A constraint-satisfaction approach to dependency translation

Constraints

Associate word in one language with word in another and optionally one or more associated dependents of the words.

[en=[lex=cook, subj=?es, dirobj=?eo],
 sw=[lex=pik-, subj=?ss, dirobj=?so],
 dep={[en=?es, sw=?ss], [en=?eo, sw=?so]}]

[en=[lex=boy], sw=[lex=mvulana]]

[en=[pos=N, relmod=?er], sw=[pos=N, relmod=?sr],
 dep={[en=?er, sw=?sr]}]

[en=[lex=be, predadj=[lex=enough]],
 sw=[lex=tosh-]]

[en=[pos=VT, +rel, dirobj=[lex=that]],
 sw=[pos=VT, rel=[cls=?rc], obj=[cls=?rc]]]

Parse source language sentence into dependency tree.
For each word, access all translation entries for target language, and for each create a candidate word in the target language. If there is an associated dependency, create a dependency and a place holder for the dependent.
Follow the procedure for generation, except that goodness also depends on translation constraints. translation constraints.