Difference between revisions of "Tiwi/Transducer"

From LING073
Jump to: navigation, search
(Initial Evaluation of Morphological Generation)
(Final Evaluation of Morphological Generation)
 
(6 intermediate revisions by 2 users not shown)
Line 83: Line 83:
 
----
 
----
  
The rest of the failing morphtests fail, because some nouns that are adjectives don't return the adjective tags or they return the word instead of a mixture of the stem and the tags.
+
Most of the rest of the failing morphtests fail, because some nouns that are adjectives don't have adjective tags associated with them or a morphophonological change in the stem, suffix, or prefix.
  
 
== Generator Evaluation ==  
 
== Generator Evaluation ==  
Line 92: Line 92:
 
* Passes and fails for the generation test: Total passes: 16, Total fails: 22, Total: 38
 
* Passes and fails for the generation test: Total passes: 16, Total fails: 22, Total: 38
  
=== Evaluation After Updating TWOL Rules ===  
+
=== Final Evaluation of Morphological Generation ===  
* passing and failing morphological analysis tests: Total passes: 39, Total fails: 6, Total: 45
+
<i>Note: after adding 9 <code>twol</code> rules for the nouns</i>
* coverage: 33/33 (1), remaining unknown forms: 0
+
* Passes and fails for the morphological analysis tests: Total passes: 45, Total fails: 7, Total: 52
* Passes and fails for the generation test: Total passes: 39, Total fails: 6, Total: 45
+
* coverage: 94 / 363 (~0.25895316804407713499): remaining unknown forms: 269
 
+
* Passes and fails for the generation test: Total passes: 45, Total fails: 7, Total: 52
== Final Evaluation of Morphological Generation ===
 
  
 
  [[Category: Sp21_Transducers]]  [[Category: Tiwi]]
 
  [[Category: Sp21_Transducers]]  [[Category: Tiwi]]

Latest revision as of 18:41, 21 March 2021

Code

Github Repo

Analyser Evaluation

Coverage: 62 / 363 (~0.17079889807162534435) and by adding:

  • ngarra<n><prn><m><p3><sg>:ngarra
  • awarra<n><prn><dem><m><sg>:awarra
  • api<cnjcoo>:api

Coverage improved from 0.17 to 0.22

Current top unknown words

TOP UNKNOWN WORDS:

  • ^kukuni/*kukuni$
  • kangi/*kangi$
  • ^arlijirrakini/*arlijirrakini$
  • api/*api$
  • ngini/*ngini$
  • Ninkiyi/*Ninkiyi$
  • mwarlijanguwi/*mwarlijanguwi$
  • karri/*karri$
  • amintiya/*amintiya$
  • naki/*naki$
  • kapi/*kapi$
  • awungarri/*awungarri$
  • yanamurluwi/*yanamurluwi$
  • waya/*waya$
  • pitipita/*pitipita$
  • papi/*papi$
  • palaji/*palaji$
  • pajinga/*pajinga$

Lexicon Information

  • Lexicons: 81
  • Lexicon entries: 130
  • Patterns: 5
  • Pattern entries: 35

Counts for individual lexicons

  • Punctuation: 22
  • I-VerbRoot: 5
  • T-VerbRoot: 3
  • V-Subject: 6
  • V-Object: 2
  • V-Sg-Imp: 1
  • NounRoot-1: 3
  • N-Suffix-Masculine: 2
  • N-Suffix-Feminine: 4
  • N-Suffix-Plural: 4
  • NounRoot-2: 5
  • NounRoot-3: 1
  • N-Prefix-Masculine: 3
  • NounRoot-4: 1
  • N-Prefix-Feminine: 2
  • NounRoot-5: 1
  • NounRoot-6: 1
  • PronounRoot-1: 1
  • Prn-Suffix-Masculine: 1
  • Prn-Suffix-Feminine: 1
  • Prn-Suffix-Plural: 1
  • PronounRoot-2: 1
  • PronounRoot-3: 1
  • Basic-Pronoun: 2
  • Conjunctions: 1
  • All anonymous lexicons: 55

Tests

In our main yaml file tiw.yaml, our transducer passes 48/88 tests (~55%) (-ci)

In our yaml file commonwords.yaml, our transducer passes 4/8 tests (50%)

Notes

  • wapa<v><iv><imp><s_sg2> ↔ taapa; fails because of morphophonology to be addressed with twol
  • pirni <v><tv><past><s_sg3><o_sg3><m> ↔ yipirni; fails because of complicated irregular case
  • kirimi<v><iv><past><s_pl3> ↔ pirikirimi; again, fails because of morphophonological change

  • wurumi<n><adj><pred><f> ↔ juwurumi
  • porimi<n><adj><pred><f> ↔ jiporimi
  • kupuranji<n><adj><pred><f> ↔ jikupuranji

all fail, because the tags appear before the stem instead of after


Most of the rest of the failing morphtests fail, because some nouns that are adjectives don't have adjective tags associated with them or a morphophonological change in the stem, suffix, or prefix.

Generator Evaluation

Initial Evaluation of Morphological Generation

Note: had to remove a few tests of suspect quality

  • passing and failing morphological analysis tests: Total passes: 16, Total fails: 27, Total: 43
  • coverage: 73 / 363 (~0.20110192837465564738), remaining unknown forms: 290
  • Passes and fails for the generation test: Total passes: 16, Total fails: 22, Total: 38

Final Evaluation of Morphological Generation

Note: after adding 9 twol rules for the nouns

  • Passes and fails for the morphological analysis tests: Total passes: 45, Total fails: 7, Total: 52
  • coverage: 94 / 363 (~0.25895316804407713499): remaining unknown forms: 269
  • Passes and fails for the generation test: Total passes: 45, Total fails: 7, Total: 52