Human Language Technology can modernize writing and grammar instruction 
Gerard Kempen 
University of Leiden 
P.O. Box 9555, 2300 RB Leiden, The Netherlands 
kempen @rulfsw.leidenuniv.nl 
The recently published Survey of the State of the 
Art in Human Language Technology does not 
spend a single word on computer-aided language 
learning (CALL). Indeed, present-day CALL 
systems hardly employ Natural Language 
Processing (NLP) techniques. Reasons for this 
state of affairs are not hard to find. First of all, 
current language teaching methods tend to 
emphasize oral language skills, especially in 
second-language instruction. But automatic 
recognition of speech, in particular speech by 
non-natives, has only taken its first steps outside 
the laboratory; and many language teachers still 
judge synthesized speech of insufficient quality 
to serve as a model for language learners. 
Secondly, modern language pedagogy stresses 
communicative success rather than formal 
correctness. This, too, works against the 
profitable deployment of NLP tools in CALL 
because automatically generating non-trivial, 
communicatively interesting and instructive 
dialogues does not yet seem within reach of NLP 
technology, let alone the evaluation of student 
responses from the point of view of successful 
interpersonal communication. More congenial 
with these priorities were multimedia 
innovations. If anything has brought about a 
metamorphosis in second-language teaching 
practices, it was the introduction of affordable 
video, audio and other graphical and acoustic 
tools that, under the control of flexible software, 
can create an illusion of 'total immersion'--the 
supposedly ideal language learning situation. 
However, language proficiency includes more 
than conversational skills alone. Equally 
important are writing skills: orthography, 
formulating well-formed sentences, composing 
clear and well-organized texts. It is in the area of 
written language instruction that NLP 
technology can find a wealth of extremely 
valuable applications. Teaching first- and 
second-language writing skills is very labor- 
intensive because teachers need to mark large 
numbers of test papers. NLP software holds the 
potential of alleviating this burden considerably, 
and even of outperforming teachers in the speed 
and quality of feedback to learners, and in the 
capability of generating well-targeted and 
attractive exercises. 
A particularly important reason why human 
language technology should begin to take 
written hmguage instruction seriously, derives 
from the following argument. Many instructional 
scientists subscribe to the view that language 
skills are best acquired in a situation similar to 
that of children learning their mother tongne. 
This explains not only the bias in favor of oral 
and conversational language skills in current 
hmguagc pedagogy but also the reluctance to 
work with explicit grammar rules. The negative 
attitude toward grammar is strengthened by the 
generally disappointing outcomes of grammar- 
based language teaching methods. 
However, these negative results may have a 
completely different origin. Although the 
reasoning that language acquisition can do 
without explicit rules may hold for oral 
language proficiency, there is no evidence that it 
generalizes to the acquisition of written language 
skills. If writing skills do require the application 
of explicit orthographic, morphological, 
syntactic, etc., rules by the learner, then anti- 
grammar attitudes must be detrimental. Learners 
will be deprived of knowledge that in fact is 
essential to solving writing problems. More 
serious is the ensuing lack of interest in the 
improvement of grammar instruction methods. 
Taught by age-old methods, many learners only 
have an inkling of the meaning of important 
grammatical concepts. Grammar rules 
referencing these terms are hard to apply 
successfully in written composition. 
The response by most linguists and 
instructional scientists to this state of affairs has 
been misguided. Instead of initiating research 
into improved grammar teaching methods, they 
have tended to play down the importance of 
grammar rules and linguistic awareness in 
learning how to write. 
1005 
Modern NLP techniques for the analysis and 
generation of written language, in combination 
with graphical tools for visualizing and 
manipulating the structure of words, sentences 
and texts, afford excellent possibilities for 
creating integrated curricula for grammar and 
writing instruction. Language engineers, take up 
the challenge, seize the opportunity! 
1006 
