Week4-2Parsing

Parsing human language

Rather different from computer languages

• No types for words(variable, comment, …)
• No brackets around phrases
• Ambiguity
  
  • words
  • parses
• Implied information

Parsing

• Parsing means associating tree structures to a sentence, given a grammar(often a CFG)
  
  • There may be 0, 1, or more than 1 such tree structures for the given sentence
• Grammars are declarative
  
  • They don’t specify how the parse tree will be constructed

Syntactic ambiguities

• PP attachment
• Gaps
  
  • Mary likes Physics but hates Chemistry.
    -Coordination scope
  • Small boys and girls are playing.
• Particles vs. preposition
  
  • She run up a large hill.
    -Gerund vs. adjective
  • Frightening kids can cause trouble.

Applications of parsing

• Grammar checking
• Question answering
• Machine translation
• Information extraction
• Speech generation
• Speech understanding
• Interpretation

Context-free grammars

• A CFG is 4-tuple $(N, \Sigma, R, S)$
  
  • N: non-terminal symbols
  • $\Sigma$: terminal symbols(disjoint from N)
  • R: rules($A \rightarrow \beta$), where $\beta$ is a string from $(\Sigma \cup N)^\ast$(a string from the union set of $\Sigma$ and N with 0 to a large number)
  • S: start symbol in N

Phrase structure grammar

• Sentences are not just bag of words!(AGAIN!)
• Context view of language
  
  • A PP looks the same whether it is part of the subject NP or part of VP
• Constituent order
  
  • SVO(subject verb object)
  • SOV
  • …
• Auxiliary verbs
• Imperative sentences
• Interrogative sentences
• Negative sentences

Leftmost derivation

• A leftmost derivation is a sequence of string $s_1, s_2, ..., s_n$
  -$s_1 = S$, the start symbol
  -$s_n$ includes only terminal symbols