TransSearch: a Bilingual Concordancer

A bilingual concordancer

TransSearch is a search engine somewhat like Google, except that it is applied to millions of pairs of sentences that are mutual translations. A TransSearch query, expressed as a single word or as an expression, returns not only all the sentences in its database that contain those words but also the corresponding sentences in the other language. TransSearch is thus a tool that enables translators to submit queries to a translation memory, in order to locate ready-made solutions to all sorts of translation problems.

translation memory is a textual database made up of groups of documents that are mutual translations and in which the various links between translated segments are explicitly recorded. In the simplest case, the database contains documents in only two languages, say English and French, and includes a translation for every document. More complex configurations are possible (e.g. multilingual or partial translation memories), but for the time being, let us restrict ourselves to the simplest case.

TransSearch allows users to search translation memories, which we call TransBases, and within which documents are aligned at the sentence level. There are various methods that can be employed to automatically produce these kinds of alignments with a high level of accuracy. The user launches a search by submitting a query to the system. The relatively simple query syntax makes it easy to describe expressions and terms. The results are presented in the form of lists of pairs of sentences that satisfy the query.


In 2001, RALI, lead by Elliott Macklovitch, launched, that features a judicial database and the Hansard database from 1986 to today . Since November 2003, this service is now handled by Terminotix.


Example of translations in French of take for a ride (click on the images to enlarge them). The first shows the output of the first version of TransSearch that returns only the sentence in which the queried expression has been found and the corresponding sentence in the other language. The second picture shows the output of the more recent version of TransSearch that also identifies the span of target text corresponding to the expression in the source language; the correspondances are then grouped in a list shown at the left for a fast exploration of the context of uses of these translations. 

Research project progress

Between May 2008 and April 2011, RALI and Terminotix, with the support of a Cooperative Research and Development grant from NSERC, completely rebuilt TSRALI and its internet interface. This new version is now available at Terminotix. 

Translation spotting was the main focus of research. Two post-doctoral fellows (Julien Bourdaillet and Stéphane Huet) developed, implemented and evaluated many algorithms in order to find an appropriate tradeoff between precision and speed. Another scientific challenge was merging equivalent translations, i.e. translations that are grammatical variations of one another or containing grammatical words that do not change the essence of a same translation. This work lead to many scientific presentations in refereed conferences and to an article in Machine TranslationThis presentation describes the main features of the new TransSearch.

As this new feature implied changing the underlying text management system, we also redesigned the architecture of the system in order to take into account the new interaction facilities of modern browsers. After consultations with our industrial partner, two research associates (Jacques Steinlin and Fabrizio Gotti) developed a new user interface that allow a user to query the system but also to a group manager to deal with subscribers of his group and to the system manager to deal with all system parts. Translation spotting and the new web architecture will make it easier to embed TSRali within new applications.

This project also involved 4 PhD students and 3 MSc students that dealt with various subproblems. The system has been competed and delivered to the industrial partner in April 2011 who then completed localisation and user migration.

A technology transfer agreement between Université de Montréal and Terminotix has been finalised in order to continue this fruitful collaboration between the RALI and Terminotix of more than 12 years.

Publications about TransSearch

Contact : Guy Lapalme.