Improving Machine Translation Quality with Automatic Named

Entity Recognition

Bogdan Babych

Centre for Translation StudiesUniversity of Leeds, UK

Department of Computer ScienceUniversity of Sheffield, UK

bogdan@comp.leeds.ac.uk

Anthony Hartley

Centre for Translation StudiesUniversity of Leeds, UK

a.hartley@leeds.ac.uk

Overview• Problems of Named Entities (NEs) for MT• Experiment set-up

– Segmentation of the MT output– Scoring scheme

• Results of the experiment• Discussion

– Improving MT with IE techniques

• Conclusions and future work

Problems of NEs for MT• NEs are the weak point for many MT systems• Distinct linguistic properties of proper nouns and

different translation strategies for NE• “NE internal” errors:

– Proper / common noun disambiguation errors– Errors in morphosyntactic categories of NEs

• “NE external” errors in the context of NEs:– Word sense disambiguation errors– Errors in morphosyntactic features in NE context– Segmentation errors

Translation strategies for NEs

• Language-dependent strategies– Eastern Slavonic languages: person names are

transcribed with Cyrillic characters

• Strategies dependent on a type of NE– [Newmark, 1982: 70-83]: organisation names are often

left untranslated– Languages with Cyrillic writing system: organisation

names are often left in original Roman orthography• E.g.: 4 articles on international economy from BBC Russian

site: Roman-script NEs cover 6% of the total 1000 tokens

Proper / common disambiguation errors– English: “Ray Rogers” – MT ProMT E-R: “Луч Rogers”

(‘A ray (beam of light) Rogers’)

– English: “Bill Fisher”– MT ProMT E-R : “Выставить счёт Рыбаку”– MT ProMT E-F : “Facturez le Pêcheur”

(‘(To) send a bill to a fisher’)

– English: “Jeff Levy”– MT Systran E-F : “prélèvement de Jeff”

(‘Jeff’s imposing of a tax’)

Contextual changes around unrecognised NEs

• Errors in morphosyntactic categories– English: “… they have been flying in United cockpits” – E-R MT: “… они летали в Объединенных кабинах”

(‘they have been flying in united (joined) cockpits’)

• Segmentation errors– English: “Eastern Airlines executives notified union

leaders …”– E-R MT: “Восточные исполнители авиалиний

уведомили профсоюзных руководителей…”(‘Oriental executives of the Airlines notified …')

Compound errors -- combining:

• “NE internal” errors and errors in the context of NEs

• Lexical disambiguation errors and errors in morphosyntactic disambiguation / segmentation– English: “In Ford-UAW talks…”– E-R MT: “В Броде - UAW говорит”

(‘In a ford (shallow place) - UAW is talking’)

Information Extraction (IE) technology

• IE: from unrestricted text to a database– specific subject domain (e.g. satellite launches)

– predefined template with fields to be filled

• IE tasks:– NE recognition

– Co-reference resolution

– Word sense disambiguation

– Template element filling

– Scenario template filling

– Summary generation

NE recognition in IE• NE recognition is specifically addressed and

benchmarked (DARPA MUC6 & MUC7 competitions)

• Manually annotated “gold standard” available• Highly accurate

– leading IE systems achieve F-score 80-90%– performance is higher and less dependent on a

subject domain (compared to Scenario Template Filling)

• Available under GPL: NE recognition module ANNIE in Sheffield’s GATE system

Using NE recognition for MT

• GATE-ANNIE system allows automatic annotation of NEs in English texts

• MT systems accept Do-Not-Translate (DNT) lists– acceptable translation strategy for many organisation names in

certain language pairs

• Suggestion: if NE recognition is more accurate for IE systems, then general MT quality will improve (compared to the baseline performance)– NE-Internal changes are predictable (DNT strategy)

– Changes in the context of NEs are more interesting and more difficult to predict

Experiment set-up• Purpose: evaluating morphosyntactic changes in

the context of NEs after DNT-processing

• Corpus: – 30 texts (news articles) from MUC6 evaluation set

(11,975 tokens, 510 NE occurrences, 174 NE types)– GATE “responses” -- NE recognition output file

generated by GATE-1 for MUC6 competition(Precision - 84%; Recall - 94%; F-measure - 89.06%)

• MT systems: – E-R ProMT 98; E-F ProMT 2001; E-F Systran 2000

Experiment set-up (contd.)

• Stage 1: Automatic generation of DNT lists from GATE-1 annotation

• Stage 2: Generating translations for 3 systems– Baseline translation (without a DNT list)– DNT-processed translation

• Stage 3: Automatic segmentation of translations into NE-internal and NE-external zones

• Stage 4: Manual scoring of NE-external differences

Segmentation algorithm

• Annotated NEs in the English original are looked up in the DNT-processed translation

• Strings between found NEs are then looked up in the baseline translation

• If a string is not found, it is highlighted (signaling a difference in the context of the NE)– Result: NE-internal and NE-external zones in the

baseline translation are separated– NE-external differences are highlighted– No complex alignment

Segmentation algorithm (contd.)

ORIGINAL DNT-PROCESSED BASELINE

Separately in its

SSEECC

filing UUSSAAiirr

disclosed details of its

plan for financing the

PPiieeddmmoonntt

Acquisition

Отдельно в его

регистрации

SSEECC,

UUSSAAiirr

раскрыл детали его

планов

финансирования

приобретения

PPiieeddmmoonntt

Отдельно в его

регистрации

СЕКУНДЫ,

USAir

раскрыл детали его

планов

финансирования

Предгорного

приобретения.

Scoring scheme

Evaluating morphosyntactic well-formedness

Score Baseline translation DNT-processedtranslation

+ 1 not well-formed well-formed+ 0.5 not well-formed; not well-formed;

some features aremore correct

= 0 equally (not) well-formed– 0.5 not well-formed;

some features aremore correct

not well-formed

– 1 well-formed not well-formed

Scoring examples: +1 score

+1 Original:(It) represents 4,400 Western Union employeesaround the country.Baseline translation:(Он) представляет 4,400 Западных служащихСоюза по всей стране.('It represents 4,400 Western employees of theUnion around the country')DNT-processed translation:(Он) представляет 4,400 служащих WesternUnion по всей стране.('(It) represents 4,400 employees of WesternUnion around the country')

Scoring examples: +0.5 score+0.5 Original:

Western Union Corp. said its subsidiary, WesternUnion Telegraph Co.…Baseline translation:Западная Корпорация Союза сказала еевспомогательную, Западную КомпаниюТелеграфа Союза…('Western Corporation of a Union said itsauxiliary (case.acc.), Western Company ofTelegraph of a Union …')DNT-processed translation:Western Union Corp. Сказанный его филиал,Western Union Telegraph Co. …('Western Union Corp. Its branch (case.nom) issaid, Western Union Telegraph Co.…')

Scoring examples: =0 score

=0 Original:American Airlines Calls for MediationBaseline translation:Американские Авиалинии Призывают Кпосредничеству(American Airlines Call(num.plur.) for Mediation)DNT-processed translation:American Airlines Призывает Кпосредничеству(American Airlines Calls(num.sing.) for Mediation)

Scoring examples: -0.5 score

–0.5 Original:USAir said that William R. Howard will be elected presidentof USAirBaseline translation:USAir сказал тот Уильям Р. Говард будут избраныпрезидентом USAIRUSAir said that (particular) (demonstr.pron,nom.) William R.Howard will be elected president of USAirDNT-processed translation:USAir сказал того Уильяма Ра. Говард будут избраныпрезидентом USAirUSAir said of that (particular) (demonstr.pron,gen.) WilliamRa. Howard will be elected president of USAir

Scoring examples: -1 score–1 Original:

to discuss the benefits of combining TWA andUSAirBaseline translation:чтобы обсудить выгоды от объединения TWAи USAIR('to discuss the benefits of the merge (noun) (of)TWA and USAir')DNT-processed translation:чтобы обсудить выгоды от объединяющегосяTWA и USAir('to discuss the benefits of the combining(participle, sing.) TWA and (of) USAir')

Manually scored part of the corpus• 50 highlighted strings for each MT system• Gain score: Overall score / Scored differences

Number of:Original– GATE

MT E-RProMT

MT E-FProMT

MT E-FSystran

Paras. withNE

218 225 225 239

Paras. withcontextualdifferences

139(61.8%)

132(58.7%)

207(86.6%)

Paras.manuallyscored

31(22.3%)

28(21.2%)

30(14.5%)

Strings withdifferences

211 212 411

Stringsscored

50(23.7%)

50(23.6%)

50(12.2%)

Results of the experiment

ProMT 1998E-R

ProMT 2001E-F

Systran 2000E-F

Mark N Score N Score N Score+1* 28 = +28.0 23 = + 23.0 18 = + 18.0

+0.5* 2 = +1.0 5 = + 2.5 24 = + 12.00* 4 = 0 7 = 0 8 = 0

–0.5* 3 = –1.5 1 = – 0.5 1 = – 0.5–1* 13 = –13.0 14 = – 14.0 10 = – 10.0

SUM 50 +14.5 50 + 11.0 61 + 19.5Gain +29% +22% +32%

Results for additional 50 strings...ProMT 1998

E-R50: 100:

Mark N Score N Score+1* 28 = +28.0 59 = +59.0

+0.5* 2 = +1.0 8 = +4.00* 4 = 0 6 = 0

–0.5* 3 = –1.5 7 = –3.5–1* 13 = –13.0 31 = –31.0

SUM 50 +14.5 111 +28.5Gain +29% +26%

Improvement in the context of NEs

• Aspects of improvement:– morphosyntactic features and categories– word sense disambiguation – word order and syntactic segmentation

• Consistency in improvement– for both languages – for all MT systems

Examples of improvement

Original:TWA stock closed at $28 …Baseline translation:E-F

Systran Fermé courant de TWA à $28 … (‘Closed (Past participle) current (Noun/Presentparticiple) of TWA at $28 …’)DNT-processed translation:L’action de TWA s’est fermée à $28 … ('The stock of TWA closed (Verb) at $28 …')

Examples of improvement:2

Original:National Mediation Board is expected to release Pan Am Corp. fromtheir contract negotiations.Baseline translation:E-R

ProMT Национальное Правление Посредничества, как ожидается,выпустит Кастрюлю - Корпорация от их переговоровконтракта.('National Mediation Board is expected to release [put on the market]a Saucepan - Corporation from their contract negotiations.')DNT-processed translation:National Mediation Board, как ожидается, освободит Pan AmCorp. от их переговоров контракта.(‘National Mediation Board is expected to release [make free] PanAm Corp. from their contract negotiations.’)

Improvement: languages and systemsOriginal:The agreement was reached by a coalition of fourof Pan Am's five unions.Baseline translation:E-R

ProMT Соглашение было достигнуто коалициейчетырех Кастрюли пять союзов Ама.('The agreement was reached by a coalition offour of a Saucepan five unions of Am.')DNT-processed translation:Соглашение было достигнуто коалициейчетырех из пяти союзов Pan Am.('The agreement was reached by a coalition offour out of five unions of Pan Am ')

Improvement: languages and systems:2Original:The agreement was reached by a coalition of fourof Pan Am's five unions.Baseline translation:E-F

ProMT L'accord a été atteint par une coalition de quatrede casserole cinq unions d'Am.(‘The agreement was reached by a coalition offour of saucepan five unions of Am.’)DNT-processed translation:L'accord a été atteint par une coalition de quatrede cinq unions de Pan Am.(‘The agreement was reached by a coalition offour of five unions of Pan Am.’)

Improvement: languages and systems:3Original:The agreement was reached by a coalition of fourof Pan Am's five unions.Baseline translation:E-F

Systran L'accord a été conclu par une coalition de quatrede la casserole étais cinq syndicats.(‘The agreement was reached by a coalition offour of the saucepan was five trades-unions.’)DNT-processed translation:L'accord a été conclu par une coalition de quatrede Pan Am's cinq syndicats.(‘The agreement was reached by a coalition offour of Pan Am’s five trades-unions.’)

Discussion• Different aspects of MT quality are

interdependent– improvements on one level help other levels

• IE techniques target specific tasks also necessary for the SL analysis stage in MT– NE recognition– co-reference resolution– word sense disambiguation

• MT can benefit from clearly defined evaluation procedures for specific IE tasks

Conclusions and Future Work• NE recognition within IE framework improves

not only treatment of NEs by MT, but also boosts the overall MT quality:– morphosyntactic and lexical well-formedness– features of the wider context of NEs

• Future work: harnessing other focused technologies for MT– co-reference resolution– word sense disambiguation– evaluating the baseline performance of MT systems