AppsApps

Who are Treq developers?

Design and implementation: Martin Vavřín, Alexandr Rosen

Tools for word-to-word alignment and wordpairs extraction: GIZA++ (Och, F. J. – Ney, H. (2003). A systematic comparison of various statistical alignment models. Computational Linguistics, 29(1), 19–51.). Our thanks are due to Ondřej Bojar and David Mareček for advice on the installation. The results of the automatic excerption were not reviewed

We are grateful to Elżbieta Kaczmarska for inciting the development of Treq.

User interface: Martin Vavřín

Data extraction: Pavel Procházka, Martin Vavřín

Design support: Jan Kocek

How to cite Treq:

  • Vavřín, M. – Rosen, A.: Treq. FF UK. Praha 2015. Available on WWW: "http://treq.korpus.cz".
  • Škrabal, M. – Vavřín, M. (2017): Databáze překladových ekvivalentů Treq. Časopis pro moderní filologii 99 (2), s. 245–260.

Version 0.1 (alpha)

Release date: September 2014
  • searching through results from Core of Intercorp

Version 0.2 (beta)

Release date: August 2015
  • adding the data from other parts of Intercorp (collections)
  • user interface redesigned
  • Integration of Treq in to the toolbar at korpus.cz main page

Version 0.3 (beta)

Release date: September 2015
  • English translation of the Interface
  • links to the respective queries in Kontext concordancer

Version 1.0

Release date: October 2015
  • final inteface polish and official release

Version 1.1

Release date: January 2016
  • added the proportion column in the result table
  • searchs logging

Version 2.0

Release date: March 2017
  • redesigned language selection
  • addition of English-foreign dictionaries
  • addition of multiword expressions
  • you can newly choose any or all collections
  • you can now use RegEx expressions
  • option to ignore case of the query
  • sum of all frequencies at the bottom

Version 2.1

Release date: July 2020
  • upload of new data from corpus Intercorp v. 12
  • addition of Spanish-foreign dictionaries
  • addition of hyphenated words

Version 2.2

Release date: January 2023
  • upload of new data from corpus Intercorp v. 15


© 2015, ICNC

Restrict search to core/collections

Choose the type of data to be included in your search.

What is core and what are collections?

Search in the lemmatized version of the database

With this option checked, Treq will use dictionary forms both for the search and results (cat-Katze, nice-gentil, sing-cantar) unlike the non-lemmatized option (cats-Katzen, nicer-plus gentil, sang-cantaban).

What is lemma?

What languages are lemmatized?

Include multiword expressions

With this option checked, Treq will search a different database which includes multiword expresions. (Note: This type of querying has a higher error rate.)

Use regular expressions in the query

Regular expresions are special characters with a specific function (e.g. a dot stands for any one character) enabling complex querying. (Note: The results may take longer to compute.) Without checking this option, you can still use '%' as a substitute for a word or a word part (sing% finds the following word forms: singing, sings, singer etc.)

Which regular expressions can I use in Treq?

Case insensitive query

With this option checked, Treq will return all forms of the word regardless of the capitalization (cat will find CAT, Cat etc.).

Ver. 2.2
Source language Target language Restrict to ?
? ? ? ?

Help

Are you wondering how to best translate a word? Do you need to come up with a synonym or other suitable expression? Try Treq! Treq is a collection of bilingual dictionaries, built automatically from the InterCorp parallel corpus. The dictionaries are bidirectional, with all languages represented in the corpus on one side and Czech, English or Spanish on the other side.

To use Treq, start by specifying the desired language pair by selecting source language (the language of the query) and target language (the language of the potential equivalents). The query can be entered either as a specific word form, as a lemma (Lemma), as a multiword unit (Multiword) or using regular expressions (Regex). The query can also be made case insensitive (A = a). Depending on the (Restrict to:) parameter, result retrieval can target different text types: the fiction-oriented core texts, specific collections, or the entire corpus. Then enter your query (Query:) and click Search. The query result is a list of all translation candidates of the given word, sorted by decreasing frequency by default. By clicking on a particular candidate, you can browse its occurrences in InterCorp and check the translation contexts. The reported frequency may differ since the corpus query may also find instances where the potential equivalent corresponds to a different word.

Treq is based on texts from InterCorp, release 15. The first step was to align the original and translated texts word-to-word using statistical methods provided by the GIZA++ program (Och–Ney 2003). The aligned word pairs were then sorted and summarized. The results of the automatic excerption were not reviewed; however, the relative frequency of a given pair may serve as an approximate indicator of reliability. The more often an equivalent of a given word occurs in comparison with other equivalents, the more useful it may be.