A Future with Language Technology
Human language technology is expected to permeate all areas of our life in the future (HOPE, 2002). Progress in computational linguistics and human language technology will not only lead to an omnipresence of language processing chips in clothes, offices, houses, shops and other locations, it also opens new perspectives in politics, society, economics, or education.
In order to sound out estimations and expectations a short online-questioning (Haenelt, 2002) has been run with language processing professionals, students and educated people from other professions. Since the questioning was run for a very short time only and the participants acceded more or less randomly, this study does not claim to be representative. The questioning focused on perspectives that are opened by computational linguistics and human language technology. Four examples from politics, society, education and science shall be provided in this article.
The most focused question with respect to politics was, whether the participants could imagine a machine which mediates in the conflicts between peoples. No one doubted the technical feasibility of such a machine. This probably results from the fact, that in these days even in the field of human language technology partial and stepwise improved solutions are becoming similarly accepted as in other technological fields. So solutions are not necessarily expected to work fully automatically and without any human participation, but rather systems can be imagined which could structure a discussion, collect the goals of different peoples, support the comparison of these goals, access background information such as laws and other cultural and historical conventions, and could support the finding of solutions.
All participants ranked such a machine as a highly desirable goal. But sadly enough, most contributors doubted the political will for using such a machine.
“Imagine your car can explain the repair it needs”. This is the type of question that is typical in these days for characterizing the potential of language processing (cf. HOPE, 2002). All participants could imagine this, and all participants found this highly desirable.
More and more language technology products appear in the market. Some are accepted readily – imperfect as some of them are, since there is no alternative (such as for instance search engines), others compete with alternative possibilities (such as dictation systems vs. keyboards). After a first phase of aiming at fully automatic high quality processing as the (more or less) only imaginable marketable state of the art in the sixties of the last century (ALPAC, 1966) human language technology has gained some experience in which “gears” and “clutch pedals” are accepted by the market, and some dialogue with the market and training has developed some experience of customers.
Examples of products which can be offered with restricted language processing capabilities are
- voice control of everyday machines (operating of telephones, ordering of cinema tickets),
- “programming” of everyday life systems: video recorder, telephone systems, house technology
- dictation systems
- dialogue systems, (telephone) information systems, call centres
- text processing with spelling and grammar correction
- translation systems
- (internet) search engines
Currently, however, many of these systems are more or less systems for “experts”, because their application requires some expertise for filling the gap between the processing provided and the processing needed. Examples of such expertise are
- speaker adaptation, dictionary updates, grammar updates, formal knowledge modelling, or providing examples and training of rules,
- knowledge of the technology for estimating the usability and for making optimal use of systems:
- knowledge of information retrieval methods in order to formulate an optimal search query,
- knowledge of parsing / generation / translation technology in order to determine or even formulate texts which are best suited for machine translation or to understand processing errors and perform pre- or post-editing efficiently)
- realistic estimation of error rates and reliability in order to determine possible application areas.
All participants of the questioning appreciated the support provided by the products which are available now and which make many tasks easier, but they also wished a better command of language of the systems in order to be able to concentrate on their communicative goals, rather than to have to deal with matters of the system. Clearly, the goal is the development of more advanced and more comfortable and broader applicable systems. The more progress will be made towards successfully managing everyday language, the more human language technology will become an everyday life technology.
All participants agreed that interdisciplinary work will become more and more important in future, and that more training in cross-method thinking will be required in education.
Computational linguistics and language technology already comprise approaches from linguistics, from mathematics, from engineering and other disciplines. Based on this interdisciplinary experience this science and technology can offer support to forward interdisciplinary thinking in school education: They can contribute educational topics and methods which provide an excellent opportunity to overcome a split of pupils into those which are “linguistically gifted” and those which are “mathematically gifted” in favour of pupils which are well-educated.
For a better understanding and integration of linguistics and mathematics it could be useful to cross the boundaries and to teach mathematics with examples from language technology and linguistics with examples from mathematics. Applications like part-of-speech-tagging with Hidden Markov Models or text retrieval with vector models are well suited for giving linguistically advanced pupils a deeper understanding of probability theory and vector algebra, and mathematically advanced pupils a deeper understanding of parts of speech, sentence structures, and text and content constitution. Integrating such applications into teaching also allows for combining example-based learning and theory-based learning, and thus also broadens the basis for understanding the subject. In addition, the applications are very good examples of the applicability of linguistics and mathematics beyond class tests and exam papers, and for developing methods by crossing the boundaries of individual disciplines. The learning process can further be supported by applying the knowledge to the development of small programs. There are applications for which basic versions can be implemented with fairly simple programs (cf. Haenelt, 2001).
In the online discussion not only computational linguists but also chemists, lawyers, and mathematicians unanimously agreed that they would appreciate examples from language technology in mathematics lessons.
Currently human language is the most simple and flexible means of knowledge and information handling. It allows for the constitution of information, for its presentation, memorization and for its communication. It can be very economically adapted to specific communication purposes. Linguistic communication succeeds on the basis of a certain breadth and depth of variation and vagueness. Language enables humans to think and thus to access and integrate heterogeneous and incomplete information as well as to develop knowledge and information in the progress of time.
The ability to model these properties and functions will be essential for the development of more flexible natural language systems. Although much progress has been made, currently there is no consistent comprehensive technology for the development of corresponding systems, and even research on individual components has not been completed. Therefore research towards this goal is still a complex undertaking. It begins with exploring and modelling language mechanisms which serve to constitute, to communicate and to understand concepts. It includes pragmatic conditions and cognitive processes, and at an even higher level of complexity it also ranges to physical and chemical processes in the brain.
With respect to time perspectives participants think that more progress will be made in the area of semantic processing within the near future. With respect to cognitive processes one participant wrote: “In fifty years we will be even more astonished how complex, unique and estimable humans are, since we still will not have succeeded in understanding cognitive processes, not to mention to imitate them.” Another participant answered with respect to biotechnological perspectives: “Currently we know at most one thousandth of how the brain works. If we aim at an electronic system which also communicates with the brain (for medical purposes for instance), we additionally need a computer technology in nm- and pm-dimensions.” The person believes that it will take at least another hundred years to investigate the processes in the brain. The person currently regards the development of a speaking computer fitted with some intelligence and emotions which can manoeuvre a spaceship as more feasible. This, however, should have existed in 2001 already according to the film “Odyssey 2001”.
As we have seen in the last years, human language technology has made much progress, and the participants believed that progress has also been made with respect to an estimation of the tasks to be solved, of their complexity and their application breadth, in the research community as well as in industrial development and in the market and with funding authorities. They believe that this will lead to a realistic estimation and better allocation of the resources needed for research and development in this area. Language technology has been estimated a multi-billion dollar industry (Spiegel, 1999). This is the more promising considering that we are only at the beginning of its scientific, technologic and economic development.
ALPAC (1966): Language and Machines. Computers in Translation and Linguistics. (ALPAC report, 1966). National Academy of Sciences.
Haenelt, Karin (2002): Mit Computerlinguistik und Sprachtechnologie in die Zukunft. Eine Einladung zum Querdenken und Mitreden. Diskussion. Fragebogen. Literatur. Links. - <http://kontext.fraunhofer.de/haenelt/hlt/CL-Futur2002/>
Haenelt, Karin (2001): Beispielprogramm zur Berechnung der Ähnlichkeit zwischen Anfrage- und Dokumentvektoren nach der Cosinus-Formel 19.12.2001. - <http://kontext.fraunhofer.de/haenelt/kurs/folien/IRVektorBerechnung.html>
HOPE (2002): Das europäische Projekt zur Förderung sprachtechnologischer Anwendungen: HOPE/EUROMAP 3 <www.bit.ac.at/4_fue_5rp_2tp_euromap_info.htm>
Spiegel (1999): Der Spiegel. Nr. 15, 12.04.1999, S. 124-127.