how do
they get an overview?  To date, lexicographers have used collocate
lists, sorted according to some statistical measure of salience, but
these lists have not been syntactically sensitive.  They have simply
looked at words co-occurring with the nodeword in some specified
window such as "between 1 and 3 words before the nodeword". I shall be
presenting a strategy for producing a set of collocate lists for each
nodeword, with a different list for each grammatical relation.

I shall sort shortly be applying the strategy in "production mode" to
the common nouns, verbs and adjectives of English --- so feedback and
suggestions are particularly welcome.