The goal of this workshop is to provide a forum for researchers working on problems related to the creation and use of parallel text. Recent events have demonstrated once again the importance of inter-language communication across a broad range of languages. This reinforces the need for advances in machine translation (MT) and multi-lingual processing tools, especially for languages with scarce resources.

This is a two-day workshop featuring two tracks:

1. Building and Using Parallel Texts for Languages with Scarce Resources (day 1)
2. Exploiting Parallel Texts for Statistical Machine Translation (day 2)

TRACK DESCRIPTIONS

1. BUILDING AND USING PARALLEL TEXTS FOR LANGUAGES WITH SCARCE RESOURCES

We invite submissions of papers addressing any of the following issues:

• Construction of parallel corpora, including the automatic identification and harvesting of parallel corpora from the Web
• Tools for processing parallel corpora, including automatic sentence alignment, word alignment, phrase alignment, detection of omissions and gaps in translations, and others
• Methods to evaluate the quality of parallel corpora and word alignments
• Using parallel corpora for the derivation of language processing tools in new languages
• Using parallel corpora for automatic corpus annotation (e.g. word sense disambiguation)
• Using parallel corpora for cross-language information retrieval and extraction
• The quality of language resources and systems that can be constructed with small amounts of parallel text and how do these scale up with the amount of text available.
• The role of external knowledge sources (e.g. bilingual dictionaries) in building resources and systems relying on parallel texts.
• Machine learning techniques for building and exploiting parallel texts (e.g. using small amounts of human-aligned parallel text to bootstrap large aligned corpora; active selection of data based on usefulness for different tasks)

Shared task

Teams that participate in the alignment exercise will be provided the training data for each language pair and development data taken from the gold standard data in order to build their systems. Thereafter they will be provided the unaligned gold standard data and asked to submit their proposed alignments in a short time frame. There will be two tracks for each language pair, one for teams that augment the training data with additional resources, and another for those that only use the training data. The resulting alignments will be evaluated relative to the previously mentioned gold standard data prior to the workshop. Short papers describing systems participating in this shared task and all evaluation methodologies employed will constitute a separate section in the workshop proceedings.

A more detailed description, training, development, and test data, and a number of other related resources will be made available from http://www.cs.unt.edu/~rada/wpt05.

2. EXPLOITING PARALLEL TEXTS FOR STATISTICAL MACHINE TRANSLATION

Translating documents from foreign languages into English (or between any two languages) by computer is one of the oldest goals in computational linguistics. Now, armed with vast amounts of digitally available translated text and powerful computers, we are witnessing significant progress toward achieving that goal. Statistical methods allow the analysis of parallel text corpora and the automatic construction of machine translation systems. Already, for some language pairs such as Chinese-English or Arabic-English, statistical machine translation (SMT) systems built at research labs outperform commercial systems.

Recent experimentation has shown that the performance of SMT systems varies greatly with the source language. In this workshop we would like to encourage researchers to investigate ways to improve the performance of SMT systems for diverse languages, including morphologically complex languages (e.g., Finnish) and languages with partial free word order (e.g., German). These issues lie on the border of linguistic analysis and statistical modeling, and the ACL conference is the most appropriate forum to investigate them, as ACL has a long tradition of hosting high-quality research in both areas.

Topics of interest include, but are not limited to:

• word-based, chunk-based, phrase-based, syntax-based SMT
• using comparable corpora for SMT
• using morphological and POS information for SMT
• integration of rule-based MT and statistical MT
• decoding
• error analysis

A more detailed description of the shared task, the test and training corpora, a freely available MT system, and a number of other resources are available from http://www.statmt.org/wpt05/mt-shared-task/

SUBMISSION INFORMATION

IMPORTANT DATES

ORGANIZERS

CONTACT

PROGRAM COMMITTEE