***Call***
"I am performing a "definitive" analysis of the meanings of subordinating
conjunctions and would be interested in linking up with anyone who has
focused on their representation in NLP systems. I am performing an
analysis of subordinating conjunction definitions in Webster's 3rd
International Dictionary, modeling these definitions using the theory of
labeled directed graphs (digraphs), using principles for identifying
primitives I have previously described (see Litkowski, K. C. (1988). On
the search for semantic primitives. Computational Linguistics, 14(1), 52
for an overview).
The "meaning" of subordinating conjunctions essentially consists of
labeling clauses and establishing discourse relationships of time,
contingency, place, condition, concession, contrast, reason, purpose, and
result (see Quirk et al. pp. 1070-1112). I am aware that subordinating
conjunctions are used as cue words in discourse processing, but I am not
aware of any systematic bringing together of these "meanings" in a
computational system. Characterizing these meanings is important in the
digraph analysis, and while I can do it myself, it would be preferable
not to reinvent the wheel. I would be grateful if anyone can point to
computational representations of these meanings.
A database of these "meanings" will eventually be made publicly available
on the web for anyone to use."
***Responses***
Many respondents noted correctly that another term which subsumes
subordinating conjunctions (SCs) is "discourse markers," for which there
is a substantial literature. Megan Duque-Estrada has a very extensive
bibliography of this literature available on the web at
http://www.ufpa.br/~megan. This literature provides substantial
information pertinent to my request.
More specific information going to the heart of my request for features
and semantic labels associated with SCs was provided by Alex Eulenberg,
Ken Barker, and Alistair Knott. Mary Dee Harris provided the link to
Ali's work; I am very grateful for this link.
Alistair Knott (http://www.cogsci.ed.ac.uk/~alik/publications.html,
particularly "A Data-Driven Method for Classifying Connective Phrases")
and Alex Eulenberg (http://php.indiana.edu/~aeulenbe/, providing features
for additive conjunctive sentence adverbials) provide an identification
of features associated with comprehensive lists of "cue phrases" (that go
beyond the smaller set of SCs). The feature names include MODAL STATUS,
POLARITY, FOCUS OF POLARITY, PRESUPPOSITIONALITY, SOURCE OF COHERENCE,
ANCHOR, PATTERN OF INSTANTIATION, and RULE TYPE.
Ken Barker (http://www.csi.uottawa.ca/~kbarker/, particularly
"Interactive semantic analysis of clause-level relationships, CLRs)
provides an identification of semantic labels assigned to clauses based
on CLR markers or clausal connectives. The semantic relationships
include CAUSAL (CAUSATION, ENABLEMENT, ENTAILMENT, PREVENTION,
DETRACTION), TEMPORAL (CO-OCCURRENCE, PRECEDENCE), and CONJUNCTIVE
(CONJUNCTION, DISJUNCTION).
These two sets of information respond precisely to my needs and are quite
useful in their own right.
Another respondent (who prefers to remain anonymous), who has
investigated diachronic processes (including SCs in German), noted a
possibly very interesting point that Old High German had no
Complementizer Phrases, so that "das" evolved into "dass". This person
also cited work describing subordinating conjunctions with prepositions
(like: bevor, nachdem, indem) having a descriptive part (-vor-, nach-)
and a referential part (expressed by the d-words or the w-words),
including pairs like "nachdem -wonach", "dadurch dass - wodurch", and
"damit - womit". There is thus the suggestion (to me, at least) that the
use of subordinating conjunctions over time might reflect an evolutionary
process of reasoning where particular feature values and semantic
relationships have become lexicalized. The characterization of feature
values and semantic relationships by Knott, Eulenberg, and Barker may
facilitate this type of diachronic analysis.
When I complete the first phase of my research (the initial digraph
analysis), I will provide notification of its availability. Then, when I
fully incorporate the analysis of features and semantic relationships, I
will post the data on the ACL SIGLEX Lexical Resources page
(http://www.clres.com/siglex.html).
I thank everyone who responded and hope that this summary responds to
those who asked to be kept informed of my findings.
Ken
-- Ken Litkowski TEL.: 301-926-5904 CL Research EMAIL: ken@clres.com 20239 Lea Pond Place Gaithersburg, MD 20879-1270 USA Home Page: http://www.clres.com