Earley-based structural transfer for Apertium

Perhaps Earley's algorithm to parse context-free grammars (which has a left-to-right longest-match philosophy as Apertium) could be used to perform more complex syntactical transformations; this could be useful for distant language pairs containing embedded structures.

Open questions

• Currently, Apertium uses text streams to communicate. I assume this would not be possible here.
• When would one call the bilingual dictionary? Apertium Level 2 calls it in the first stage.
• We should check whether this has been done before.
• In case there is more than one parse of a sentence, there should be a way to select the most likely.

Existing parsers

Current free-software parsers which might be worth looking at:

• AGFL parser (GPL)

Further reading

• Koichi Takeda Pattern-Based Context-Free Grammars for Machine Translation

This paper proposes the use of "pattern-based" context-free grammars as a basis for building machine translation (MT) systems.

• Randall Sharp and Oliver Streiter Simplifying the Complexity of Machine Translation