5. Conclusion

5.1 Current Situation

Today, Machine Translation is a useful tool if one does not simply try to replace human translators by computers. There exist useable commercial systems and there has already been large-scale success. But all these successes in MT were based on thorough analysis, practical evaluation of demands and thoughtful application of available means, not marketing descriptions of inferior products.

Yet, there is no system keeping the promise of instant, accurate translation from language X to language Y. Pre- and/or postprocessing is always neccessary and, depending on the purpose of the translation, should be done by an able translator.

To get a rough overview of the contents of a document in a foreign language, some products might suffice. But usually it makes more sense to follow the machine aided human translation approach, that is the human translates and uses the computer's facilities to look up words, idioms, synonyms etc.

For special tasks, human aided machine translation can be used to reduce costs. Again, sound analysis is neccessary as the texts to be translated must be suitable for MT -- just installing "some" software will not do it.

I agree with Jeanette Pugh:

From the human translator's perspective, MT has gradually become more of an ally than a threat.

[Pugh 1992, p. 18]

5.2 Perspectives

I only covered text-based MT here while translation covers more than that. There are promising projects, for example speech-to-speech translation in realtime like Verbmobil⁹, needing completely new techniques and provoking new approaches which might lead to improvements in text-based MT as well.

Another feature [of Verbmobil] is the presence of a special dialogue management module keeping track of the course of the dialogue. In several places, considerations about human interpreting strategies have found their way into the system modules. This tendency increases in the second project phase [...]

These characteristics make the Verbmobil a typical representative of the new tendencies in MT. It is, however, only one of several interesting examples. The efforts of many other researchers are going into the same direction [...]

[Hauenschild, Heizmann 1997, p. vii]

Research in MT continues. A lot of big problems are to be solved. I assume that for perfect MT we need to fully understand the human brain because language is the result of a very complex mental process. Linguistics hits several borders of human knowledge. Very fundamental questions are risen.

• How do we construct sentences?
• What is the link between language and thinking?
• How does the human brain resolve ambiguities?
• What is knowledge? Can it be appropiately represented by machines?
• What is meaning?
• Where does meaning come from?
• How do we think?

(Computer) Linguistics has already allied with Artificial Intelligence, Psychology and probably Neurobiologics as well. There have been experiments with neural networks which imitate the way the human brain works although they are very limited in capacity. There will probably be more attempts in that direction; we may end up simulating a whole human brain just to translate Ich bin drin into proper English.

Footnotes

...Verbmobil⁹

Verbmobil's first project phase was finished in 1996 with the presentation of a research prototype