[Corpora-List] ad-hoc generalization and meaning

[Paula Newman]

Rob, 

Re:
YW> Like many bystanders, Ive been the resisting the temptation to join in this fight, with its wonderful mix of the odd and the very familiar.... 

Ditto. But can't resist any further.

Maybe the oddness relates to the question of the purpose of  the "informal grammar" , which is never stated.   

There are many overlapping purposes to investigations of human language, i.e, to "linguistics".  Among them are:   
- to identify the different ways in which languages pattern, at different levels (morphological, syntactic, semantic, discourse)
 - to understand what kinds of meanings are conveyed by language and how those relate to surface forms, again at different levels.
- to explore how languages evolve.   
- to investigate how language is acquired.  
- to understand how the brain processes language

It is always (or almost always) understood that studying one aspect is just that.  For example, noone  studying morphological patterns believes that the result will be a total description of a language. 

Natual language processing (NLP), on the other hand, has different purposes relating not so much to understanding how language works, but to doing something useful with language automatically.   Like any kind of computational task  the problem is to design systems that can perform the desired functions in an acceptable amount of space and time.   For this purpose, the linguistic determinations and formalizations of regularities in  different aspects of  human language and human language processing can supply bases for modularization .  Those base can make mnay applications feasible even though a comprehensive formalization of natural language meaning in all its aspects remains beyond our reach.  

Admittedly, the term "computational linguistics"(CL) tends to muddy the waters, because in any particular use it encompasses one or more of:  (a) the development of formal notations for the results of linguistic studies, (b) linguistic studies whose results are formally expressed,  and (c) symbolic NLP and (d) statistical NLP ( which in the last decade+ has dominated ACL conferences).    To sloppily generalize, statistical NLP  allows a blending of language aspects usually treated separately in linguistic studies and in symbolic NLP,  

So the question, Rob, is what are you proposing?  Is it a new approach to linguistic investigation, or to NLP, or to ??

Paula
 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/corpora/attachments/20070913/5c77e184/attachment.htm>
-------------- next part --------------
_______________________________________________
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora