Example-based Machine Translation without Saying Inferable Predicate

Example-based Machine Example-based Machine Translation without Saying Translation without Saying Inferable PredicateInferable Predicate

Eiji Aramaki* **, Sadao Kurohashi* ** ，Hideki Kashioka** , Hideki Tanaka***

* University of Tokyo ** ATR *** NHK

EBMT FrameworkEBMT Framework

conference└held └in Canada

conference└held └in Canada

EBMT using Structural Translation Examples

Input

Output

For each phrase in an input,

a plausible TE is retrieved

English phrases are combined to generate output

Translation Examples

カナダで

開かれた

五月初めに to be held

in Canada

会議で

議論されました

conference

for negotiations

カナダで(Canada)

開かれた(be held)

会議で(conference)

HUMAN : in Hokkaido as the tourism season

PurposePurpose

INPUT ：

Learning Inferable Expression

HUMAN : a conference in Canada

Predicate

OUTPUT: in Hokkaido greeted tourism season

OUTPUT: a conference held in Canada

(Canada) (be held) (conference)カナダで開かれる会議で

(tourism season) (start) (Hokkaido)

観光シーズンを迎えた北海道ではINPUT ：

Redundant

Failure

OutlineOutline

• Verb phase (VP) aligned corpus – How to build VP-aligned-corpus– Investigation of VP-aligned-corpus

• Method

• Experiment

• Conclusion

VP-aligned-corpusVP-aligned-corpus

• Motivation– Verb-phrase translation is difficult⇒To learn how to translate verb-phrases, we annotated

verb-phrases and their correspondences.

• Process– Estimate alignment automatically– Modify alignment by hand

• Target– NHK-News-Corpus

ＮＨＫＮＨＫ -News-Corpus-News-Corpus

• NHK is a Japanese broadcasting corporation• Japanese articles are translated to be natural as

English news not parallel corpus⇒

田植え体験は農作業を通して米作りの意義などを考えていこうという地球環境平和財団の呼び掛けで開かれたもので、海外三十四ヵ国の大使や書記官、それに一般の参加者ら合わせておよそ千人が集まりました。 The event was organized by the private Foundation for

Global Peace and Environment.田植えに使われた苗は去年の秋、天皇陛下が皇居で収穫された稲籾から育てたものです。

参加者たちは裸足になって水田に足を踏み入れ地元に伝わる田植え歌に合わせて慣れない手つきで苗を植えていました。

田植えフェスティバル石川県輪島市で外国の大使や一般の参加者など千人あまりが急な斜面の棚田で田植えを体験する催しが行われました。

輪島市白米町には（しろよねまち）千枚田と呼ばれる（せんまいだ）大小二千百枚の棚田が急な斜面から海に向かって拡がっています。

きょうの輪島市は雲が広がったもののまずまずの天気となり、出席された高円宮さまも海からの風に吹かれながら田植えに加わっていました。地球環境平和財団では今年の夏休みに全国の子どもたちを対象に草刈りや生きものの観察会を開く他、秋には稲刈体験を行なう予定にしています。

Ambassadors and diplomats from 37 countries took part in a rice planting festival on Sunday in small paddies on steep hillsides in Wajima, central Japan.

About one-thousand people gathered at the hill, where some two-thousand 100 miniature paddies, called Senmaida, stretch toward the Sea of Japan.

The rice seedlings are grown from grain harvested by the Emperor at the Imperial Palace in Tokyo last autumn.

Barefoot participants waded into the paddies to plant the seedlings by hand while singing a local folk song about the practice of rice planting.

ＮＨＫＮＨＫ -News-Corpus-News-Corpus

• Some phrases has no parallel expressions in the other language.











Alignment EstimationAlignment Estimation• DP matching Sentence alignment using translation

dictionaries• Extract 1-to-1 sentence pairs • Parse and Phrasal alignment [Aramaki et al., 2003]





















Alignment ModificationAlignment Modification• We modified alignment by hand (5,500 sentence-pairs)

(1) VP or notVP:= phrase includes a verb or an adjective that has an argument

(2) For VP, its corresponding phrases

VP has no corresponding phrases

Investigation of CorpusInvestigation of Corpus

Where does a Japanese phrase corresponds in English?

Japanese : English #

VP : VP 9779

VP : φ 6831

VP : NP or PP 716

VP : Others 319

カナダで

開かれた

通商会議で in Canada

at a trade conference(be held)

(Canada)

(trade conference)

Condensed Alignment PatternCondensed Alignment Pattern

OthersOthers

11491149

56825682

Investigation of CorpusInvestigation of Corpus

Where does a Japanese phrase corresponds in English?

Japanese : English #

VP : VP 9779

VP : φ 6831

VP : NP or PP 716

VP : Others 319

Condensed Alignment PatternCondensed Alignment Pattern

OthersOthers

11491149

56825682

• VP-φ often occurs with Condensed Alignment Pattern (or CAP)

P’

C’C

PVP

カナダで

開かれた

通商会議で in Canada

at a trade conference(be held)

(Canada)

(trade conference)

If a in a CAP is always inferable ⇒We can achieve a compact translation by omitting the .

VP

VP

VP in CAP is inferableVP in CAP is inferable• We randomly extracted 80 CAPs, checked

whether s are inferable or not

inferable 56

not Parse Error 3

Alignment Error 11

Phrase Chunking Error 1

Others 9

P’

C’C

PVP

VP

VP in CAP is inferableVP in CAP is inferable

Inferable P-CONTEXT 21

C-CONTEXT 16

PC-CONTEXT 19

Not Parse Error 3

Alignment Error 11


Others 9

P’

C’C

PVP

• We randomly extracted 80 CAPs and checked whether s are inferable or notVP

P (Parent)-CONTEXTP (Parent)-CONTEXT カナダで開かれた会議で～ (Canada) (be held) (conference)

カナダで開かれた会議で～ (Canada) (be held) (conference)

カナダで

in Canada

at a conference

開かれた

会議で

P’

C’

C

P

at a conference in Canada …at a conference in Canada …

VP

(conference)

(be held)

(Canada)

C (Child)-CONTEXTC (Child)-CONTEXT 肺の病気にかかった男性の～(lung disease) (suffer) (man)

肺の病気にかかった男性の～(lung disease) (suffer) (man)

肺の

with a lung disease

the man

かかった

男性の

P’

C’

P

the man with a lung disease … the man with a lung disease …

VP

(man)

(suffer)

(lung)病気に

C (disease)

PC-CONTEXTPC-CONTEXT各国から派遣された救助チームの～ (countries) (sent) (rescue team)

各国から派遣された救助チームの～ (countries) (sent) (rescue team)

from counties

rescue teams派遣された

救助チームの

P’

C’P

rescue teams form countries … rescue teams form countries …

VP

(rescue team)

(sent)

各国からC (countries)

VP in CAP is inferableVP in CAP is inferable

Inferable P-CONTEXT 21

C-CONTEXT 16

PC-CONTEXT 19

Not Parse Error 3

Alignment Error 11


Others 9

P’

C’C

PVP

• We randomly extracted 80 CAPs and checked whether s are inferable or notVP

OutlineOutline

• Verb phase aligned corpus

• Method

• Experiments

• Conclusion

カナダで

開かれた

会議で

in Canada

at a conference

(be held)

(conference)

(Canada)

カナダで

Basic IdeaBasic Idea

• Incorporate CAPs into translation examples

開かれた

会議で

in Canada

at a conference

(be held)

(conference)

• Generalize TEsP-CONTEXT Generalize C⇒C-CONTEXT Generalize P⇒

(Canada)


カナダでin Canada(Canada)

会議で at a conference(conference)

X X

Alignment

Automatic ClassificationAutomatic Classification

Divide CAPs into 2 fragments and count frequencies

P’

C’C

PVP

シンガポールで

開かれた

会議で

in Singapore

at a conference(be held)

(Singapore)

(conference)

P’

C’C

PVP

ロンドンで

開かれた

会議で

in London

at a conference(be held)

(London)

(conference)

Fragment FrequencyFragment Frequency

V C C’ Freq

開く(held)

カナダ Canada 4

リヨン Lyon 4

シンガポール Singapore 3

ロンドン London 3

P V P’ Freq

会議開く

(held)

conference 17

会合 meeting 2

集会 gathering 1

Fragments which include “ 開く (held)”

Automatic ClassificationAutomatic Classification

freq(C) =4

freq(P) =17

freq(C) > freq(P) x Ｃ ⇒ C-CONTEXT

freq(P) > freq(C) x Ｃ ⇒ P-CONTEXT

Otherwise PC-CONTEXT⇒

Divide CAPs into 2 fragments and count frequencies

2

2

Translation FrameworkTranslation FrameworkInput

東京で(Tokyo)

開かれた(be held)

会議で(conference)

Translation Examples (=CAP)

開かれた

＊会議で

in Canada

X(be held)

X

( ＊ conference)

in Tokyo

at a conference


Tokyo(Tokyo)

Output

東京

OutlineOutline


• Method

• Experiments

• Conclusion

Preparation for Preparation for Experiments Experiments

• Extract 30,000 1-to-1 sentence pairs• Automatically estimate phrasal alignments • Extract and classify 4,219 CAPs

#

Ｐ -CONTEXT 1120

Ｃ -CONTEXT 297

ＰＣ -CONTEXT 2802

ExperimentsExperiments

Judgments of CAP Classifications are subjective task. ⇒ Evaluation by full Translation using generalized

CAPs

• Translation– Direction:

Japanese-English translation

– Test-set: 240 lead (top)-sentences of NHK-News-Corpus

– Evaluation:BLEU-score (4 references)

Methods & ResultsMethods & Results

BLEU

BASELINE 24.6

ＣＡＰ MT 24.8

ＣＡＰ MT ＋ 25.0

• Methods– BASELINE ： EBMT without CAPs– CAPMT ： EBMT using CAPs– CAPMT+ ： EBMT using generalized CAPs

ExampleExample

INPUT アメリカのメリーランド州で十四日に行われた航空ショーで・・・

REF An air show in the US state of Maryland on the 14th

BASELINE Air show was held in maryland of the united states on the 14th ...

CAP+ Air show in maryland of the united states on the 14th ...

ExampleExample

INPUT 二十五日に行われる日韓首脳会談に ...

REF ... summit due to be held on the 25th

BASELINE ... summit meeting conducted on 25th

CAP+ ... summit meeting on 25th

OutlineOutline


• Method

• Experiments

• Conclusion

ConclusionConclusion

• VP-aligned-Corpus– How to build the corpus– Investigations of the corpus

• Method for translations without saying inferable predicate

• Future work– A human translates inferable expressions with a

certain context– We have to deal with larger context

Example-based Machine Translation without Saying Inferable Predicate

Documents