Latin and Mandarin Chinese/Structural transfer

From LING073
Revision as of 12:06, 19 April 2018 by Cdalton2 (talk | contribs)

Jump to: navigation, search

This is the page for the structural transfer of Latin and Mandarin Chinese. The main page for this language pair can be found here.

Pre-evaluation

Latin corpus coverage

Number of tokenised words in the corpus: 380

Coverage: 88.68%

Top unknown words in the corpus:

2 potest

2 facet

2 possit

2 quo

1 tibi

1 Mariaene

1 quid

1 audit

1 possum

1 matrae

1 James

1 loquent

1 audire

1 not

1 Videbasne

1 duo

1 poterunt

1 eae

1 aliquid

1 posset

Chinese corpus coverage

Number of tokenised words in the corpus: 447

Coverage: 100.00%

lat → zho

WER: 586.79%

PER: 586.79%

zho → lat

WER: 97.42%

PER: 93.55%

Implementation

lat → zho

For translating into Chinese, we implemented macros that take case off of nouns and add the associative particle 的 to nouns that, in Latin, had the genitive case.

Maria caput pueri videt.

"Maria saw the boy's head."

tagger:

^Maria<np><ant><f><sg><nom>$ ^caput<n><nt><sg><nom>$ ^puer<n><m><sg><gen>$ ^videre<vblex><pri><act><p3><sg>$^.<sent>$

biltrans:

^Maria<np><ant><f><sg><nom>/小红<np><ant><f><sg><nom>$ ^caput<n><nt><sg><nom>/头<n><nt><sg><nom>$ ^puer<n><m><sg><gen>/男孩<n><m><sg><gen>$ ^videre<vblex><pri><act><p3><sg>/看到<vblex><pri><act><p3><sg>$^.<sent>/。<sent>$

chunker:

apertium-transfer: Rule 2 caput<n><nt><sg><nom>/头<n><nt><sg><nom>

apertium-transfer: Rule 2 puer<n><m><sg><gen>/男孩<n><m><sg><gen>

apertium-transfer: Rule 1 .<sent>/。<sent> ^default<default>{^小红<np><ant><f><sg><nom>$}$ ^noun<SN>{^头<n>$ }$ ^noun<SN>{^男孩<n>$ ^的<pr>$}$ ^default<default>{^看到<vblex><pri><act><p3><sg>$}$^sent<SENT>{^。<sent>$}$

interchunk:

Rule 1 noun<SN>{^男孩<n>$ ^的<pr>$} ^default<default>{^小红<np><ant><f><sg><nom>$}$ ^noun<SN>{^头<n>$ }$ ^noun<SN>{^男孩<n>$ ^的<pr>$}$ ^default<default>{^看到<vblex><pri><act><p3><sg>$}$^sent<SENT>{^。<sent>$}$

postchunk:

^小红<np><ant><f><sg><nom>$ ^头<n>$ ^男孩<n>$ ^的<pr>$ ^看到<vblex><pri><act><p3><sg>$^。<sent>$

lat-zho:

#小红 头 男孩 的 #看到。

zho → lat

For translating into Latin, we implemented rules to take away the preposition 给 and change the following noun into dative case.

我给狗听。

"I listen for the dog (i.e., for the dog's sake/at the dog's behest)."

tagger:

^我<prn>$ ^给<pr>$ ^狗<n>$ ^听<vblex>$

biltrans:

^我<prn>/ego<prn>$ ^给<pr>/$ ^狗<n>/canis<n><m>$ ^听<vblex>/auscultare$

chunker:

apertium-transfer: Rule 2 给<pr>/

apertium-transfer: Rule 3 狗<n>/canis<n><m> ^default<default>{^ego<prn>$}$ ^pr<SPR><dat>{}$ ^nom<SN><ND><CD>{^canis<n><m><2><3>$}$ ^default<default>{^auscultare$}$

interchunk:

apertium-interchunk: Rule 2 pr<SPR><dat>{} nom<SN><ND><CD>{^canis<n><m><2><3>$} ^default<default>{^ego<prn>$}$ ^nom<SN><sg><dat>{^canis<n><m><2><3>$}$ ^default<default>{^auscultare$}$

postchunk:

^ego<prn>$ ^canis<n><m><sg><dat>$ ^auscultare$

zho-lat

#ego cani #auscultare

Post-evaluation

Latin corpus coverage

Number of tokenised words in the corpus: 380

Coverage: 88.68%

Top unknown words in the corpus:

2 possit

2 potest

2 quo

2 facet

1 abest

1 duo

1 Id

1 Te

1 quid

1 vult

1 Utrique

1 poterunt

1 accipiunt

1 se

1 veta

1 Scisne

1 tibi

1 not

1 possum

1 posset

Chinese corpus coverage

Number of tokenised words in the corpus: 447

Coverage: 100.00%

lat → zho

WER: 97.84%

PER: 97.04%

(after adding spaces to zho.sentences.txt)

zho → lat

WER: 95.45%

PER: 89.39%