Our discussion of flexible translation between Swedish and Norwegian is based on unpublished work by [Dyvik1992]. The standard references on sign-based approaches to linguistic representation are [Pollard and Sag1987,Pollard and Sag1993]. The view of constraint based translation that we describe is loosely modelled on that used in `Shake and Bake' [Whitelock1992,Beaven1992]. See [Kaplan et al. 1989], [Sadler1991] and [Sadler1993] for a slightly different approach. General discussion of how multi-dimensional representations can be used in MT can be found in [Sadler and Arnold1993].
On knowledge-based MT see [Goodman and Nirenburg1991], and the special issue of the journal Machine Translation, [Goodman1989].
On the processing of corpora, and their use in linguistics generally, see [Garside et al. 1987], and [Aijmer and Altenberg1991].
The idea of example-based MT was first discussed in a paper by Nagao [Nagao1984]. For a review of more recent work along these lines, see [Somers1992].
The pure statistical approach to MT is based on the work of a team at IBM , see for example [Brown et al. 1990]. As regards aligned, bilingual corpora, the most common form of alignment takes the sentence to be the organizing unit for chunking, see [Brown et al. 1991] and [Gale and Church1991b] for relevant discussion. On automatic extraction of word correspondences across bitext, see [Gale and Church1991a]. Techniques involving the use of corpus resources for automatic sense disambiguation have also been explored within the DLT project, see [Sadler1989].
The translation of no, which was described around page above, is discussed by [Sumita et al. 1990]. The discussion of o-negaishimasu is from [Furuse and Iida1992b], see also [Furuse and Iida1992a], and [Sumita and Iida1991].
The frame for computer on page above is taken from [Goodman and Nirenburg1991, page 25,].
For up to date reports on research in the field of MT, there are several journals, and several major international conferences. The specialist Journal is Machine Translation, edited by Sergei Nirenburg, from Carnegie Mellon University in Pittsburg, USA, and published by Kluwer Academic Publishers. However, the journal Computational Linguistics, published by the MIT Press for the Association for Computational Linguistics (ACL), also publishes research which is directly about MT.
The specialist conference for research on MT is called TMI ---
for `Theoretical and Methodological Issues (in Machine Translation)'.
This has been held every two years since 1986, and proceedings are
published ([TMI11985],[TMI21988][TMI31990],[TMI41992]). Many of
the papers in the last
of these are directly or indirectly about the issue of `rationalist'
(i.e. rule-based ) vs. empirical approaches to MT.
The proceedings of
the main Computational Linguistics conferences, namely ( COLING),
the conferences of the Association for Computational Linguistics
(ACL) and the conferences of the European Chapters of the ACL, also
contain a high percentage of papers about MT. ACL conferences are held
annually in the USA (for example, [ACL281990,ACL291991,ACL301992]). The EACL
conferences are held biennially,
[EACL11983,EACL21985,EACL31987,EACL41989,EACL51991], as is COLING:
Coling 84 [Coling841984] was held in Stanford, California,
COLING 86 [Coling861986] in
Bonn, Coling 88
Budapest, Coling 90 [Coling901990] in Helsinki, and Coling 92 [Coling921992] was held in Nantes.