LINGUIST List 8.501

Fri Apr 11 1997

Qs: Parser testing, Universal, Software

Editor for this issue: Susan Robinson <suelinguistlist.org>

Directory

1. Gordon Franck, Concept based Parser Testing
2. Arthur Merin, Testing for a Universal (`but', `also')
3. MEYER, TOM, CALL Software

Upon reading the comparison of web based parsers with some interest, I
am considering resurrecting a web version of some parser-tester software
that an advisee of mine (Dave Zeitler) wrote last summer. Our aim was
to provide a test of parsers based upon linguistic concepts rather than
raw online corpora. Our method was to extract starred and grammatical
sentences from linguistic texts, e.g., Radford 81, Quirk 72, and add
them to a database. The sentences in the database were indexed by the
concepts involved, e.g., subject/verb agreement, extraction, etc., much
as they were organized in the original texts. When this database was
partially complete we created an interface to feed these sentences to a
particular parser, receive the parsers grammaticality judgements, and
score these judgements for agreement with the database. The parser
tester would then provide output stating the parsers degree of success
(a true/false positive/negative table) organized by linguistic concept.
A user (parser developer) could click on the concept score to get
feedback on the particular sentences involved (in any table cell). We
judged our system to be successful (Dave upon graduation) as it provided
broad linguistic coverage, and linguistically informative feedback, with
minimal effort (e.g., small lexicon) involved by the parser developer.
However, there would be a significant amount of work involved in
completing the database and making this software net accessible, so
before going any further I am hoping to get some feedback on the
following questions. I will post a summary to the list.

1. Does this software already exist?
2. Who would find this method of parser testing meaningful?
(alternatively, which folks couldn't give a hoot whether their parser
handles parasitic gaps)
3. Would the database be useful to linguists (with a reasonable user
interface) independent of the issues of parser testing?

Thanks in advance for your feedback. I will summarize any replies sent
directly. Sincerely, Gordon

Dr. Gordon Franck
E-mail: csgfmizzou1.missouri.edu

REQUEST FOR PARTICIPATION IN TESTING FOR A LANGUAGE UNIVERSAL

The candidate, engaging a phenomenon parts of which Zellig 
Harris noted as a syntactic fact of English, is:


 Any language having bona fide translation equivalents 
 of the coordinating conjunction `but' will have 
 equivalents of
 
 (1) `Kim walks but {Kim/(s)he[K]} talks' (Pa but Qa)

 acceptable in suitable contexts (here, e.g.: we are 
 looking for a very silent messenger), whereas 
 equivalents of

 (2) *`Kim walks but Sandy walks' (Pa but Pb)

 will never be acceptable (assuming default prosody 
 or a suitable equivalent - see below), while in any 
 language having, in addition, a bona fide translation 
 equivalent of `also' or `too', equivalents of

 (3) `Kim walks but Sandy {also walks/walks too}' 

 will always be acceptable again. 


Explanatory Note: "Default prosody" for English means at the 
very least: absence (or nonobligatoriness) of a marked pause 
preceding `but'. I.e. the unacceptability judgment for 
(2) is claimed to be stable for that single-speaker reading 
(both with regard to prosody and interpretation) which is NOT 
paraphraseable as

(4) `Kim walks, but then Sandy walks'

where `then' is NON-TEMPORAL, as evidenced by preservation of 
its intended interpretation in

(5) `Kim has walked, but then Sandy has walked'.

(The intended and, for atemporal `then', presumably obligatory 
reading for the English ex. (4) is one where the second clause 
introduces an explanation for the eventuality designated by the 
first.)


Native or born-again judgments are solicited on bona fide 
translation equivalents of (1), (2) and (3); if possible 
also on those of (4)/(5) and on the feasibility of a 
(4)-style reading for (2)-equivalents. Most helpful would 
be transliterations of exx. into Roman characters, if necessary 
making use of ASCII diacritic conventions in use among students 
of the language concerned, with a word-by-word English or
(e.g., where word order differs) quasi-English translation 
underneath.

A summary of the results of the experiment will be posted,
along with a list of all those contributors who do 
not express a preference for not being listed.
For participants or anyone else interested in the explanation 
of the putative universal, a dense outline of the argument 
(2 pages worth of PostScript or, on special demand, 
typographically approximate ASCII) is available by e-mail on 
request (repr. from the Abstracts of the 10th Int. Cong. Logic 
Methodology & Philosophy of Sci., Florence 1995). A mathematical 
background or access thereto is presupposed. An extensive version, 
taking account, it is hoped, of the results of the experiment 
proposed above, will be available in English soon. Participants 
expressing an interest in being notified of its venue of 
appearance will be kept informed. 


Arthur Merin
Institute for Language and Computation (IMS)
University of Stuttgart
Azenbergstr. 12
70174 Stuttgart 
Germany

I am in the process of reorganizing our computer lab here at the
University of Miami, Intensive Language Institute and am looking for
some good CALL software in the areas of ENGLISH pronunciation,
grammar, vocab. development, writing, etc.

Any suggestions?

Thank you.

Please respond directly to Tom Meyer via email at
 tmeyercstudies.msmail.miami.edu