Kickapoo/Transducer

From LING073
Jump to: navigation, search

Our code is listed here: https://github.swarthmore.edu/Ling073-sp26/ling073-kic

Evaluations

Transducer size

Using lexd -x apertium-kic.kic.lexd > /dev/null, the current transducer contains:

Category Count
Lexicons 28
Lexicon entries 105
Patterns 14
Pattern entries 26

Total number of stems in the transducer

We counted only open-class lexical stem lexicons and excluded punctuation, affixes, person markers, tense/mode markers, and other closed-class/function-word lexicons.

Stem lexicon Count
A-Stems 8
I-Stems 5
L-Stems 1
Verbstem 2
VStem_AA 4
Total 20

Current coverage over the combined corpus

Coverage before adding common words:

  • 83 / 584 = 14.21%

Current coverage:

  • 133 / 584 = 22.77%

Improvement in coverage:

  • +50 analysed tokens
  • +8.56 percentage points

Additional corpus statistics:

  • Remaining unknown forms: 451
  • Ambiguity: 847 / 584 = 1.45

Current top unknown words

Rank Unknown word Frequency
1 kwiyatheeha 4
2 awiyeehi 4
3 yoóh 3
4 yaaha 3
5 sehe 3
6 iskweetheehani 3
7 epapakethomehaki 3
8 aia 3
9 yoohi 2
10 yeetoke 2
11 witheniikaani 2
12 weepenaanwi 2
13 tahkiai 2
14 seeski 2
15 see 2
16 poohkamáaki 2
17 papakethohahi 2
18 nootha 2
19 niihpya 2
20 nekoti 2


Unimplemented

  1. Vaux in Aspect not yet added; we haven't figured out a way to correctly analyze it;
  2. Negation also not yet added due to the same reason;
  3. Mode not yet added, may require further discussion;


Regtest results

The following table shows the number of tests that pass in each corpus.

Corpus Tests passed Match gold
analysis 2 / 2 0 / 2
generation 1 / 1 0 / 1
Animate Nouns-morph 2 / 2 2 / 2
Animate Nouns-gen 2 / 2 1 / 2
Inanimate Nouns-morph 2 / 2 1 / 2
Inanimate Nouns-gen 2 / 2 1 / 2
Noun Inflection (Posession)-morph 5 / 5 5 / 5
Noun Inflection (Posession)-gen 5 / 5 1 / 5
Person-morph 10 / 10 9 / 10
Person-gen 10 / 10 0 / 10
Future Tense-morph 2 / 2 0 / 2
Future Tense-gen 2 / 2 0 / 2
First Aorist-morph 1 / 1 0 / 1
First Aorist-gen 1 / 1 0 / 1
Second Aorist-morph 1 / 1 0 / 1
Second Aorist-gen 1 / 1 0 / 1
Third Aorist-morph 1 / 1 0 / 1
Third Aorist-gen 1 / 1 0 / 1
Aspect-morph 1 / 1 0 / 1
Aspect-gen 1 / 1 0 / 1
Mode-morph 6 / 6 0 / 6
Mode-gen 6 / 6 0 / 6
Negation-morph 6 / 6 0 / 6
Negation-gen 7 / 7 0 / 7
Locative-morph 1 / 1 1 / 1
Locative-gen 1 / 1 1 / 1
commonwords-morph 4 / 4 4 / 4

Test status summary

All 27 corpora currently pass all tests:

  • 79 / 79 tests pass (100%)

Commonwords evaluation

We added a new commonwords-morph test file for frequent unknown words from the corpus.

Corpus Tests passed Match gold
commonwords-morph 4 / 4 4 / 4

Notes

Unresolved Tests

Analyser Evaluation

Twol Rules

No Twol rules are added; No verification has been noticed in Kickapoo grammar in the new script.

Updated Stem Counts

Lexicons: 29 Lexicon entries: 106 Patterns: 13 Pattern entries: 26

Counts for individual lexicons:

Punctuation: 22

Demonstrative: 2

Particle: 1

A-Stems: 8

A-Number: 2

A-Poss: 5

I-Stems: 5

I-Number: 2

I-Poss: 5

L-Stems: 1

Loc: 1

Adverb: 1

Conj: 1

PrefPerson: 2

Vaux: 15

Verbstem: 2

VerbAnimacy: 2

SuffPerson: 8

Tense: 3

VStem_AA: 4

V-Mode: 5

Numeral: 3

All anonymous lexicons: 6

Generator Evaluation

Corpus 1 of 12: Animate Nouns-morph

 2/2 (100.0%) tests pass (2/2 (100.0%) match gold)

Corpus 2 of 12: Inanimate Nouns-morph

 2/2 (100.0%) tests pass (1/2 (50.0%) match gold)

Corpus 3 of 12: Noun Inflection (Posession)-morph

 5/5 (100.0%) tests pass (5/5 (100.0%) match gold)

Corpus 4 of 12: Person-morph

 10/10 (100.0%) tests pass (3/10 (30.0%) match gold)

Corpus 5 of 12: Future Tense-morph

 2/2 (100.0%) tests pass (0/2 (0.0%) match gold)

Corpus 6 of 12: First Aorist-morph

 1/1 (100.0%) tests pass (0/1 (0.0%) match gold)

Corpus 7 of 12: Second Aorist-morph

 1/1 (100.0%) tests pass (0/1 (0.0%) match gold)

Corpus 8 of 12: Third Aorist-morph

 1/1 (100.0%) tests pass (0/1 (0.0%) match gold)

Corpus 9 of 12: Aspect-morph

 1/1 (100.0%) tests pass (0/1 (0.0%) match gold)

Corpus 10 of 12: Mode-morph

 6/6 (100.0%) tests pass (0/6 (0.0%) match gold)

Corpus 11 of 12: Negation-morph

 6/6 (100.0%) tests pass (0/6 (0.0%) match gold)

Corpus 12 of 12: Locative-morph

 1/1 (100.0%) tests pass (1/1 (100.0%) match gold)

All tests pass.


Final Evaluation For Morphological Generation

The morphological generation component now passes 95 out of 95 tests, with 0 failing tests.

Overall, the system achieves a match gold rate of 72/95 = 75.8%.

For this part of the assignment, we added 2 new twol rules.

When we reran the coverage test on the largest corpus, the new coverage was 24.01%.