Diana McCarthy
Research Interests
1) Lexical Acquisition:
(enabling computers to automatically learn about words)
selectional preferences
(preferences that words have for
co-occurring words)
For example, so that a computer can learn the sorts of words that occur as the
object of verbs like eat or open. This might help a computer work out what
it refers to in the following sentences
"The packet contained chocolate but nobody was allowed to eat it."
"The packet contained chocolate but nobody was allowed to open it."
diathesis alternations
finding out which verbs can transform like break in "The boy broke the window." <-> "The window broke."
non-compositional multiwords
phrases which mean something different to
what might be expected given the constituent words
like phrasal verbs e.g. cough up (which can mean pay in
British English)
or verb object combinations e.g. take the biscuit (which can mean
be worse than anything in British English)
2) Word sense disambiguation
(enabling computers to determine the meaning of a
word in a given context)
I have used unsupervised methods for determining the sense of
words in text using selectional preferences (see above)
I work on methods for acquiring rankings over word
senses to provide NLP systems with information on sense frequency
given a particular corpus. Whilst the "celebrity" sense of
star might be common in popular newspaper text, the "celestial
body" sense might be more predominant in a scientific domain. See out project Ranking
Word Senses for Disambiguation: Models and
Applications
I am interested in techniques for evaluating methods which a) learn what
meanings words have and b) learn ways of determining the right meaning in a
given context. I co-organised an International lexical substitution task with Roberto Navigli
Back to Diana McCarthy home page