Yes! NLP-based FL-ITS will be Important
Henry Hamburger
Department of Computer Science
George Mason University
Fairfax VA 22030 USA
henryh@cs.gmu.edu
In this note, a straw man is destroyed, optimism is
expressed, an existing system is sketched, and some
issues are laid out.
A Direct Approach.
It is useful to consider a straightforward argument
for according NLP a central role in CALL. I will
call it SCW, since it is simple, clear and wrong. It
runs like this: intelligent tutoring systems (ITS)
with simulated problem environments are
potentially excellent for learning; a key module of
an ITS is a computational model of expertise in the
domain; NLP systems are such models in the
domain of hmnan languages; so let's use NLP
knowledge bases as the expert module of a foreign
language ITS.
see Chanier et al. (1992). For a rule to be learned or
even to be useful in explanation, it should be
congruent with learners' cognition and expressed in
a way that is meaningful to them. Anyone familiar
with NI,P knowledge bases could hardly think
seriously about promoting their direct use in
explanations to learners.
Stay the Course.
The lailure of SCW certainly does not mean we
should all go home. On the contrary, I strongly
believe that valuable ITSs will be built and that
they will use NLP, or I would not spend so much of
my time on one. It does, however, suggest that it
will take significant work, in design as well as
implementation, to create an NLP-driven ITS for
language that engages students and helps ttlem
learn efficiently.
But Language is Different.
But language use is different fl'om most ITS
domains, in a way that invalidates SCW. In a
typical domain a student shows progress by success
in explicit stepwise reasoning to a solution.
Successful language use, in contrast both to other
ITS domains and to the ability to state grammatical
rules, need not demand articulating the stepwise
reasoning about sentence construction, but it does
require the capacity to arrive quickly at a result
(either a sentence that expresses one's current
thought or the meaning of someone else's
sentence), and moreover to do so while thinking
about something else, namely the substance of the
conversation. Such a capacity may well imply a
grammar in the brain but not awareness and
articulation of it.
Expertise and ITS.
Even for explicit teaching of grammatical, lexical
or other knowledge of a particular language, SCW
is flawed. Experience teaches that a performanceoriented
representation of domain knowledge may
be tmsuitable tbr direct use in an ITS; see Ciancey
(1987) and in the world of language specifically,
Let Many Flowers Bloom.
Proclaiming a future for NLP in ITS is not to deny
benefits from other kinds of devices or the
continuing importance of hmnan teachers, tutors
and other conversational partners. Indeed N1,P
itself may have variegated ways of contributing to
the mix of learning resources, by systems that use
different modules of NLP and that aim to foster
various kinds of language knowledge in the
student. Automated aid has been undertaken lot
parts of languages all the way from spelling,
pronunciation and morphology, through syntax and
semantics, to discourse and cultural knowledge. For
the most part, the NLP-based work has been at the
low end, rarely going above syntax.
Two-Medium Conversation.
A sketch of our own (Hamburger, 1995) twomedium
conversational system will give a concrete
sense of one approach to NLP-intensive CALL.
FLUENT-2 is a language learning and tutoring
system, helping students learn and letting teachers
influence what is learned. It has features of both
intelligent tutoring systems and microworld
learning environments. Using a pedagogical
1007
strategy of situational immersion, the system
engages the student in meaningful multi-media
communicative acts in graphically depicted real
world situations.
The student interacts with the system by direct use
of the target language. The system uses Felshin's
(1993) multilingual NLP system, which can expose
students to a wide variety of linguistic forms. The
intent is that new words, phrases and grammatical
usage will become comprehensible through
meaningful exposure and use. The system makes
this possible by tightly coupling the language to
graphical acts and system generated animations
within a realistic ongoing situation. Actions
available to both the student and the system include
selecting and moving objects and making human
figures walk, turn, point, grasp and release objects,
and so on.
The teacher interacts with the system through
graphical (GUI) tools that facilitate the designing of
exercises and the construction of appropriate
microworlds. These tools let a teacher construct
exercises that invoke specific linguistic concepts in
the target language without having to deal directly
with the NLP system. One parameter of an exercise
can be a plan for a goal in a situation, a capacity
that makes exercises portable across microworlds.
Real language teachers have given design advice
and now use the system.
Three Issues.
Fidelity and interactiveness are key issues tbr ITS.
Extensibility has been a key issue in NLP. All three
will be important for NLP-based CALL. I'll
subdivide each, mention some interplay among
them and comment on Fluent-2 in light of them.
Fidelity is the accuracy of a presentation. For
example the visual fidelity of a photo or video
exceeds that of simple graphics or animation. A
technical drawing may have good conceptual
fidelity if it connects related concepts. Two aspects
of fidelity that may rightly concern language
educators are cultural authenticity and the
situational continuity of a conversation.
Interactiveness of an ITS can include its immediate
responsiveness by faithfully updating the visible
situation after a student makes some kind of move
as well as longer-range responsiveness to an
individual student, based on a model that it builds
of that student's knowledge. In addition, the student
may be offered control over parameters of the
systems behavior, including subject matter,
difficulty and style. We have used this last
approach, in response to arguments made by Self
(1988).
A system can be designed to be extendable within a
language. It takes extra effort to make it possible
for a teacher to do so, as opposed to a programmer.
Portability across languages is familiar to NLP
researchers, and, as noted above, portability can
also refer to moving exercise types into new
situations. Schoelles and Hamburger (1996) show
how this capacity lets one present a language
concept in one situation and test it in another.
There are tradeoffs here. As an example, Murray's
(1995) video system achieves exceptionally fine
cultural fidelity, but for that reason little of it is
language portable. Interactiveness also is difficult
to achieve with video, since there is a finite amount
of video material produced in advance. Our
generative, recombinative animation approach does
not encounter this constraint.
In the End.
It will not be easy, but I hope and believe that
before too long there will be a variety of exciting,
effective NLP-based CALL systems.
References
Chanier, T., Pengelly, M., Twidale, M. and Self, J.
(1992) Conceptual modelling in error analysis in
computer-assisted language learning systems. In
Swartz, M.L. and Yazdani, M. (Eds.) Intelligent
Tutoring Systems for Foreign Language Learning.
Berlin: Springer-Verlag
Clancey, W. (1987) Knowledge-Based Tutoring:
The GUIDON Program. Cambridge, MA, USA:
MIT Press.
Felshin, S. (1993) A Guide to the Athena Language
Learning Project Natural Language Processing
Systems. MIT.
Hamburger, H. (1995) Tutorial tools for language
learning by two-medium dialogue. In Holland,
V.M., Kaplan, J.D. and Sams, M.R. (Eds.)
Intelligent Language Tutors. Mahwah, NJ, USA:
Erlbaum.
Murray, J. (1995) Lessons learned from the Athena
Language Learning Project. In Holland, V.M.,
Kaplan, J.D. and Sams, M.R. (Eds.) Intelligent
Language Tutors, Mahwah, NJ, USA: Erlbaum.
1008
Schoelles, M. and Hamburger, H. (1996) Teacherusable
exercise design tools. Montreal: Proceedings
of ITS-96.
Self, J.A. (1988) Bypassing the intractable problem
of student modelling. Montreal: ITS-88.
1009


source: http://acl.ldc.upenn.edu/C/C96/C96-2173.pdf