Differences
This shows you the differences between two versions of the page.
| Both sides previous revision Previous revision Next revision | Previous revision | ||
|
external:addicter [2011/05/16 07:47] mphi |
external:addicter [2011/05/22 09:31] (current) mphi |
||
|---|---|---|---|
| Line 4: | Line 4: | ||
| This page lies in the external name space and is intended for collaboration with people outside of ÚFAL. | This page lies in the external name space and is intended for collaboration with people outside of ÚFAL. | ||
| + | |||
| + | ==== TODOs ==== | ||
| + | * test alignment with synonym detection (cz_wn required) = separating '' | ||
| + | * try misplaced phrase detection | ||
| + | * parse the reference, extrapolate unto the hypothesis via word alignment, get phrases from there | ||
| + | * group adjacent words aligned to the same word? (imitation) | ||
| + | * order evaluation | ||
| + | * currently finds misplaced items, but their shift distances are off | ||
| + | * not important for '' | ||
| + | * to fix -- for every misplaced token | ||
| + | * if it (and only it) were to be moved in the original permutation, | ||
| + | * evaluate with nr. of intersections | ||
| + | * try domain adaptation for word alignment with the "via source" | ||
| + | * technical | ||
| + | * comb and comment the code | ||
| + | * add help files | ||
| + | * integrate with the rest of Addicter | ||
| + | * approach applicable to learner' | ||
| + | * see Anne Lüdelig, TLT9 | ||
| + | * try Blast (Sara' | ||
| + | * alternative to reference-based evaluation: " | ||
| ==== Word Alignment -- Progress and Results ==== | ==== Word Alignment -- Progress and Results ==== | ||
| - | Latest best results: [[http:// | + | === Latest best results |
| + | [[http:// | ||
| === Alternative model comparison === | === Alternative model comparison === | ||
| + | |||
| + | hmm = lightweight direct alignment method (in our ACL/TSD article) | ||
| + | gizainter = GIZA++, intersection -- applied to hypotheses+references directly | ||
| + | gizadiag = GIZA++, grow-diag -- applied to hypotheses+references directly | ||
| + | czenginter = align source+CzEng to reference+CzEng, | ||
| + | czengdiag = same, but with GIZA++ grow-diag | ||
| + | |||
| | | | | ||
| | | | | ||
| - | ^ meteor | ||
| ^ ter* |0.106/ | ^ ter* |0.106/ | ||
| + | ^ meteor | ||
| ^ hmm | ^ hmm | ||
| ^ lcs | ^ lcs | ||
| ^ gizainter |0.170/ | ^ gizainter |0.170/ | ||
| ^ gizadiag* |0.183/ | ^ gizadiag* |0.183/ | ||
| - | ^ berkeley* |0.200/ | ||
| - | |||
| - | === Explicit wrong lex choice detection === | ||
| - | * align input+czeng to reference+czeng and input+czeng to hypotheses+czeng | ||
| - | * extract hypothesis-to-reference alignments from there | ||
| - | |||
| - | | ^ Precision/ | ||
| - | | ^ | ||
| ^ czengdiag* |0.187/ | ^ czengdiag* |0.187/ | ||
| + | ^ berkeley* |0.200/ | ||
| ^ czenginter |0.197/ | ^ czenginter |0.197/ | ||
| + | |||
| + | * non-1-to-1 alignments, converted to 1-to-1 via " | ||
| === Alignment combinations === | === Alignment combinations === | ||
| Line 34: | Line 58: | ||
| | | | | ||
| | | | | ||
| - | ^ meteor+hmm | ||
| ^ ter+hmm | ^ ter+hmm | ||
| + | ^ meteor+hmm | ||
| ^ gizadiag+hmm | ^ gizadiag+hmm | ||
| ^ gizainter+hmm | ^ gizainter+hmm | ||
| Line 42: | Line 66: | ||
| ^ czenginter+hmm | ^ czenginter+hmm | ||
| | ||||| | | ||||| | ||
| - | ^ berkeley+czenginter+hmm |0.219/ | + | ^ berk+czengint+hmm |0.219/ |
| - | ^ berkeley+czenginter+gizainter+hmm |0.220/ | + | ^ berk+czengint+gizaint+hmm |0.220/ |
| - | ^ berkeley+czenginter+meteor+hmm |0.220/ | + | ^ berk+czengint+meteor+hmm |0.220/ |
| - | ^ berkeley+czenginter+meteor+gizainter+hmm |0.221/ | + | ^ berk+czengint+meteor+gizaint+hmm |0.221/ |
| - | ==== TODOs ==== | ||
| - | * test alignment with synonym detection (cz_wn required) = separating '' | ||
| - | * order evaluation | ||
| - | * a lot of background research | ||
| - | * currently finds misplaced items, but their shift distances are off | ||
| - | * to fix -- for every misplaced token | ||
| - | * if it (and only it) were to be moved in the original permutation, | ||
| - | * evaluate with nr. of intersections | ||
| - | * try domain adaptation for word alignment, EMNLP 2011 paper | ||
| - | * comb and comment the code | ||
| - | * add help files | ||
| - | * integrate with the rest of Addicter | ||
| - | * approach applicable to learner' | ||
| - | * see Anne Lüdelig, TLT9 | ||
| - | * adapt Addicter to Sara's program | ||
| - | * alternative to reference-based evaluation: " | ||