Munich – 05.02.2015
From automatic text analysis to text comprehension, ExB sets the standard by winning another international competition
- ExB takes both first and second place in the Semantic Evaluation Exercises competition at SemEval-2015.
- ExB technology learns to understand texts the way people do.
Munich-based technology and development company ExB (www.exb.de) once again takes top positions in an international technology competition, leaving no doubt that it is setting the standard in the field of natural language processing.
ExB’s innovative text analysis technology has demonstrated its superior performance in the Semantic Evaluation Exercises competition at SemEval-2015, following up its extraordinary success in late 2014 at GermEval, an international software competition in named entity recognition (NER). ExB has amply demonstrated its outstanding position and competitiveness at the cutting edge of the IT industry, paving the way for the use of natural language processing technologies in everyday life.
The SemEval competition is organized by the Association for Computational Linguistics (ACL), an international scientific and professional society for people and institutions working in natural language processing. Teams from both academic institutions and technology companies participated in a total of 18 different challenges covering a variety of concrete tasks requiring the understanding of natural language.
Language understanding can mean comparing apples to oranges
People recognize similarities between statements even when they use completely different words. For example, it is easy to see the similarity between the sentences “Paul is eating an apple” and “Peter enjoyed his orange”, even though they do not share a single word. In the same way, the sentence “NASA is launching a satellite” has little in common with them, even though it is superficially just as much like the previous sentences as they are from each other. For computers, recognizing this kind of similarity and difference is very hard, even though it seems obvious to people. Solutions to this problem involve more than just the words and grammar of language, they require some genuine understanding of what language means.
ExB competed in all parts of the Semantic Textual Similarity challenge, including English and Spanish language understanding, against 28 teams with 74 proposed solutions for the English task, and just 7 teams with 17 entries for Spanish. This level of competition highlights ExB’s extraordinary performance, achieving the second best ranking in the English-language task, just behind the winning entry from the University of Colorado and ahead of Samsung, the next highest commercial participant.
ExB performance is even more striking in the Spanish competition, where it took the first three places by a large margin. This result reinforces the outstanding quality of ExB’s multilingual, robust, comprehensive language technology portfolio, which has grown to include 40 languages.
According to Dr. Ramin Assadollahi, founder and CEO of ExB: “Computer understanding of text has made extremely strong progress in recent years, not least through the innovative scientific achievements of pioneers such as ExB. This technology is no longer about comparing keywords, but about how close machines can get to a human rating in their assessment of textual similarities.”
Automatic understanding and interpretation of natural language has grown dramatically more important in the era of Big Data, since according to analysts roughly half of all digitized data is written natural language. Computer text analysis has the potential to revolutionize areas where machines have to understand people’s written language and act in response to its contents. For example, matching a doctor’s description of a patient’s symptoms to a potentially life-saving clinical trial, or matching a lawyer’s case dossier to relevant precedents and case law, saving countless hours of clerk’s labour.