/novamente - Novamente work
query debugging notes how did he do that ?
LEFT-WALL how did he do that ?
Object (do[4], that[5])
Subject (do[4], he[3])
Tense (do[4], past_infinitive)
Number (that[5], uncountable)
Number (he[3], singular)
Quantification (he[3], Definite)
[head [name <>
tense <>
links [_subj [name <>
DEFINITE-FLAG <>
noun_number <>]
_obj [name <>
noun_number <>]
how [name <<_$qVar>>
TRUTH-QUERY-FLAG <>]]
HYP <>]]
uncountable(that)[1]
past_infinitive(do)[1]
hyp(do)[1]
_obj(do, that)[1]
singular(he)[1]
definite(he)[1]
how(do, _$qVar)[1]
_subj(do, he)[1]
truth-query(_$qVar)[1]
should be .
how-query(_$qVar)[1]
|
start implosions in earnest
And .. I'm done. Just sent out email announcing this.
The last step here is to document what's going on on the wiki page,
go through a review cycle, and polish up any rough edges, or make
any desired changes.
|
Completion of the Pen tree-bank stye markup.
Finished with the Penn tree-bank style phrase structure markup. Checked into both
the disco tree, and also into my experimental branch on the novaente servers.
From the file documentation:
/**
* The PhraseMarkup class adds Penn tree-bank style phrase structure
* markup to the relex ParsedSentence/FeatureNode graph.
*
* An example Penn tree phrase structure is
* (S (NP I) (VP am (NP a big robot)) .)
*
* The head of the phrase structure is always given by "phr-head".
* This feature node will always contain the feature node
* "phr-next": pointer to the next feature node.
*
* It might also contain any one (and only one) of the following
* three feature nodes:
* "phr-type": can be S, VP, NP, ADJP, etc., or absent
* "phr-word": pointer to the feature node for the word.
* "phr-head": pointer to a subphrase. So, for example,
* (NP a big robot) is a sub-sub-phrase of the whole.
*
* The mapping is such that CDR is always given by "phr-next",
* while CAR is either "phr-type", "phr-word" or "phr-head".
* Here, CAR and CDR are the standard LISP pair primitves.
*/
|
Contemplate new bases classes
Contemplate new base classes for feature node.
I want to add "pairs" in the lisp sense, so that I can represent the lisp-like trees
(NP John (VP is (OBJ a dork))) .
Nope. Scratch that. Feature node offers a workable paradigm, trying to
jam something else in there would just be painful, and would increase confusion.
|
More code review & cleanup
Bust up SentanceView into two pieces
a per-word piece, and a traversal piece
|
Cleanup and make my current cerego code publically presentable
back to hacking.
Still have problems, e.g.
physical limits to transistor miniaturization and performance are being approached .
Object (approached[10], limits[2])
Tense (approached[10], present_progressive)
Adjective (physical[1], limits[2])
to (limits[2], oh no mr bill!)
Number (limits[2], plural)
Modifier (transistor[4], miniaturization[5])
OK, that wwas a bug in my code. Now fixed.
Object (approached[10], limits[2])
Tense (approached[10], present_progressive)
Adjective (physical[1], limits[2])
to (limits[2], miniaturization[5])
to (limits[2], performance[7])
Number (limits[2], plural)
Modifier (transistor[4], miniaturization[5])
|
skim categorial grammar
Replying to boris' email made me look up Lambek grammar.
(aka categorial grammar!?)
Appearently, Boris took a course taught by Lambek.
|
fiddling with relex
The disco repository still doesn't work for me, as the hypergraph code is still crashing.
|
Register link grammar parser with sourceforge
wrote several emails to lafferty, temperley, sleator in the last month, no reply yet.
|
|
/novamente/linkparser - Link Gramar Parser Work
Discover that there is an existing link grammar parser maintenance site.
There is an existing link grammar maintenance project, its been in operation for a
few years, it seems. They've gone through many versions now, and appearently have
dictionaries for verious non-english languages.
Its at: http://www.abisource.com/projects/link-grammar/
I'm contacting the author, I want to integrate in my patches, and the novamente
patches as well (e.g. the jni bindings, and the couple of fixes that Mike has made.)
|
|