Word Sense Disambiguation Flashcards
Where can we get a list of all the different senses of a word
wordnet
what is a ‘supersense’
gives a lexicographic category based on word meaning e.g. location
what is a ‘sense relation’
describes things like ‘part of’, or subbordinate
What is the word sense disambiguation task
given a word in its context and lots of different possible word senses
output the correct word sense given the context
describe the hypernym relationships
from concepts to superordinates
breakfast->meal
describe the hyponym relationship
from concepts to subtypes
meal->lunch
describe the instance hypernym relationship
from instances to their concepts
jane austen->author
describe the instance hyponym relationship
from concepts to their instances
composer->bach
describe the part meronym relationship
has part
from wholes to parts
table->leg
describe the part holonym relationship
part of
from parts to wholes
course->meal
describe the antonym relationship
semantic opposition between lemmas
leader->follower
describe the derivation relationship
lemmas with the same morphological root
destruction -> destroy
what are the two possible tasks to tackle word sense disambiguation
- the lexical simple task- the sense of just one word
- all words task- the sense of every word in the sentence
what is the lesk algorithm
a simple but good baseline approach
- choose the word whose dictionary or definition shares the most words with the target words neighbourhood
what are the benefits of the lesk algorithm (2)
simple
no training data
what are the disadvantages of the lesk algorithm (3)
not all words have word net definitions
ambiguous context words
poor performance compared to other methods
what is semcor
the largest corpus with word net labelled senses
how can we use supervised learning for all word WSD
tag each word with its word net senses
POS tags
collocation features of words and n-grams
weighted average of word embeddings
^ as features into a svm
