Forum for Science, Industry and Business

Sponsored by:     3M 
Search our Site:

 

USC Researchers Build Machine Translation System -- and More -- For Hindi in Less Than a Month

02.07.2003


In less than a month, researchers at USC’s Information Sciences Institute and collaborators nationwide have built one of the world’s best systems to translate Hindi text into English and query Hindi databases using English questions.

This effort was part of the "Surprise Language" project, a test of the computer science community’s ability to create translation tools quickly for previously unresearched languages sponsored by the Defense Advance Research Project Agency (DARPA). The exercise ended July 1.

"A month ago, we didn’t even know what language we would be working on," explained Ulrich Germann, a computational linguist at ISI, which is part of the USC School of Engineering.



Then, at 10:55 p.m. PDT on June 1, the manager for DARPA’s TIDES (Translingual Information Detection, Extraction, and Summarization) program fired the starting gun with an email: "Surprise Language is Hindi.... Good luck!"

Teams at 11 different sites across the US and one in the UK jumped into action, and twenty-nine days later can present an impressive array of information processing tools for Hindi.

"We succeeded in all aspects of the exercise," said Douglas W. Oard, an associate professor at the University of Maryland who is currently spending a sabbatical year at ISI. "A month ago, we had no information retrieval for Hindi, no machine translation, no named entity identification, no question answering. Now we have all of that."

ISI’s researchers focused on four aspects of cross-lingual information processing: resource building, machine translation, summarization, and providing an efficient interface for the human to navigate the information space. Of these,"clearly, machine translation is the pivotal technology in this scenario," said Germann.

ISI research scientist Franz Josef Och, a leading specialist in machine translation, did much of this key task for ISI.

"Our approach uses statistical models to find the most likely translation for a given input," Och explained. "Instead of telling the computer how to translate, we let it figure it out by itself. First, we feed the system collection of parallel texts, material in the foreign language and their translations into English. The system tries to find the English sentence that is the most likely translation of the foreign input sentence, based on these statistical models."

Och’s Hindi system was one of four developed independently during the exercise. Trials scheduled for coming weeks will rate his against those developed at other sites.

Finding and creating parallel texts for Och and his colleagues to analyze was a major effort during the exercise, said Germann. While for most European languages, there are one or two predominant standardized ways of encoding them, e.g."Latin-1" or Unicode, Hindi has a wildly mixed potpourri of encodings.

"It’s ridiculous," said Germann, "almost every single Hindi language web site has its own encoding." Tools had to be made to convert all of these various systems to a single common one to present parallel texts to Och and other machine translation experts.

"Most of the conversion work was done by our partners at other participating sites, and it was absolutely critical to the success of the exercise," Germann said.

In addition to Och’s translation work, researchers applied search, summarization, and visualization tools developed at ISI to make Hindi texts more accessible to English language speakers. ISI researchers Anton Leuski and Chin-Yew Lin collaborated on a super-Google-like mutli-document search, summarization, adn translation system that allows users to enter search terms in English and generate results grouped by similarities found in the text, using refinements on a multi-document summarization technique developed by Lin.

Graduate student Liang Zhou developed a way to generate a headline for each group of similar stories found. Leuski’s unique Lighthouse visualization system displayed these results at spheres floating in groupings on the screen, with the most similar closest together.

The bottom line: a user can then view individual documents, or automatically generated summaries for whole groups of documents. Even though all documents were originally in Hindi, all the added value is available in English, thanks to the machine translation engine. In addition, references to locations in the documents are spotted (using a third-party tool, the BBN IdentiFinder) in the text and plotted on a map.

"It’s just wonderful to see so many of the technologies that we have developed at ISI come together and interact in such a useful way," said Eduard Hovy, head of ISI’s Natural Language Group.

Eric Mankin | USC
Further information:
http://www.usc.edu/isinews/stories/98.html

More articles from Information Technology:

nachricht Ultra-precise chip-scale sensor detects unprecedentedly small changes at the nanoscale
18.01.2017 | The Hebrew University of Jerusalem

nachricht Data analysis optimizes cyber-physical systems in telecommunications and building automation
18.01.2017 | Fraunhofer-Institut für Algorithmen und Wissenschaftliches Rechnen SCAI

All articles from Information Technology >>>

The most recent press releases about innovation >>>

Die letzten 5 Focus-News des innovations-reports im Überblick:

Im Focus: Traffic jam in empty space

New success for Konstanz physicists in studying the quantum vacuum

An important step towards a completely new experimental access to quantum physics has been made at University of Konstanz. The team of scientists headed by...

Im Focus: How gut bacteria can make us ill

HZI researchers decipher infection mechanisms of Yersinia and immune responses of the host

Yersiniae cause severe intestinal infections. Studies using Yersinia pseudotuberculosis as a model organism aim to elucidate the infection mechanisms of these...

Im Focus: Interfacial Superconductivity: Magnetic and superconducting order revealed simultaneously

Researchers from the University of Hamburg in Germany, in collaboration with colleagues from the University of Aarhus in Denmark, have synthesized a new superconducting material by growing a few layers of an antiferromagnetic transition-metal chalcogenide on a bismuth-based topological insulator, both being non-superconducting materials.

While superconductivity and magnetism are generally believed to be mutually exclusive, surprisingly, in this new material, superconducting correlations...

Im Focus: Studying fundamental particles in materials

Laser-driving of semimetals allows creating novel quasiparticle states within condensed matter systems and switching between different states on ultrafast time scales

Studying properties of fundamental particles in condensed matter systems is a promising approach to quantum field theory. Quasiparticles offer the opportunity...

Im Focus: Designing Architecture with Solar Building Envelopes

Among the general public, solar thermal energy is currently associated with dark blue, rectangular collectors on building roofs. Technologies are needed for aesthetically high quality architecture which offer the architect more room for manoeuvre when it comes to low- and plus-energy buildings. With the “ArKol” project, researchers at Fraunhofer ISE together with partners are currently developing two façade collectors for solar thermal energy generation, which permit a high degree of design flexibility: a strip collector for opaque façade sections and a solar thermal blind for transparent sections. The current state of the two developments will be presented at the BAU 2017 trade fair.

As part of the “ArKol – development of architecturally highly integrated façade collectors with heat pipes” project, Fraunhofer ISE together with its partners...

All Focus news of the innovation-report >>>

Anzeige

Anzeige

Event News

Sustainable Water use in Agriculture in Eastern Europe and Central Asia

19.01.2017 | Event News

12V, 48V, high-voltage – trends in E/E automotive architecture

10.01.2017 | Event News

2nd Conference on Non-Textual Information on 10 and 11 May 2017 in Hannover

09.01.2017 | Event News

 
Latest News

New Study Will Help Find the Best Locations for Thermal Power Stations in Iceland

19.01.2017 | Earth Sciences

Not of Divided Mind

19.01.2017 | Life Sciences

Molecule flash mob

19.01.2017 | Physics and Astronomy

VideoLinks
B2B-VideoLinks
More VideoLinks >>>