Most of existing natural language parsers are designed to analyze structurally
correct sentences. For the purpose of second language learning, this is not
enough, since beginners of a new language often make some structural errors. I
am currently looking for some parsers that can not only work as a recogniser
(that is, simply say 'no' to a structurally ill-formed sentence) but are
actually able to do some sort of analysis to ill-formed sentence. I find some
parsers based on Marcus' deterministic parsing theory belong to this type, but
I think there must be other approaches but don't know what they are. The
particular questions I have are:
1. are there any approaches better than using ad-hoc methods for analysing
sentences containing structural errors?
2. if there are, then are there any parsers actually developed (programmed)
based on those approaches? Where can I find them or who can I contact to get
more information about them?
3. there must be some parsers (with or without error tolerance) implemented
in Common Lisp that are publicly available, where can I get them?
I'm a student doing Masters and working on the field of robust language
analysis for second language learning. It will be very much appreciated if
anybody there can provide me any sort of information related to the above
questions. You can either post to these newsgroup or email me personally.
Thanks in advance!
Jiansheng Jiang (email: ·····@latcs1.lat.au.oz)
Department of Computer Science and Computer Engineering,
La Trobe University
Australia
·····@latcs1.lat.oz.au (Jiansheng Jiang) writes:
> 1. are there any approaches better than using ad-hoc methods for analysing
>sentences containing structural errors?
Well, I am not sure how you classify "ad-hoc", but most of the practical
NLP systems deal with ill-formedness one way or another. E.g. Wilks developed
the Preference Semantics which he showed to work well with ill-formed
sentences (whether syntactically or semantically), Schank worked on the
concept of "interestingness" which would let you deal with semantically
incorrect sentences, Zadeh came up with fuzzy interpretation of meaning
(not on the level of syntax though), and then there is the whole heap
of people working on connectionist parsing which would accomplish exactly
what you want, here is a list of refs (which may not be entirely what you
want):
Fass, D. and Y. Wilks (1983). Preference Semantics, Ill Formedness,
and Metaphor.S Americal Journal of Computational Linguistics 9(3-4):
[JLC-190]
Howells, T. (1988). VITAL: A Connectionist Parser. Cognitive Science
88, [JLC-069]
Pollack, J. B. and D. L. Waltz (1984). Parallel Interpretation of
Natural Language. Proc. FGCS, ICOT. [JLC-054]
Schank, R. C. (1979). Interestingness: Controlling Inferences.
Artificial Intelligence 12: 273-297. [JLC-063]
Small, S., G. Cottrell, et al. (1982). Toward Connectionist Parsing.
Proc. AAAI-82, AAAI. [JLC-106]
Zadeh, L.A. (1981). PRUF - a meaning representation language for natural
languages in Mamdani and Gaines, Fuzzy Reasoning and its Applications,
Academic Press.
Jacob
=============================================================================
Jacob L. Cybulski, Deputy Director, Amdahl Australian Intelligent Tools
Programme, Dept of Comp Sci & Comp Eng, La Trobe University, Bundoora,
Vic. 3083, Australia. Ph: +613 479 1270, Fax: +613 470 4915, Telex: AA 33143,
E-Mail: ·····@latcs1.lat.oz.au. ... G'Day
=============================================================================
In article <·····················@latcs1.lat.oz.au>, ·····@latcs1.lat.oz.au (Jiansheng Jiang) writes:
|> Most of existing natural language parsers are designed to analyze structurally
|> correct sentences. For the purpose of second language learning, this is not
|> enough, since beginners of a new language often make some structural errors. I
|> am currently looking for some parsers that can not only work as a recogniser
|> (that is, simply say 'no' to a structurally ill-formed sentence) but are
|> actually able to do some sort of analysis to ill-formed sentence. I find some
|> parsers based on Marcus' deterministic parsing theory belong to this type, but
|> I think there must be other approaches but don't know what they are. The
|> particular questions I have are:
|>
|> 1. are there any approaches better than using ad-hoc methods for analysing
|> sentences containing structural errors?
Yes, semantics-based parsers are good at this. I've used this approach in
several commercial NLP systems accepting ill-formed input in electronic messages,
keyboard input and handwritten input. Some of the older variants of this approach
are considered by some to be ad hoc, but the more recent research has been on
memory-based parsing or case-based parsing, which limits the amount of "knowledge
engineering" (read "ad hoc hacking").
|> 2. if there are, then are there any parsers actually developed (programmed)
|> based on those approaches? Where can I find them or who can I contact to get
|> more information about them?
Here are some introductory references on semantics-based NLP. There's been a very
large amount of research published on this approach, mostly by Roger Schank's
students at Yale and the Institute for the Learning Sciences at Northwestern University.
Dyer, Michael, _In-Depth Understanding, MIT Press, Cambridge, 1983.
Cullingford, Richard, _Natural Language Processing_, Rowman & Littlefield, 1986.
Schank, Roger and Christopher Riesbeck, _Inside Computer Understanding_, Lawrence
Erlbaum Assoc., Inc., Hillsdale, N.J., 1981
|>
|> 3. there must be some parsers (with or without error tolerance) implemented
|> in Common Lisp that are publicly available, where can I get them?
Cullingford's book includes source code in Franz Lisp, which someone may have
translated to CL by now.
|> I'm a student doing Masters and working on the field of robust language
|> analysis for second language learning. It will be very much appreciated if
|> anybody there can provide me any sort of information related to the above
|> questions. You can either post to these newsgroup or email me personally.
|>
|> Thanks in advance!
|>
|> Jiansheng Jiang (email: ·····@latcs1.lat.au.oz)
|> Department of Computer Science and Computer Engineering,
|> La Trobe University
|> Australia