Artificial Intelligence will change every industry and every line of business. However, human experts are still needed to guarantee commercial accuracy and quality.
Palo Alto, California (PRWEB) March 16, 2017
Lilt, a Palo Alto-based startup, outperformed titans of the technology industry in a rigorous test of language translation tools. The company’s interactive translation system–the first such commercial system–outperformed the most heavily used systems from Google, Microsoft, and Systran, and SDL by a large margin. This outcome shows that systems incorporating human feedback beat systems that don’t, and are on par with the latest, expensive artificial intelligence from Google.
Lilt’s interactive system that adapts to human input is one of the most exciting developments in language translation. The company tested this system against Google, Microsoft, Systran and SDL by comparing industry-standard BLEU scores on an English to German and English to French translation of Swiss government documents. Lilt matched Google’s latest neural machine translation system, and outperformed all of the other systems by a healthy margin.
“Artificial Intelligence will change every industry and every line of business. However, human experts are still needed to guarantee commercial accuracy and quality. Our interactive system makes those experts more productive, and learns from their feedback, which is an overlooked source of valuable data,” said Ash Fontana, Managing Director of Zetta Venture Partners - the first venture capital fund in the world dedicated to artificial intelligence.
Neural-network based systems also hold a lot of promise in improving language translation. Google released a neural network-based system in November 2016 that improved their system more in a single leap than they’d seen in the last ten years combined. However, their system is expensive, requiring a week of training on 96 advanced, GPU-based systems, costing about $11,000. Lilt’s system costs just $29 to train and produces results on par with Google’s neural network system with a BLEU score of 33.1 vs 33.2 for English to French and 28.2 vs 28.6 for English to German. Lilt’s system beat Microsoft’s latest neural system by a large margin with a BLEU score of 33.1 vs 30.7 for English to French and 28.2 vs 23.8 for English to German.
Last month, Lilt announced the launch of Lilt Labs, a forum for computational linguists, scientists, and language professionals to publish evaluations, insights and research. The first article on Machine Translation Quality Evaluation offered quantitative assessments of major translation providers that combined the standard evaluation protocol with a typical translation workflow in which a human translator progressively translates a document.
“We want to drastically increase levels of translation productivity by uniting artificial intelligence and translators together, in order to bring our customers into new markets,” said John DeNero, Co-Founder and Chief Scientist at Lilt, and Assistant Professor of Computer Science at UC Berkeley. “This bench marking is an important step in showing the world that interactive systems are the way forward. We welcome the entire community to join us in publishing results and insights that will accelerate adoption of this powerful technology.”
Anyone wishing to join the discussion or register for more information can do so at http://labs.lilt.com.
Press Contact: press(at)lilt.com
Appendix: BLEU Score Comparison from latest results.