| Age | Commit message (Collapse) | Author |
|---|
|
prediction
|
|
|
|
|
|
|
|
|
|
|
|
write Haskell bindings
|
|
|
|
completely loaded
|
|
|
|
function is linearizable in a given language. This is used in the Android UI for better vizualizations
|
|
write for instance 'ab.c' and then everything between the quites is identifier. This includes Unicode characters and non-ASCII symbols. This is useful for automatically generated GF grammars.
|
|
|
|
|
|
|
|
|
|
|
|
runtime from messing up the content.
|
|
|
|
some sentences
|
|
between ordinary tokens. It is also used in the English RGL to attach the commas to the previous word.
|
|
loading. The parser still seems to be 8 times slower than before I introduced the BIND. At least the Android UI now feels a lot more comfortable
|
|
|
|
|
|
later is now retrained. Once the grammar is compiled with the .probs file now it doesn't need anything more to do robust parsing. The robustness itself is controlled by the flags 'heuristic_search_factor', 'meta_prob' and 'meta_token_prob' in ParseEngAbs.gf
|
|
|
|
|
|
yet but HOAS is implemented. The API is accessible from Python as well
|
|
|
|
order to test it properly
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
support BIND. The following things are still broken: parseval, word completion, handling 'pre', the robust mode
|
|
|
|
|
|
|
|
|
|
PgfSymbolKS, PgfExprFun and PgfLiteralStr now keep their strings as embedded flexible arrays. The latest change gives us the same compactness as the old representation but it is a lot easier to use.
|
