Building a parallel corpus for translation research

and much more"

Ana Frankenberg-Garcia

The study of human translation

Traditionally not a hard scienceDifficult to be systematic

With the advances of corpus linguistics,

things can change …

What is a corpus?

large

specific criteriatext-retrieval software

machine-readable

naturally occurring texts

Advantages of using corpora to study human translation

An enormous amount of translated texts

Systematic analyses

Quantifiable results

Corpora used in translation practice and research

1. Bilingual comparable corpora Farmhouse holidays (EN) & Agroturismo (IT)

2. Monolingual comparable corpora Translational English Corpus (EN)

3. Simple parallel corpora Tectra (EN-GL)

4. Bidirectional parallel corpora COMPARA (PT-EN and EN-PT)

Building parallel corpora text selection

• Genre (scientific, imaginative, technical, etc.)

• Mode (oral? written?)

• Variety (standard? regional?)

• Time (contemporary? older?)

• Languages (which? just two or more?)

• Translations (professional? native speakers? different translators? )

• Simple or bidirectional?

Are there translations?

Building parallel corpora example of interrelated factors

PT-EN or EN-PT PT-EN ↔ EN-PT

scientificacademic

tourism

literaturepolitics (EP)

Languages: PT-ENGenreoral popular

Building parallel corpora

Personal use Shared use

copyright permissions

results verifiable

more users and uses

copyright

no hassle

Building parallel corporacopyright

• Two permissions, double the work

• Publishers, authors and translators generally don’t know what a corpus is

• Protect

• Advertise

Building parallel corpora alignment

Text?

Paragraph?

Sentence?

Clause?

Word?

Which parts of ST and TT match?

Building parallel corpora tagsAlignment tags

e.g. textual, grammatical, semantic

What do we want tags for? More pre-processing, less post-processing

Optional tags

<id=EBJT1 1845>Joe watched Robin climb into the trailer and man-handle the calves one by one towards the ramp, their winglike ears pierced with plastic identity tags.

<id=EBJT1 1845>Joe ficou a ver Robin subir para o atrelado e encaminhar as vitelas uma a uma para a rampa, com as suas orelhas, que faziam lembrar asas, furadas e umas etiquetas de plástico a identificá-las.

Our options for

A bidirectional parallel corpus of English and Portuguese

Funding Portuguese Government and European Union (FEDER and FSE) contract ref. POSC/339/1.3/C/NAC

Project leaders Ana Frankenberg-Garcia & Diana SantosResearch assistants Pedro Sousa, Rosário Silva & Susana Inácio

PT Source texts EN Source texts

Corpus structure

EN TranslationsPT Translations

parallel

bi-directional

parallel

PT ENPT1 PT2

EN1 EN2

ST TT

Language varieties

Portugal

Brazil

Angola

Mozambique

UK

US

South Africa

PORTUGUESE ENGLISH

Unbalanced distribution!

Publication dates

1837

2002

1880

1997

1988

1914

Genre

Published fiction other genres

EXTENSIBLE

Portuguese authors

PortugalCamilo Castelo BrancoEça de QueirósJosé Cardoso PiresJosé SaramagoJorge de SenaLídia JorgeMário de CarvalhoSá Carneiro

Brazil Aluísio AzevedoAutran Dourado Chico Buarque Jô SoaresJosé de AlencarMachado de AssisManuel Antônio de AlmeidaMarcos ReyPatrícia MeloPaulo CoelhoRubem Fonseca

MozambiqueMia Couto

AngolaJosé Eduardo Agualusa

English authors

British IslesDavid Lodge

Ian McEwan

Julian Barnes

Joseph Conrad

Joanna Trollope

Kazuo Ishiguro

Lewis Carrol

Mary Shelley

Oscar Wilde

United StatesHenry JamesEdgar Allan PoeRichard Zimler

South AfricaNadine Gordimer

Portuguese translators

Ana Maria Amador, Ana Falcão Bastos, Ana Luísa Faria, Aníbal Fernandes, Carlos Grifo Babo, Cristina Ferreira de Almeida, Cristina Rodriguez, Eduardo Guerra Carneiro, Fernanda Pinto Rodrigues, Geraldo Galvão Ferraz, Helena Cardoso, Januário Leite, José Viera Lima, J. Teixeira de Aguilar, Lídia Cavalcante-Luther, Lucinda Santos Silva, Luís Lobo, Manuel João Gomes, M. F. Gonçalves de Azevedo, Maria Carlota Pracana, Maria do Carmo Figueira, Mário Martins de Carvalho, Nina Videira, Paula Reis, Yolanda Artiaga.

English translators

Adria Frizzi, Alan Clarke, Alexis Levitin, Alice Clemente, Cliff Landers, David Brookshaw, David Rosenthal, Elizabeth Lowe, Ellen Watson, Helen Caldwell, Giovanni Pontiero, Graeme Mac Nicoll, Gregory Rabassa, Isabel Burton, John Gledson, John Parker, John Byrne, John Vetch, Margaret Jull Costa, Mary Fitton, Natália Costa, Peter Bush, Richard Zenith, Ronald W. Sousa.

Can any text be included in the corpus?

Only published source texts and translations

Only English translated directly from Portuguese

Portuguese translated directly from English

Only human translations!

72 source texts (extracts)

75 translations

Texts

Size

1,549,551 1,436,493words words in in English Portuguese

Possibly the largest existing edited parallel corpus

Interface

Free

Easy to use by people who have never heard of corpora before

Powerful and flexible tool for experienced corpus users

Results good for research and education

www.linguateca.pt/COMPARA/

“nodded”

ST

TT

0

2

4

6

8

10

12

14

100 K words

Distribution of “nodded” in source texts and translations

Users and uses

Language learners and anyone working with PT-EN bilingual dictionary with examples

Language teachers exercises and tests

Translators language equivalents

Translation lecturers exercises & problems

Translation theorists test translation hypotheses

Lexicographers bilingual dictionaries

Computational linguists and language engineers machine translation and other applications

Backstage options

Text tags

EBJB1.ptele revelou-me o seu interesse por Gosse <tnote> Edmund William Gosse (1849-1928), crítico inglês </tnote> e pela sociedade literária inglesa dos finais do século passado.

EBDL2T1.enWhen we sat on the sofa together to watch <title>News at Ten</title>

EBDL1T1.pt passou-me uma receita de <named> Valium </named>

EBJB1.en the white bear, <foreign> thalassarctos maritimus </foreign>, is the aristocrat of bears...

EBDL1T1.ptacaba por se esquecer de ter medo, até que acaba por verificar que não há <emph> de que </emph> ter medo.

Text tags

1 alignment unit = 1 source-text sentence

S

S

S

S

S2

S S(+S)

S

S½

Ø

ST TT

Alignment options and tags

Portuguese: PALAVRAS

Petrus/PROP pediu/V_fmc a/DETartd especialidade/N da/PRP+DETartd casa/N --/PU uma/DETarti paella/N valenciana/ADJ --/PU que/SPECrel comemos/V em/PRP silêncio/N ,/PU acompanhados/V apenas/ADV do/PRP+DETartd saboroso/ADJ vinho/N Rioja/PROP ./PU

Grammar tags

[pos="V.*"] "silêncio"

English: CLAWS (coming soon)

Petrus/NP1 asked/VVD for/IF the/AT specialty/NN1 of/IO the/AT house/NN1 --a/AT1 Valencia/NP1 paella/NN1 --which/DDQ we/PPIS2 ate/VVD in/II silence/NN1 ./.

Grammar tags

I did, too --changed over to the knitted tie at a <sem=“cor”> red </sem>light.

People interested in creating specific tags for their research can do so, as long as they do the tag insertion and revision work

Specific tag revision interface underway (Sousa, in preparation)

e.g. semantic tag for colour (Inácio et al. 2007)

Other tags

1. Observing source texts and translations

2. Constrasting Portuguese and English

3. Comparing translated and untranslated language

4. Examining the characteristics of translated texts

Research work

Studies unthinkable before corporaMany other studies possible!

www.linguateca.pt/COMPARA/ComparaPublications.html