GeneralizeDisambiguatr

treelet: a portion of a dependency tree. (learn what that is, it's like a subtree)

what are examples of phrases that need disambiguation? Check out the Carpuat and Wu PSD paper again.

cross-language ambiguity in parsing? What does that even mean? prepositional attachment ambiguity is really an English problem -- but it would produce different text in a target language.

In Malt parser, what kinds of features can you put into it?

• can you get probabilities out?
• how do you divide up the different kinds of situations, where you know which classifier to ask? are the output classes just SHIFT or REDUCE?

how do you know when you've got a case where different parses would give the same target-language sentence? (eg, english->spanish "the boy saw the girl on the hill with the telescope.") HOW DO YOU KNOW WHEN THE AMBIGUITY MATTERS?

allowing a parser to accept incomplete parses?

remain silent about the different ambiguities in the source language... you may not need to represent them.

3 kinds of ambiguity:

• real actual ambiguity
• disambiguated by the structure of the sentence
• disambiguated by the context. (prefer this over this)

we need to read about tree transducers because we haven't and we didn't know if it was equivalent to XDG.

consider:

• chunk
• try to translate
• if necessary, parse
• then try to translate again

disambiguatr for other MT systems

• could it be integrated into Apertium? They used to have a WSD module; how does this compare to CL-WSD? The really interesting point here is that the cross-language variety takes word senses to be ''different word types in the target language''.
• http://wiki.apertium.org/wiki/Word-sense_disambiguation
• http://wiki.apertium.org/wiki/Lextor
• http://wiki.apertium.org/wiki/How_to_get_started_with_lexical_selection_rules
• http://wiki.apertium.org/wiki/Lexical_selection_in_target_language