DIY corpora: the WWW and the translator

Federico Zanettin

Abstract

The WWW is the single largest existing repository of electronic texts, and has recently attracted the attention of researchers involved in translator training as a suitable source of texts for the creation of "disposable corpora". These are small, specialized corpora created ad-hoc to serve the needs of the translator for a specific translation project, and their value lies not only in their analysis but even more so in their creation. This approach complements a number of studies which have been carried out on the use of small corpora for language learning and translator training, where the main focus is on methods and techniques for analysing texts already collected by the teacher. This paper presents an experiment which was carried out at the School for Translators and Interpreters of the University of Bologna in Forlì with third and fourth year translation students in the context of a course on computer assisted tools. Students were given a text to translate and asked to search the Internet, select suitable web pages in the target language, and download them on disk. In this way, while cyclically performing the translation and adding material to the corpus as the translation proceeds, they were able to familiarize themselves with the topic of the translation at hand, to select texts according to text type, to assess the reliability of text sources and evaluate the perspective readership. These DIY corpora were then browsed switching between a full text mode and a concordancing, and learners were able to tackle many translation problems related to specific terminology and phraseology..

1 Introduction

Traditionally, translators have used "parallel texts", i.e. collections of printed texts produced in similar communicative situations, as a way of checking text-typological conventions in the source and target languages. In the last few years information technology has brought about a completely new scenario. The availability of vast quantities of texts in many languages and on all kinds of subjects is a dream come true for translators as well as for all types of discourse professional, text processors and language services providers.

The WWW is the largest existing repository of texts. The number of publicly accessible web pages has reached 2 to 4 billions (Fletcher 2001). The vast majority of these are in English (about 80% according to estimates), but the number of users whose first language is not English is increasing (Fletcher 2001), and the pages in languages other than English are increasing at a faster pace than pages in English. Table 1 gives an estimation of the growth of the main European languages between October 1996 and February 2000, sorted according to their rate of growth (which is shown in the last column).

	Millions of words (October 1996)	Millions of words (February 2000)	Growth factor
Spanish	104	1,894	18
German	229	3,333	15
French	223	2,732	12
Italian	124	1,338	11
Portuguese	106	1,161	11
Norwegian	106	947	9
Finnish	21	166	8
English	6, 082	48, 064	8

Table 1: Growth of languages on the Web (data from Grefenstette and Nioche, 2000)

For instance, while the number of words in Italian available on the web in February 2000 was eleven times larger that of October 1996, the number of words in English increased "only" by eight times.

With availability comes ease of access: A few clicks of the mouse are often worth several trips to the library or consultations with clients and colleagues. There is also the added bonus that electronic texts can be analysed through corpus linguistics techniques rather than just read sequentially, thereby uncovering linguistic information which would be otherwise very difficult to obtain.

Recent research in translation studies has stressed the contribution which corpora of electronic texts can bring to translators. By using appropriate software translators can look up words in a matter of seconds, and highlight patterns by sorting contexts around search words. If a corpus is appropriately designed, it can provide reliable evidence of authentic linguistic behaviour and text-structuring conventions by highlighting recurrent patterns. Terminological and collocational information can be especially useful.

Experiments in the literature have reported the uses of bilingual corpora and monolingual corpora (Zanettin 1998, 2001; Bowker 1998, 2000; Pearson 1998; Gavioli and Zanettin 2000) as sources to compile term banks, and as aids during the translation task. One problem with these typically small and domain specific corpora is the limited range of topics and text types for which they are available. Recent work has concentrated on increasing this range and availability, and a number of well crafted bi- or multilingual corpora, comparable and/or parallel, such as the English Norwegian Parallel Corpus, the COMPARA corpus and the CEXI corpus are complete or under way. Some work has also been done using large monolingual corpora such as the BNC (Stewart 2000, Bernardini 2001). These resources will certainly prove to be very useful. However, even the 100 million words BNC is ill-equipped to meet the needs of translators working with very specialised texts and confronted with specific terminology.

But now - for this - translators can turn to the web.

There are two sets of problems related to the use of the WWW documents as corpus material. The first concerns procedures for assessing relevance and reliability: Information is dispersed in the WWW through vast quantities of documents, and it is thus crucial for the translator to retrieve this information in the most efficient and effective way.

The second relates to strategies and techniques for searching electronic texts: Search engines provide access points to Internet documents either through lists generated by full text searches or by pre-selected lists organized by topic, and are thus catalogues rather than corpora.

The WWW is not a corpus, but it can be used as a corpus. Some search engines (e.g. Google, Copernic) display, next to document pointers (hyperlinks), a concordance-like context with the search word(s) highlighted. Some applications which, building on search engines, are designed more specifically for the needs of language professionals are also available on the web. WebCorp (Kilgarriff 2001) and KWICFinder (Fletcher 2001), for instance, download and produce keyword in context abstracts of web pages which match the user's search criteria. KWICFinder permits more targeted searches by using a number of restricting criteria and allows for the display of the output in a number of formats. These enhancements go a long way towards solving the problem of using the web as a suitable source of text-linguistic information, but they still do not solve the problems of the relevance and the reliability of the document abstracts retrieved. The Internet is full of a large number of ephemeral texts of dubious authorship and authority, and the relevance criteria of search engines are different from those of translators of specific texts (Fletcher 2001).

In this paper I would like to take another approach, which has been already explored by a number of researchers and trainers (e.g. Pearson 1998, Maia 2000, Varantola 2000, Bertaccini and Aston 2001) and look at the web as a source of texts for a DIY corpus.

A DIY web corpus can be characterized as follows:

- it is a collections of Internet documents, or more precisely of web pages in HTML.

- it is created ad hoc as a response to a specific text to be translated

- it is an open corpus. More material can be added as the need arises

- it is disposable (Varantola 2000) or virtual (Ahmad et al.1994). It is not destined to be part of a more permanent corpus, and can be disposed of as soon as the translation is completed. Copyright permissions are not required

- like "parallel texts" it can be either bilingual comparable or target monolingual.

In the following sections I report on an experiment with DIY corpora at the school for translators and interpreters of the University of Bologna at Forlì with a group of advanced translation students.

2 The experiment

This experiment was carried out within a course of CAT tools, which comprised a number of different modules and was designed as a general introduction to computer assisted translation, providing an overview of existing technologies available to professional translators such as terminology management systems and translation memory tools, and of resources such as online term banks, machine translation programmes, mailing lists for translators, etc. One module, of which this experiment was a part, was on "corpus management in translation", defined by Varantola (forthcoming) as "the knowledge and skills needed in the compilation and use of corpus information for individual translation assignments".

At the time of the experiment, which took place over two weeks in five two-hour sessions, the students had already been exposed to the use of various professional tools and online resources, and while not all of them were skilled Internet users, only one of them was a novice. They were also already familiar with the main features of WordSmith Tools (Scott 1996), the corpus analysis programme used in the experiment.

In preparation for the task students were given a brief survey of the tricks and treats of Internet Explorer and WordSmith Tools, and a task sheet was distributed with operational instructions and a set of quick reference guidelines. As regards the browser, for instance, students were instructed to take advantage of the "chronology" feature, which lists all visited documents in a side window together with their title and address and allows for off-line browsing. They were told to open documents in multiple windows and save all relevant web pages in individual "corpus folders" on the hard disk of their computers, in order to use them as a corpus to be analysed with the concordancer. They were also given a list of some search engines (some international, e.g. Google, Altavista, Yahoo, and some Italian, e.g. Virgilio, Arianna, Kataweb). Those who translated into/from German, French or Spanish found language/area specific search engines.

Students where asked to engage in a real translation task, using a DIY corpus as a resource to help them translate either a text which they had as an assignment from another course or - for some of them - as a paid translation job. No restrictions were given as to source and target languages. Two additional texts (one in English, one in Italian) were provided for those who didn't have a better text handy.

The following are some of the texts which students translated

- an encyclopaedia article on prostate cancer, from English into Italian

- (part of) a textiles catalogue (web page), from Italian into Spanish

- (part of) a bicycle locks catalogue (web page), from Italian into English

- an article on earthquakes from a science magazine, from Italian into English

- a promotional leaflet on diamonds, from English into Italian

Students were encouraged to translate their text using a number of tools they were already familiar with, such as online terminology banks (e.g. Eurodicautom, Logos), translator workbenches (e.g. Trados, Déjà Vu, WordFisher), electronic dictionaries (e.g. Babylon), etc. After setting up their workstation (opening the text editor/workbench, Internet browser and WordSmith Tools), they read the source text and started their translation. They also began to search the Web for suitable texts to include in their corpus and help them in the specific translation task. Some students chose to work alone, others worked in pairs or groups of three. The teacher acted as a facilitator, helping to solve problems. At the end of the experiment the students were asked to write a brief report on the benefits and shortcomings of creating and using a DIY web corpus as a translation resource.

A first step usually consisted in trying to get a better understanding of the source text. To this end, some students focused on unknown terminology, using paper and electronic dictionaries, term banks and other online resources. Online glossaries, usually found by searching for the word "glossary" along with words identifying the topic (e.g. "diamonds" or "prostate cancer") were reported to be especially useful. By checking for equivalents in source and target language glossaries students were able to identify some key terms to be used as key words for searching for relevant corpus candidates. Other students started by looking at "Internet directories", i.e. lists of web sites organized into categories which are provided by "portals". A combination of the two techniques with the use of multiple search engines seemed the most useful strategy.

Students were free to choose the type and number of texts to download from the Internet and include in their corpus. They were advised to open all candidate documents, scanning them for content rather than for specific linguistic items, and save relevant pages in their corpus folder. After having found and saved a first group of texts, students used WordSmith Tools to analyse the corpus while proceeding with the translation. As the translation proceeded they added more material, refining searches according to their needs. The size of their corpora eventually varied between 10 and 50 texts, or 5,000 to 40,000 words.

The relevance of a document was usually decided after scanning the text for both verbal and visual clues, such as titles, layout and images. Decisions were taken at the level of overall content, text type, style and register. For instance, the student translating a catalogue of luxury textiles discarded a number of texts from the web site of a museum dealing with African traditional textiles and clothing; the student translating a bicycle locks catalogue discarded a newspaper article on urban safety.

As for reliability, students discarded bad translations (into which they sometimes ran by using tentative translation equivalents as search terms) and privileged texts produced by recognizable entities and authorities within the relevant discourse communities (experts, producers, public and private agencies and associations).

Finding useful web pages in the target language was also an exercise in audience design, giving a change to students to form an idea of the potential perspective readership of their translations.

Having created their corpus, they still had the problem of how to find the information they were looking for, but they could use the knowledge derived from their prior acquaintance of the texts to conduct searches around known equivalents or come up with informed hypotheses.

The corpus was mainly used for finding information on terminology, phraseology and collocations. For instance, after having established while constructing the corpus that an "antifurto per biciclette" is a "bicycle lock", it was easy for a student to get a list of different types of locks (cable lock, coiling cable lock, U-lock, disc lock, etc.) by sorting the output of a concordance for "lock" according to the first word to the left. The student translating a scientific report on earthquakes learned that while in Italian both walls and buildings "crollano" in English buildings usually "collapse" while the word "wall/s" collocates more frequently with "fall/s". When looking for a translation for "cedimenti strutturali gravi" one student generated a concordance of "structural*" and quickly found that she could use the phrase "heavy structural damage".

When they were uncertain or presented with multiple translation candidates, students relied on frequency of occurrence as and indicator of reliability, stressing that since the texts in the corpus were carefully selected, it was unlikely that they would produce spurious examples.

Some students resorted to concordancing mainly while revising the translation. That is, they first wrote a draft of the translation while finding parallel Internet texts, then went through their text checking their hypotheses and intuition against the corpus with WordSmith Tools.

3 Benefits and problems

In their reports, many students noted the advantage of a corpus of electronic texts over more traditional reference material. While in paper dictionaries the information is usually buried in small-type heavy columns, web pages often contain images and other multimedia features which aid understanding. They stated that constructing the corpus was as useful as generating concordances from it, and that they often went back to view a web page in full after looking at concordance lines.

However, in line with similar observations made by both trainee and professional translators who participated in similar experiments (Varantola forthcoming, Jääskeläinen and Mauranen 2000), students complained about the lack of user-friendliness of WordSmith Tools. Despite its many capabilities, the current version of WordSmith Tools is still not fully equipped to work with tagged texts written in HTML/XML and, while it is possible to exclude tags from view, it is not possible to jump from a concordance line to the corresponding web page. To take advantage of important information from layout and images, students had thus to switch between a concordance in WordSmith Tools and the corresponding web page, with each file having to be located by its name and opened in the browser.

One group of students spent much time inspecting web pages for specific terminology rather than looking for suitable texts. In this respects, they used the web itself as corpus rather than as a source for creating a corpus. They spent more time reading individual documents and looking for exact equivalents rather than deciding whether a text was likely to contain useful terminology and save it for later inspection with the concordancer. They also felt frustrated when they later found that, having saved too few documents, their corpus was not large enough to be of use.

Students noted that searching for web pages, creating the corpus and analysing it with the concordancer was time-consuming, and argued that the translation task would have taken less time if done with dictionaries alone. But they also stated that they felt more confident about the solutions adopted, especially in translating into the foreign language, and that the balance between costs and benefits would be different with longer translation assignments.

Other observations by students concerned the use of the web as a corpus resource: while some believed it would be mostly useful when translating into a foreign language, others chose to translate into their first language. While all students created a target language corpus not all of them created a source language one. [1]

4 Conclusions

DIY corpora are one of a number of different types of corpus resources which translators can use in their work. [2]

Research on corpus use in translator training environments generally takes a bottom-up approach, which could be termed "from words to texts". This approach is mostly concerned with finding appropriate ways of analysing corpus resources provided by the teacher, be they large monololingual corpora or smaller mono or bilingual corpora, created ad hoc either from electronic text archives on CD-ROM or from printed sources.

This approach has been complemented by a top-down approach, i.e. one going "from texts to words", which assumes no pre-existing corpus to be analysed, and which has been made possible thanks to the availability of large quantities of texts on the Internet. By compiling their DIY corpus prior, during or even after the translation task (Aston 2000), students (and translators) can get a first acquaintance with texts, and take full advantage of web pages prior to word prompted analysis.

Hopefully software producers and developers will create professional applications in which the functions of browser and concordancer will be better integrated, and DIY will find their place in the translator's workstation together with other corpus resources and computer assisted tools.

Notes

[1] An assessment of the of the quality of the assignments was outside the scope of the experiment. However, better translations seemed to have been produced by those students who adopted successful strategies in the creation and analysis of the corpus.

[2] Another type of corpus resources are translation memories, which are a very specialized kind of parallel corpus, and are usually relevant, reliable and well integrated into the translation work-flow. But of course translators do not have a translation memory ready for all occasions.

References

• Ahmad, K., P. Holmes-Higgin, R. Sibte Abidi (1994) "A description of texts in a corpus: "Virtual" and "real" corpora". In Willy Martin, Willem Mejis, Margaret Moerland, Elsemiek ten Pas, Piet van Sterkenburg and Piek Vossen (eds.) EURALEX 1994 Proceedings. Amsterdam: Vrije Universiteit, 390-402.
• Aston, Guy (2000) "I corpora come risorse per la traduzione e l'apprendimento". In Silvia Bernardini and Federico Zanettin (eds.) I corpora nella didattica della traduzione. Bologna: CLUEB, 21-29.
• Bernardini, Silvia (2001) "Spoilt for choice: A learner explores general language corpora". In Guy Aston (ed.) Learning with corpora. Houston, TX: Athelstan, 220-249.
• Bertaccini, Franco and Guy Aston (2001) "Going to the Clochemerle: Exploring cultural connotations through ad hoc corpora". In Guy Aston (ed.) Learning with corpora. Houston, TX: Athelstan, 198-219.
• Bowker, Lynne (1998) "Using Specialized Monolingual Native-Language Corpora as a Translation Resource: A Pilot Study", in META, 43: 4, 631-651.
• Bowker, Lynne (2000) "Towards a methodology for exploiting specialized target language corpora as translation resources, in International Journal of Corpus Linguistics, 5: 1, 17-52.
• Fletcher, William (2001) "Concordancing the web with KWiCFinder", presentation given at the Third North American Symposium on Corpus Linguistics and Language Teaching, Boston, MA, 23-25 March 2001. Available at http://miniappolis.com/KWiCFinder/Corpus2001.htm
• Grefenstette, Gregory and Julien Nioche (2000) "Estimation of English and non-English language use on the WWW". In Proceedings of the RIAO (Recherche d'Informations Assistee par Ordinateur), Paris.
• Gavioli, Laura and Federico Zanettin (2000) "I corpora bilingui nell'apprendimento della traduzione. Riflessioni su un'esperienza pedagogica". In Silvia Bernardini and Federico Zanettin (eds.) I corpora nella didattica della traduzione. Bologna: CLUEB, 61-80.
• Jääskeläinen, Riitta and Anna Mauranen, (s.d.) "Work Package 5: Development of a Corpus on the Timber Industry - Final Report", Project SPIRIT MLIS-programme: MLIS-3008 SPIRIT 24637, University of Joensuu, Savonlinna School of Translation Studies.
• Johansson, Stig (1998) "On the role of corpora in cross-linguistic research". In Stig Johansson and Signe Oksefjell (eds.) Corpora and cross-linguistic research: Theory, method and case studies. Amsterdam and Atlanta: Rodopi, 3-24.
• Kilgarriff, Adam (2001) "Web as corpus". In Paul Rayson, Andrew Wilson, Tony McEnery, Andrew Hardie and Shereen Khoja (eds.) Proceedings of the Corpus Linguistics 2001 conference, UCREL Technical Papers: 13. Lancaster University, 342-344.
• Maia, Belinda (2000) "Making corpora: A learning process", In Silvia Bernardini and Federico Zanettin (eds.) I corpora nella didattica della traduzione. Bologna: CLUEB, 47-60.
• Pearson, Jennifer (1998) Terms in context. Amsterdam & Philadelphia: Johns Benjamins.
• Pearson, Jennifer (2000) "Surfing the Internet: teaching students to choose their texts wisely". In Lou Burnard and Tony McEnery (eds.) Rethinking Language Pedagogy from a Corpus Perspective. Frankfurt am Maim et al: Peter Lang, 235-239.
• Scott, Mike (1996) WordSmith Tools. Oxford: Oxford University Press.
• Stewart, Dominic (2001) "Conventionality, creativity and translated text: The implications of electronic corpora in translation". In Maeve Olohan (ed.) Intercultural faultlines. Manchester: St Jerome, 73-91.
• Varantola, Krista (2000) "Translators, dictionaries and text corpora", In Silvia Bernardini and Federico Zanettin (eds.) I corpora nella didattica della traduzione. Bologna: CLUEB, 117-133.
• Varantola, Krista (forthcoming) "Translators and disposable corpora". In Zanettin, Federico, Silvia Bernardini and Dominic Stewart (eds.) Corpora in translator training (provisional title). Manchester: St Jerome.
• Zanettin, Federico (1998) "Bilingual Comparable Corpora and the Training of Translators", in META, 43: 4, 616-630.
• Zanettin, Federico (2001) "Swimming in words: corpora, language learning and translation". In Guy Aston (ed.) Learning with corpora. Houstox,TX: Athelstan, 177-197.