Revisingmachinetranslation a marketableskill Belinda Maia Universidade do

Revisingmachinetranslation – a marketableskill Belinda Maia Universidade do Porto 6 June 2013 1

Overview Perspective: what to do when you have limited access to MT Teaching revision of MT Related technology ◦ General introduction to MT ◦ Rule based MT (RBMT) - language study – an experiment ◦ Revision of Statistical MT (SMT) – e. g. Google Translate and Bing ◦ Revision of ‘mixed’ systems - http: //itranslate 4. eu/en/ ◦ Corpora – OPUS, Statistical ‘dictionaries’ – Lingueehttp: //www. linguee. com/ Making your own SBMT system – using parallel corpora and translation memories – Moses for Mere Mortals 2

Revision European Norm 15038 ◦ General linguistic revision ◦ Special domain revision Revising and Editing for Translators (Mossop, 2007) draws attention to ◦ Different levels of revision of Human Translation (HT) ◦ Different demands / needs for revision of HT 3

Revision of MT Different levels / needs /demands Remember this when revising MT Accept that revising MT may require balance between acceptable <> good (<> excellent? ) according to needs Accept use of MT for rough or gist as a commercial possibility 4

Teaching about MT A short history of MT The main attempts to create MT ◦ RBMT – Rule Based Machine Translation ◦ SBMT – Statistically Based Machine Translation MT has opened up the mere possibility of translation to millions = Multilingualism rather than ‘lingua franca’ = Need for a positive attitude to MT MT as a tool for translators The interest of MT as an object of research 5

Experiments with RBMT 2003 -5 Linguateca project – http: //www. linguateca. pt METRA – Systran, Babelfish, Altavista, Amikai… BOOMERANG EN>PT>EN>PT……. Tr. Ava – analysis of EN>PT MT Objectives ◦ Raise awareness of grammar rules and HT v. MT > similar problems to HT into 2 nd language ◦ Test EN<>PT MT 6

Revision of Statistical MT (SMT) Free online - Google Translate, Bing Warn students of copyright and security problems …. Also detection of plagiarism! HOWEVER… It appears the majority of users want it as a dictionary or for terminology (Hutchins – 2010) But revising the results tells us a lot about MT 7

News items – PT (3 June 2013) A polícia checa confirmou nesta segundafeira a morte de cinco pessoas devido ao efeito das chuvas. Um homem e uma mulher morreram domingo no desmoronamento de uma casa em Trebenice, 30 quilómetros a Sul de Praga. Dois outros corpos sem vida foram encontrados no Norte do país e um homem morreu em Benesov, 40 quilómetros a Sul de Praga. 8

RBMT The Czech police confirmed in this Monday the death of five people due to the effect of rains. A man and a woman had died Sunday in the collapse of a house in Trebenice, 30 kilometers south of Praga. Dois other bodies without life were found in the North of the country and a man died in Benesov, 40 kilometers south of Plague. 9

Result? Fair 10

Google Translate > EN (same day) The Czech police confirmed on Monday the death of five people due to the effect of rainfall. A man and a woman died Sunday in a house collapse in Trebenice, 30 kilometers south of Prague. Two other dead bodies were found in the north and a man died in Benesov, 40 kilometers south of Prague. 11

Result? Very good Possible plagiarism? ! 12

Enciclopedia item - PT Descendente, pelo lado de seu pai, de uma família de raízes rurais de Entre Douro e Minho, desde muito nova que se interessou por livros, começando por ler alguns da biblioteca do avô materno. Foi através destas primeiras leituras que tomou contacto com alguns dos melhores escritores franceses e ingleses, os quais lhe despertaram a arte narrativa. Em 1932 vai para o Porto estudar, onde passa parte da adolescência, mudando-se para Coimbra em 1945, e, a partir de 1950 fixa definitivamente a sua residência no Porto. . 13

RBMT Descendant, for the side of its father, of a family of agricultural roots of Between Douro and Minho, since very new that were interested for books, starting reading some of the library of the maternal grandfather. Foi through these first readings that took I contact with some of the best French and English writers, which it despertaram the narrative art. In 1932 Porto goes to study it, where it passes part of the adolescence, changing itself for Coimbra in 1945, and, from 1950 itdefinitively sets its residence in Porto. 14

Result Very weak But absent subject > ‘it’ 15

Google Translate > EN Descending the side of his father, a family of rural roots of Entre Douro e Minho, from very young to get interested in books, starting with reading some of the library's maternal grandfather. It was through these early readings that [? ] came into contact with some of the best writers in French and English, which hearoused the narrative art. In 1932 [? ] goes to Porto study, where he spends part of [? ] adolescence, moving to Coimbra in 1945, and in 1950 finally fixed his residence in Porto. 16

Result Marginally better? At least with certain phrases But sexist! The article was about Agustina. Bessa-Luis, a well known Portuguese writer – but a woman! 17

SO …. MT works for some texts better than others Practice with MT will show: ◦ When it will (not) help the translator ◦ Why it does (not) translate reasonably well ◦ How far it can be useful – e. g. With specific text types For terminology For ‘set phrases’ etc Knowledge of its strengths and limitations will allow revision of MT as a marketable skill 18

Using related technology Statistical ‘dictionaries’ – e. g. Linguee http: //www. linguee. com/ Corpora – e. g. OPUS corpora http: //opus. lingfil. uu. se/ Making your own SBMT system – using parallel corpora and translation memories See Moses for Mere Mortals http: //code. google. com/p/moses-for-meremortals/ 19

Moses for Mere Mortals You need: Linux – Ubuntu 10. 4 -11. 2 on your computer At least 4 GB RAM and enough ‘storage’ Corpora ◦ Either - the bigger the better - e. g. download OPUS & EC corpora ◦ Or – smaller – your own TMs Training ◦ Your computer ◦ Yourself ◦ Your private MT system! 20

Our experience Course (14 hours) given by Hilário. Fontes (DGT- Luxembourg) – Portuguese section (pioneers in MMM!) Problems with ◦ Linux ◦ Actual use of MMM But. . we survived Students tended to find it easier than the teachers! 21

Our experience Master’s students. ◦ Motivated to learn more ◦ Understood implications for their future Ph. D students plan research in this area ◦ Testing the system ◦ Adapting results between Portuguese and Brazilian Professional translators ◦ Plan to train their TMs with MMM Several marketable skills 22

Future developments We hope to Encourage a more positive attitude to MT Develop understanding of MT Extend this knowledge to a larger group Make use of material supplied 23

References Hutchins, John, 2010 ‘A short history of Machine Translation’ & several more articles - see http: //www. hutchinsweb. me. uk/ Koehn, Philip. 2010. Statistical Machine Translation. Cambridge: CUP. Mossop, Brian. 2007. Revising and Editing for Translators. Manchester: St. Jerome Pub. Bing Translator - http: //www. bing. com/translator Free Online Translations Online Language Translator http: //itranslate 4. eu/en/ Google Translate - http: //translate. google. com/ Linguee - http: //www. linguee. com/ Moses for Mere Mortals http: //code. google. com/p/moses-for-mere-mortals/ OPUS corpora - http: //opus. lingfil. uu. se/ 24

If I have time. . Google Translate is learning…. 25

News item (PL in 2009) Ksidz kradl wódke w Biedronce Ksiadz Pawel z Kesowa nie ma watpliwosci. To reka boska sprawila, ze zoztal przylapany na kradziezy wodki w Biedronce. Bóg zeslal na niego wieczny wstyd i ponizenie, bo duchowny ulegl grzesznej pokusie i chcial w Wielkim Poscie napic sie gorzalki. 26

GT (2009 version) Chaplain, Fr Peter steals vodka in Biedronka Rev. Pawel bites there is no doubt. It is a divine hand has made zoztal caught stealing vodka in Biedronka. God sent down upon him eternal shame and humiliation, he has met the priest wanted to temptation and sin in Lent drink some vodka. 27

GT - 2013 version Chaplain steal vodka in Biedronka Paul Priest of billets there is no doubt. It's God's hand made the zoztal caught stealing vodka in Biedronka. God sent down to the eternal shame and humiliation, because the minister succumbed to temptation and wanted to sin during Lent drink some vodka. 28