Ptrifon1 Final project

What We Did

We created a transducer for Romansh and evaluated based on its coverage of a corpus. With no available Romansh bible translation, we used the Swedish Penal Code as our corpus. Our transducer supports 14 Noun Stems and 3 Verb Stems as well as five parts of speech overall.

Evaluation Results

Morph/Gen Test Evaluation Corpus 1 of 12: analysis

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

Corpus 2 of 12: generation

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

Corpus 3 of 12: Nouns-morph

 14/14 (100.0%) tests pass (12/14 (85.71%) match gold)

Corpus 4 of 12: Nouns-gen

 14/14 (100.0%) tests pass (14/14 (100.0%) match gold)

Corpus 5 of 12: Adjectives-morph

 4/4 (100.0%) tests pass (4/4 (100.0%) match gold)

Corpus 6 of 12: Adjectives-gen

 8/8 (100.0%) tests pass (7/8 (87.5%) match gold)

Corpus 7 of 12: Articles-morph

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

Corpus 8 of 12: Articles-gen

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

Corpus 9 of 12: Conjunctions-morph

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

Corpus 10 of 12: Conjunctions-gen

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

Corpus 11 of 12: Verbs-morph

 33/33 (100.0%) tests pass (2/33 (6.06%) match gold)

Corpus 12 of 12: Verbs-gen

 40/40 (100.0%) tests pass (40/40 (100.0%) match gold)

---

Coverage Evaluation

TOP UNKNOWN WORDS:

   914 ^da/*da$
   556 ^il/*il$
   552 ^la/*la$
   336 ^u/*u$
   333 ^l/*l$
   283 ^in/*in$
   267 ^d/*d$
   237 ^en/*en$
   222 ^e/*e$
   202 ^per/*per$
   198 ^è/*è$
   191 ^dal/*dal$
   174 ^ch/*ch$
   168 ^chasti/*chasti$
   166 ^ina/*ina$
   159 ^che/*che$
   154 ^delinquent/*delinquent$
   148 ^tenor/*tenor$
   139 ^art/*art$
   133 ^derschader/*derschader$

coverage: 5762 / 21301 (~0.27050373221914464110) remaining unknown forms: 15539 ambiguity: 22843 / 21301 (~1.07239096756020844092)

---