From: AAAI-94 Proceedings. Copyright © 1994, AAAI (www.aaai.org). All rights reserved.
Processing Pragmatics
for Computer-Assisted
nstruct ion
Language
Keiko Horiguchi
Computational
Linguistics Program and Center for Machine
Carnegie Mellon University
5000 Forbes Avenue
Pittsburgh,
PA 15213
keiko@cs.cmu.edu
Computer-assisted
language instruction systems that
only perform syntactic
processing of input sentences
are not able to offer advice on pragmatic
aspects of
language use, and they cannot handle the variability
generally afforded by natural languages to express a
given propositional
content.
This paper describes a solution to this problem implemented as an extension to the existing Japanese tutorial system ALICE-than
(Evans & Levin 1993). We
to represent pragmatic infordesigned the p-structure
mation as well as the propositional
content of a senThe pragmatic
content is encoded in terms
tence.
of the speech situation,
the speaker’s attitude toward
the addressee, and the felicity conditions for the inLinguistic features that express
tended speech act.
the speaker’s uncertainty
are interpreted
as reducing
factors that weaken felicity conditions of the intended
speech act of the sentence.
We implemented a p-structure mapping program that
generates p-structures
from syntactic
structures.
As
an example, the p-structure
for the sentence Tegamiwo kaite itadakenai
darou ka to omou n desu ga (“I
wonder if you might be able to write a letter for me”)
is shown below.
SPEEC:H-AC:T
FEATIJRE
AC:TION
requesting-action
receive-favor-potential
ACTEE
+#Q
(tegami-wo)
SENSE “letter”
pbr
(kaite)
SENSE “write”
BELIEF
DESIRE
EXPECTATION
PLAC:EMENTOF-ADDRESEE
SPEECHSITUATION
1458
FEATURE
REDUCING-FAC!TOR
EXTENDEDPREDIC!ATE
C:ONJUNCTION
THINK
NEGATIVE
TENTATIVE
INTERROGATIVE
FAVOR
higher
formal
Student Abstracts
receive-favor-potential
/;-CT (n desu)
S (gal
L:,g 3 (toomou)
+
Eis 5 (darou)
~3%
(ka)
IziVh~
(itadakenai)
Translation
The system stores the pragmatic analysis template.
The error analysis matcher compares this template with
the student input.
It then reports any features that
are missing or different, as well as error features that
are inserted during the analysis. Based on this, the error matcher then formulates appropriate feedback. For
example, if a student used the verb sashiagerarenai
infor the above sentence, the system
stead of itadakenai
would respond as follows:
You seem to have used
the wrong direction.
the
verb
with
inward
the
giving
verb with
You
should
have
used
direction.
Our computer assisted instruction system benefits from
adopting p-structure
in two ways. First, the system allows students flexibility for expressing propositions
in
different ways, since the system can accept similar information
expressed in different structures.
Second,
the system is able to detect errors and give finer feedhack on pragmatic
usage of the language.
Students
can now express the required proposition
more freely
without burdening the teacher with the task of typing
in all possible correct answers and incorrect answers
with appropriate feedback.
Acknowledgments:
I am grateful to Lori Levin,
David Evans, Martin Thurn, and Steve Handerson for
their help and guidance.
References
AIIen, J. F. 1983.
Recognizing
intentions
from natural
language utterances.
In Brady, M., and Berwick, R., eds.,
Computational
Models of Discourse, 107-166.
Cambridge,
MA: MIT Press.
Evans, D. A., and Levin, L. S. 1993. Intelligent computerasisted language learning theory and practice
in ALICEthan. In Army Resecarch Institute
Workshop
on Advanced
Technologies
for Language
Learning.
Kogure, K.; Iida, H.; Yoshimoto,
K.; Maeda, H.; Kume,
M.; and Kato, S. 1988. A method of analyzing Japanese
speech act types. In Prceedings
of the 2nd International
Conference
on Theoretical
Methodological
Issues
in Machine Translation
of Natural
Languages.