Wamesa/Transducer

From LING073
Revision as of 20:16, 1 March 2017 by Twarner2 (talk | contribs) (Evaluation)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Analyser

Evaluation

My analyser covers 32 stems.
My main yaml file (wad.yaml) has 54/100 tests passing. Instances of infixation (<p2><sg> and <p3><sg>) make up the vast majority of the tests that are not passing.
At first, my aq-covtest only covered 2.22% of my lexicon. The most egregiously uncovered words were yau and pai, and when I added these, coverage increased to 10.77%. Now, my commonwords.yaml file has 2 out of 3 tests passing.

Notes

It is hard to meaningfully say what words my transducer covers, because most of the top uncovered words are actually morphemes, just ones that in some scholars' transcriptions are written as a separate word and in others' transcriptions are glommed onto the stem.


Generator

Evaluation

My analyser/generator now has 67/106 tests passing. I added 6 rules to my twol file. In my twoc file, I split the sets of square-bracketted features into four smaller sections each, to alleviate the former painful compile time.


Disambiguator

Disambiguator