In the previous chapters, we have tried to give an idea of what is currently possible in MT. In this chapter, we look to the future. Our aim is to give a flavour of current research in MT, indicating what issues are receiving attention and what techniques are thought to be promising.
Of course, not all the ideas that are currently important are really
new ones. A great deal of current research is directed at how
familiar techniques can be improved --- for example, how standard
`Linguistic Knowledge' approaches can be improved by using better
linguistic analyses (analyses based on better linguistic theories, or
a better understanding of existing theories), and developing or
adapting more efficient processing methods, and better tools for use
in constructing and modifying systems. Likewise, an important feature
of current research involves work on sublanguage MT
(cf.
Chapter 
), but though the design of tools to aid
sublanguage analysis is an increasingly important area, it is hardly a
new development. Other currently important work is concerned
with integration, which can relate either to the integration
of MT with other Natural Language Processing technologies, or
to the (non-trivial) problems of integration of MT into general
document processing technology that arise as one tries to make a
practically and commercially usable system out of a research prototype
MT system. A particularly important example of the former is research
on `speech-to-speech' MT systems --- that is, systems that can take
spoken input, and produce spoken output (e.g. for more or less
simultaneous interpreting of telephone conversations). Such work is
clearly important, and often throws up interesting differences of
emphasis (for example, in speech-to-speech work, there is an emphasis
on speed, and on dealing with sentence fragments, since one would like
to be able to translate each utterance as it is spoken, without
waiting for the end. This gives importance to `bottom up' methods of
analysis, and severe restrictions on the input in terms of text-type ,
etc). However, there is an obvious sense in which such work it is
`more of the same' --- it involves improving one aspect of an existing
idea, rather than presenting a genuinely new direction, and would be
accessible on the basis of the earlier chapters of this book. In
this chapter, we will concentrate on what we think may turn out to be
more radical ideas.
The chapter has three main sections. In Section , we
outline some current issues and trends in the design of sets of
linguistic rules for MT, that is, work within the established
`Linguistic Knowledge', or `Rule-Based' paradigm. The next section
() gives an overview of some of the corpus and machine
readable dictionary resources which have recently become available.
These resources have stimulated a great deal of research within the
traditional LK/rule-based paradigm, and have also been of key
importance in the trend towards so-called empirical approaches
to MT, which are sketched in Section .