16 September 2010

Lecture 6: Part of Speech Tagging

Woohoo, some linguistics!

Parts of speech are useful for several things:
  • Knowing how to pronounce a word (object noun vs verb)
  • Knowing what words come around it (personal pronouns vs possessive pronouns)
  • Knowing how it behaves syntactically -- most verbs behave the same and are "substitutable" from a syntactic perspective (though probably not from a semantic perspective)
Broadest categorization: open class versus closed class

Note: not all parts of speech exist in every language, and if a words is in one part of speech in language X, it's translation isn't necessarily the same part of speech in language Y.  (Eg., beautiful in English/Korean, or green in English/Japanese).

Some major open classes:
  • Nouns: things.  Concrete (ship, chair), abstract (happiness) or verb-like terms "gerunds" (running), proper "Hal"
  • Verbs: actions.  Eat, drink, sleep; English has auxiliaries but these are usually considered closed class (see below)
  • Adjectives: mostly modify nouns (blue, big, old)
  • Adverbs: mostly modify verbs (slowly, very) but also other weird things (every, unfortunately, not)
Almost every language has Nouns and Verbs, and many have Adjectives and Adverbs.

Some closed class:
  • Prepositions: the bunny ran ___ the hill
  • Determiners: articles: a, an, the; others: this blog
  • Pronouns: (personal) she, I, who, (possessive) my, her, (wh) what, how
  • Conjunctions: and, but, either/or
  • Auxiliary verbs: would, could, can, to -- but not the preposition "to"
  • Particles:  like prepositions, but go with verbs (put down, raise up, etc.)
  • Numerals: one, two, three, first, second, third, etc.
Part of speech tagging is the task of assigning parts of speech to words in a sentence.  About 11% of English word types are ambiguous, but over 40% of word tokens are.

Rule based approaches usually consist of:
  • A small lexicon that covers some common open-class words and most common closed-class words
  • A set of rules for disambiguation
  • A set of heuristics for guessing the tag of unknown (open class) words
Lexicon is easy.  Heuristics are things like:
  • Does it end in "ed"?  "ly"?  "ing"?  "s"?
  • Is it capitalized?  (Doesn't apply to P1)
If these don't help, then it's probably an unknown open class word and we use context (rules) to disambiguate.

Rules get more complicated...
  • Noun/Verb disambiguation is usually the biggest issue
    • Is there a determiner or adjective to my left?  How far?
    • Am I at the end of a sentence?
    • Is the an auxiliary to my left?  An adverb?
  • There's also Verb/Aux disambiguation
    • Verb Aux is rare order, but Aux Verb is common
    • Is there a verb anywhere in the sentence?
  • Noun/WH-complementizer disambiguation
    • John saw that book
    • John saw that book burn up
    • Is there a verb to the right, nearby?
There are other systematic ambiguities.  The book talks about some, and you can try to find more in P1!

No comments:

Post a Comment