RGyalrong (Situ) and English

From LING073
Jump to: navigation, search

Machine Translation between JiaMuZu rGyalrongic Situ and English

Developed Resources

GitHub jya-eng repository

GitHub Situ Transducer repository

Apertium English Transducer GitHub repository

Evaluation

Coverage

Mini-corpus (~1000 characters) coverage rose to 68.56%

$ ./coverage-lt.sh path/to/corpus jya-eng.automorf.bin
Generating hitparade (might take a bit!)
TOP UNKNOWN WORDS:
      6 ^shnə/*shnə$
      5 ^mak̚/*mak̚$
      5 ^kə/*kə$
      4 ^sɬo/*sɬo$
      3 ^rək̚/*rək̚$
      3 ^dxi/*dxi$
      2 ^xⁱe/*xⁱe$
      2 ^xiŋ/*xiŋ$
      2 ^və/*və$
      2 ^tsoŋ/*tsoŋ$
      2 ^tshi/*tshi$
      2 ^te/*te$
      2 ^sɬob/*sɬob$
      2 ^shod̚/*shod̚$
      2 ^sdok̚/*sdok̚$
      2 ^niⁱan/*niⁱan$
      2 ^le/*le$
      2 ^kʰa/*kʰa$
      2 ^bən/*bən$
      1 ^ʒak̚/*ʒak̚$
coverage: 181 / 264 (~0.68560606060606060606)
remaining unknown forms: 83
ambiguity: 760 / 264 (~2.87878787878787878788)
jya.sentences.txt Thu Mar 26 02:05:32 AM EDT 2026       181/264 ~0.68560606060606060606 \|      760/264 ~2.87878787878787878788 


Sentences

Original Biltrans Output Our Translation Actual Translation
mbndəd dshəʼdə mbndəd magənndɽa nagənndoʼ naŋos ^mbnded<n>/demon<n>$ ^dsheʼ<det><dem><mod><sg>/this<det><dem><mod><sg>$ ^mbnded<n>/demon<n>$ ^magenndra<adj>/different<adj>/naughty<adj>/terrible<adj>$ ^nndoʼ<v><prt_stat><perf><vdn><npst>/exist<vaux><prt_stat><perf><vdn><npst>$ ^ŋos<v><perf><vdn><npst>/be<vaux><perf><vdn><npst>$ this demon #exist different demon #be This demon was a very terrible demon
rə mbndəddə mənaŋos nə mbndəddə wərniidə ^mbnded<n><mod>/demon<n><mod>$ ^ŋos<v><perf><vdn><qst><npst>/be<vaux><perf><vdn><qst><npst>$ ^ne<cnjsub>/<cnjsub>$ ^mbnded<n><mod>/demon<n><mod>$ ^wujo<prn><pers><p3><sg><gen>/he<prn><pers><p3><m><sg><gen>/she<prn><pers><p3><f><sg><gen>$ ^rnii<v><prt_n><npst><mod>/name<n><npst><mod>$ and #be the demon #<cnjsub> the demon his #name The demon, eh, the demons name—
mbndəddə qabba raŋliŋ gərniigə manjuʼ ^mbnded<n><mod>/demon<n><mod>$ ^qabba<n>/Chap-pa<n>$ ^raŋliŋ<n>/Rang-ling<n>$ ^rnii<v><prt_stat><npst><mod>/be.called<v><npst><mod>$ ^manjuʼ<cnjadv>/moreover<cnjadv>$ the demon #Chap-pa #be.called #Rang-ling #moreover besides the demon called Chap-pa Rang-ling
wanndɽiʼdə mənaŋos nə warjaʼbdə mənaŋos nə ^wujo<prn><pers><p3><sg><gen>/he<prn><pers><p3><m><sg><gen>/she<prn><pers><p3><f><sg><gen>$ ^anndriʼ<n><mod>/friend<n><mod>$ ^ŋos<v><perf><vdn><qst><npst>/be<vaux><perf><vdn><qst><npst>$ ^ne<cnjsub>/<cnjsub>$ ^wujo<prn><pers><p3><sg><gen>/he<prn><pers><p3><m><sg><gen>/she<prn><pers><p3><f><sg><gen>$ ^arjaʼb<n><mod>/wife<n><mod>$ ^ŋos<v><perf><vdn><qst><npst>/be<vaux><perf><vdn><qst><npst>$ ^ne<cnjsub>/<cnjsub>$ #be his friend #<cnjsub> #be his wife #<cnjsub> there was his lover, eh, his wife, yeah,
mbndədmo hashaŋ magə nagərnii nagəŋos 'nəŋos ^mbndedmo<n>/demonness<n>$ ^hashaŋ<n>/Hashang<n>$ ^mage<n>/Make<n>$ ^rnii<v><prt_stat><perf><vdn><npst>/be.called<v><perf><vdn><npst>$ ^ŋos<v><prt_stat><perf><vdn><npst>/be<vaux><prt_stat><perf><vdn><npst>$ ^ŋos<v><ev><npst>/be<vaux><ev><npst>$ #demonness #Hashang #be.called Make #be #be who was a demonness called Hashang Make.
rə nndzamlaŋ gərgidə rjarbo gərgidə mənaŋos nə ^re<cnjcoo>/and<cnjcoo>$ ^nndzamlaŋ<n>/world<n>$ ^gergi<n><mod>/one<n><mod>$ ^rjarbo<n>/king<n>$ ^gergi<n><mod>/one<n><mod>$ ^ŋos<v><perf><vdn><qst><npst>/be<vaux><perf><vdn><qst><npst>$ ^ne<cnjsub>/<cnjsub>$ and world the #one king #be the #one #<cnjsub> In the whole world there was one king—
rə dərmudə ganndza dshəʼdshe rə rjarbo gərgidə ^re<cnjcoo>/and<cnjcoo>$ ^rmu<n><prt_n><mod>/person<n><prt_n><mod>$ ^nndza<v><prt_dyn><npst>/eat<v><npst>$ ^dsheʼ<det><dem><sg>/this<det><dem><sg>$ ^LOC<loc>/in<pr>/on<pr>/at<pr>/from<pr>$ ^re<cnjcoo>/and<cnjcoo>$ ^rjarbo<n>/king<n>$ ^gergi<n><mod>/one<n><mod>$ and #eat person this in and king the #one When he ate people he grabbed a king
wəsbog gəgənərgow konə ^wujo<prn><pers><p3><sg><gen>/he<prn><pers><p3><m><sg><gen>/she<prn><pers><p3><f><sg><gen>$ ^sboʼg<n><prt_n>/underside<n><prt_n>$ ^nergo<v><s_3sg><o_3><prt_stat><perf><sup><npst>/put<v><s_3sg><o_3><prt_stat><perf><sup><npst>$ ^kone<cnjsub>/so<cnjsub>$ #put his underside #so and put him under his rule;
nagəŋosyn 'nəŋos ^ŋos<v><s_3pl><o_3><prt_stat><perf><vdn><npst>/be<vaux><s_3pl><o_3><prt_stat><perf><vdn><npst>$ ^ŋos<v><ev><npst>/be<vaux><ev><npst>$ #be #be so it was
gərjiyn 'nəŋos ^rji<v><s_3pl><o_3><prt_stat><ifi>/went<v><s_3pl><o_3><prt_stat><ifi>$ ^ŋos<v><ev><npst>/be<vaux><ev><npst>$ #went #be it was

Additions

For the final system, I expanded the MT pair in three ways:

Firstly, I added five new morphology elements.

Secondly, I added two lexical-selection rules.

Lastly, I added two transfer-rule patterns for demonstrative noun phrases.

Please find these additions in the latest commit by me in the bilingual repository

Final Evaluation

Coverage on the large corpus (obtained by concatenating the two existing corpora):

$ coverage-hfst ../initial_corpus/jya.corpus.large.txt jya.automorf.hfst
Generating hitparade (might take a bit!)
TOP UNKNOWN WORDS:
     62 ^də/*də$
     57 ^na/*na$
     40 ^dogəcəs/*dogəcəs$
     37 ^sog/*sog$
     34 ^jo/*jo$
     28 ^da/*da$
     24 ^wu/*wu$
     23 ^shi/*shi$
     23 ^gacəs/*gacəs$
     23 ^a/*a$
     22 ^bə/*bə$
     21 ^jig/*jig$
     20 ^bəjəg/*bəjəg$
     19 ^ma/*ma$
     18 ^yə/*yə$
     18 ^gəngu/*gəngu$
     16 ^shə/*shə$
     14 ^txi/*txi$
     14 ^shnə/*shnə$
     13 ^sdo/*sdo$
coverage: 1756 / 3983 (~0.44087371328144614612)
remaining unknown forms: 2227
ambiguity: 6150 / 3983 (~1.54406226462465478283)
jya.corpus Thu May 14 04:49:02 AM EDT 2026      1756/3983       ~0.44087371328144614612 \|      6150/3983       ~1.54406226462465478283