Natural Language Parsing
- Humans take strings of words (e.g. textual sentence, or speech)
and combine them to get a semantic representation (and maybe
syntax).
- Generally people do this in one pass (no back tracking).
- There was a big argument in the psycholinguistic community in
the 80s and 90s if parsing was syntax first or semantics first,
which has been resolved with both first.
- Sentences are often ambiguous (e.g. I saw the girl with
the telescope.) and people often don't notice the ambiguity.
- There is a lot of other ambiguity in NL processing.
- We don't have a complete model of parsing.
- We use parsing for text engineering, and that's not perfect either.
- Similarly, there isn't agreement to the grammar of English,
- and semantics is poorly understood.
- NL is thought to be context free (S-> NP VP.).
