Question:
You may know this already, but the sad fact of the matter is that there
is no reference grammar for any human language that is really suitable
for NLP applications. There are reference grammars that provide a
frame of reference for discussion (probably these are the best answer
to your question). There are also reasonably effective grammars for
very special small tasks like making a travel reservation or
translating weather reports or understanding air traffic control or
(kind of) extracting limited kinds of information from news reports,
but all of these make serious compromises in order to work at all and
none can be considered a grammar of English. Many of these examples
work by focussing on very small pieces of the input and essentially
ignoring the rest. Others work by limiting the domain of discussion
and possibly by parsing text using systems so complex that they are not
easily inspectable by humans.
That said, have you looked at the "Oxford English Grammar" or "A
Comprehensive Grammar of the English Language"? You have to love the
fact that the first author of the second work is named "Quirk". Both
of these will give you the terminology you ask for with the second
being less readable, but more comprehensive.
The commentary on the Comprehensive Guide make it sounds like the
Cambridge reference is also a serious contender.
http://www.amazon.com/exec/obidos/tg/detail/-/0198612508/102-2010582-...
http://www.amazon.com/exec/obidos/tg/detail/-/0582517346/102-2010582-...
http://www.amazon.com/exec/obidos/tg/detail/-/0521431468/ref=pd_sim_1...
Answer:
The disenchantment with grammars in NLP stems from past failures of
parsers that are primarily grammar driven. That is, parsers whose main
mechanism is to execute a recursive grammar specification.
However, systems that use such grammar execution as one type of
component can do well. Similarly for systems that can use grammar as a
source of knowledge or corroboration. For example...
TAIParse is a multi-pass analyzer that uses recursive grammar passes as
well as pattern-based passes that emulate the knowledge and processing
expressed in grammars -- but with the control afforded by a natural
language engineering approach.
Submit Your
Own Answer!
