Lexical Semantics Flashcards
Hyponymy
IS-A Relationship i.e. subclass relationship. e.g. dog is a hyponym of animal.
Hypernymy
Superclass relationship e.g. animal is the hypernym of dog.
Meronymy
Part-of relationship. Only applies to functional parts not to all pieces.
Synonymy
Two words with the same meaning or nearly the same meaning.
Antonymy
Opposite meaning - mainly applicable to adjectives.
WordNet
Main resource for English Lexical Semantics.
Synset
Wordnet set of near synonyms.
Uses of hyponymy
Semantic classification - selection restrictions (e.g. object of eat must be edible) and for named entity recognition.
Shallow inference - ‘X murdered Y’ implies ‘X killed Y’
Back-off to semantic classes in classification
Word-sense disambiguation
Query expansion for IR.
Polysemy
A word having more than one sense.
Homonymy
A word having two unrelated sense (rare most sense are actually related).
Collocations
A group of two or more words that occur together more often than would be expected by chance.
Yarowsky’s Unsupervised Learning Approach to Word Sense Disambiguation
Seed collocates are chosen for each sense. These are used to accurately identify distinct senses. Sentences in which the disambiguated sense occur can be used to learn other discriminating collocates automatically. This process is then iterated.
Steps in Yarowsky’s
Identify all example of the word to be disambiguated in the training corpus and store their context.
Identify seeds which reliably disambiguate a few of these uses. Tag these automatically and count rest as residual.
Train decision list classifier on these examples.
Apply this classifier to the training set and add all examples tagged with greater than threshold reliability to the Sense A and Sense B sets.
Iterate steps 3 and 4 until convergence.
Apply classifier to unseen test data.
