Electronic Courses: towards Unified Curricula in Language and Speech Technology Education

This paper is a plea for concerted effort towards preparing unified curricula in Language and Speech Technology (L&ST) education. To be maximally effective, the text courses should be translated in student’s native languages and should be accompanied by easily accessible NLP demonstrators. The software companions, demonstrating the basic concepts and allowing for simple experiments could be selected from the public domain and organised as a coherent toolbox. A possible scenario for this global aim is discussed and a couple of available educational programs in the area of Language and Speech Technology are briefly presented. 1. AIMS AND MOTIVATION In an area with such a dynamics as Language & Speech Technology (L&ST), keeping up with the latest trends is of paramount importance. However, given the broad spectrum of problems dealt with by this domain, in general, students taking courses in L&ST are inherently biased and limited to the areas that are within the coverage and interest of the instructor. Where the courses are accompanied by hands-on laboratories, the subjective factors are augmented by the objective ones: lack of software that could be used in broad education. From the point of view of the instructor, keeping up his lecture notes and searching for appropriate software that could exemplify the theoretical presentations are in many cases too cumbersome, so most of them stick with some (presumably) good text-books and limit topic demos (if available) to those for which appropriate software is at hand (frequently, in-house programs or specialised systems related to projects the instructor is involved with). On many web sites one could find various texts meant to be used as support for courses in L&ST and some of them are very good. Investing some time, one could also find several pieces of software that could be used in demonstrating basic ideas and approaches and also main applications of the current issues in LS&T. That is to say that the building blocks are there, what is needed is a team of experts to assemble them in a coherent educational construct. This idea has been suggested, with respect to other goals, under several metaphors by various researchers: “lingware shelf” (Zock) and “lego approach” (Veronis) to name just two of them. Within an environment such as ELSNET, with its tremendous potential both in academic and industrial activity, it would be easier and it would take less time to develop a balanced curriculum for L&ST with texts and associated demo systems. Contributed by several researchers from both academia and industry, and distributed world-wide via the net, such an educational pool would have a highly beneficial impact on the development of the domain: • the educational process would better follow the mainstream of research and development; • international cooperation, mainly the one addressing multilinguality, would be significantly facilitated by relying on a common view of the domain; • difficulties in dissemination due to the language barriers, could be partially overcome by translating the curriculum in all European languages. The last point is particularly relevant and given the wide European coverage of ELSNET it is perfectly feasible. Among other benefits, such an endeavour would provide a basis for terminology unification (especially in the case of Central and East European countries). Coming to this point I cannot refrain from dreaming of an electronic encyclopaedic dictionary of L&ST (WordNet like), translated in all European languages. ELSNET is not only bridging Language and Speech communities but also Western and Central European traditions and schools. Most of the multilingual projects that were lately carried within the INCO-Copernicus program have shown that this gap was larger than it was believed (not thematically, but mainly from the methodological point of view). And it suffices to say that none of the standards or standardisation proposals I know of in the area of language resources (EAGLES, PAROLE, MULTEXT etc.) developed in Western Europe, could be fully applicable to all languages of Central and Eastern European Countries. Before concluding this section, I would like to make another suggestion. With the advent of the multilingual applications and new language markets, more and more requests on assistance with language issues can be seen on the net. Coming from various users, ranging from Ph.D. students and individual researchers to industrial companies, these requests vary in complexity and formal Proceedings of the International Conference “ELSNET in Wonderland”, Soesterberg, 25-27 March, 1998 aspects. Sometimes the requests reach the right persons, sometimes not. Establishing an ELSNET multilingual electronic consultancy desk acting as a dispatcher for such services might be a solution in the benefit of the international community. 2. A POSSIBLE SCENARIO This section will hypothetically assume an established consortium, with a large spectrum of expertise in language and speech, native speakers of the European target languages, reliable links between experts, consortium commitment and necessary funds for doing the work. Most of the assumptions above are obviously met by ELSNET association. A preliminary phase in the project would be to investigate the current status in Europe as far as L&ST teaching is concerned. Similar actions were taken in the past years, mainly under the ACL auspices [1], [2], [3] so that good models are available. Given that the above mentioned surveys were concerned only with computational linguistics, and more than 4 years passed since the last one, the questionnaire has to be updated and extended. Considering the audience of the courses, their duration and based on the answers provided by respondents an near ideal curriculum could be sketched. What I mean by an ideal curriculum is a collection of topics that would maximally cover the area of L&ST from both academic and industrial perspectives. The topics should be arranged in a dependency network, so that prerequisites for one subject be easily identifiable. Certainly an ideal curriculum, is unteachable (at least given the limitations on duration and available competence) and it is just a framework for customising a realistic instruction program aligned to the state of the art. In principle, two main categories of students with interest in L&ST can be envisaged: students in hard sciences (computer science, mathematics, engineering etc.) and students in humanities (linguistics, cognitive sciences, literature, journalism, language teaching, psychology etc.). Given different backgrounds of these two student categories (the differentiating could be certainly finer grained) seems natural to consider different perspectives, emphasis and approaches. According to [3] the average duration for a computational linguistics course is one term (but two, three or even 4 terms are not rare cases) . Depending on this factor, the number of topics (in a coherent dependency) and the level of details for the selected topics would vary from one site to the other. The second phase would be gathering of relevant texts on the topics included into the ideal curriculum and finding experts to translate the texts in each of the considered languages (ideally, all languages in Europe) and make these texts available on the net. A third phase (potentially in parallel with the second one) should consider collecting demonstration software for as much as possible of the topics included into the ideal curriculum. This work should rely on several public software registries and/or contributions from the consortium. Given the variety of hardware and software platforms, the collection of teaching support software should be organised not only thematically, but also on specific requirements criteria. To keep track of the various options in providing hands-on training on specific subjects of the curriculum, the consortium should provide and maintain an on-line data-base, whose organisation would allow answering simple questions like: “give me details on the tools for demonstrating chart parsing” and providing basic information like the name of the programs, the platform they runs on, associated demos and so on. 3. EDUCATIONAL LINGWARE In this section I will briefly present some programs we or our Romanian colleagues developed (some of them in cooperation with other partners), which we consider to be useful for educational purposes (but not only) and which are freely available on request. The Appendix contains snapshots of interactions with the systems described. It should be mentioned that the systems we describe here are just a very small set of public software which could be found in several archives. Even the Romanian systems are more numerous, but given the space limitation we will restrict ourselves to presentations of EGLU, PAIL, GULIVER, KRIL, HEADGEN and PROSODICS. Occasionally we will mention some other systems that could be used to support specific topics covered by the above-mentioned systems.