Re: Tree-Adjoining Grammars and Cross-Serial Dependencies in Dutch Embedded Clauses (was: Re: Unsupervised learning of natural languages)
|From:||Henrik Theiling <theiling@...>|
|Date:||Tuesday, December 20, 2005, 1:06|
Thomas Hart Chappell <tomhchappell@...> writes:
> My library finally borrowed "Tree Adjoining Grammars: Formalisms,
> Linguistic Analysis, and Processing", ...
Nice! I had a seminar on TAGs at uni and TAGs are quite interesting
since they are parsable in O(n^6) (IIRC) instead of the O(e^n) for,
e.g., HPSG. I found they are much more the way to go in analysing
language than the more linguistically inspired grammars like HPSG. It
tries to solve the parsability problem for a certain class of langs
that cannot be handled with CFGs, while still being polynomial.
Obviously, I'm a programmer so I want to approach the problem from the
implementational point of view all the time. O(n^6) is awfully
complex to me!
> In their overview, the editors mention the Dutch sentence fragment:
> "... omdat Wim Jan Marie de kinderen zag helpen leren zwemmen"
That's just like the sentence that I made up! :-) Only the proper
names seem to make it parsable more easily. That one's quite easy!
> They show how it can be handled by just one rule of a T.A.G., (_with_ _no_
> _subsequent_ _transformations_, btw), and, like you, say it can't be
> generated by context-free grammar.
Please tell us what you think when you've read the article.