key to abbreviations for selection methods
now exploring elimination of type or type-tag pair duplication w/in each individual selection round
seeded randomly w/ 50 sentences, initial step size 200 words, doubling every three rounds
| file | selection methods | graph cutoff |
|---|---|---|
| :morph:earl_al:gr_randuniq.pdf | rand v. rand,uniq | 10K words |
| :morph:earl_al:gr_uniqre.pdf | uniq: rand v. wlEnt | 10K words |
| :morph:earl_al:gr_combore.pdf | combo: rand v. wlEnt | 10K words |
| :morph:earl_al:gr_entauc.pdf | wlEnt: basic v. uniq v. combo | 10K words |
comparing wlEnt and randbase, but now using weaker features for training the tagger
| file | selection methods | graph cutoff |
|---|---|---|
| :morph:earl_al:wk_graph1.pdf | weak:wlEnt-nopool, randbase | 2.5K words |
| :morph:earl_al:wk_graph2.pdf | weak:wlEnt-nopool, randbase | 50K words |
| :morph:earl_al:wk_graph3.pdf | weak:wlEnt-nopool, randbase | 20K words |
| :morph:earl_al:wk_graph4.pdf | weak:wlEnt-nopool, randbase | 12K words |
wlEnt: seeded (randomly) with 25 sentences, no pool, initial step size of 100 words being compared to randbase as below (24 march)
| file | selection methods | graph cutoff |
|---|---|---|
| :morph:earl_al:graph10.pdf | wlEnt-nopool, randbase | 50K words |
| :morph:earl_al:graph11.pdf | wlEnt-nopool, randbase | 20K words |
| :morph:earl_al:graph12.pdf | wlEnt-nopool, randbase | 12K words |
| :morph:earl_al:graph13.pdf | wlEnt-nopool, wlEnt-pool | 20K words |
wlEnt: seeded (randomly) with 25 sentences, pool of 100 sentences, initial step size of 100 words
randbase: seeded (randomly) with 25 sentences, initial step size of 25 sentences
seqbase: seeded (randomly) with 25 sentences, initial step size of 25 sentences
all: step size double every three cycles
| file | selection methods | graph cutoff |
|---|---|---|
| :morph:earl_al:graph4.pdf | wlEnt, randbase, seqbase | 50K words |
| :morph:earl_al:graph6.pdf | wlEnt, randbase, seqbase | 20K words |
| :morph:earl_al:graph5.pdf | wlEnt, randbase, seqbase | 12K words |
| :morph:earl_al:graph7.pdf | wlEnt, randWord | 50K words |
| :morph:earl_al:graph9.pdf | wlEnt, randWord | 20K words |
| :morph:earl_al:graph8.pdf | wlEnt, randWord | 12K words |
| file | selection methods | graph cutoff |
|---|---|---|
| :morph:earl_al:graph1.pdf | wlEnt: all v. unks | 50K |
| :morph:earl_al:graph2.pdf | all: wlEnt, randbase(nov07) | 50K |
| :morph:earl_al:graph3.pdf | unks: wlEnt, randbase(nov07) | 50K |