'Too bee by the see' (Extract)

By ZIV HELLMAN
Extract from an article in Issue 12, September 29, 2008 of The Jerusalem Report. To subscribe to The Jerusalem Report click here. Cn u cee wot is rong wiht this sntenc? Most of us have become accustomed to relying on computerized spell checkers to catch and correct spelling errors - and consider them a godsend compared to the alternative: leafing through a dictionary. But standard spell-checkers leave much to be desired. Consider, for example, the sentence: "My idle holiday wood bee on an iland were the sea is fild with moultey coulerd fish." Inputting that sentence into Microsoft Word's spell checker, laboriously verifying each word underlined in red, and selecting one of the suggested alternatives might result in: "My idle holiday wood bee on an island were the sea is field with molten coiled fish." Not very impressive. But type the sentence into the spell checker developed by Ginger Software, a young Tel Aviv-based start-up, and with one click of a button it returns a flawless version true to the writer's intention: "My ideal holiday would be on an island where the sea is filled with multi-colored fish." For millions who wage an aggravated daily struggle with the vagaries of spelling in the English language, Ginger offers new hope. "Many people suffering from dyslexia, especially those categorized as 'hard core' dyslectics, are deeply frustrated with the difficulties they have with writing and spelling, and have given up on spell checkers because they often fail to help them," says Yael Karov, CEO of Ginger Software. "They often resort to having somebody else correct their errors or write for them, for lack of alternatives. And there are many people around the world learning English who find the non-phonetic spelling of many English words bewildering and cannot rely on standard spell checkers. Ginger's automated spell correction enables anyone to generate high-quality written texts for business and personal use." Ginger's patent-pending product, currently still in limited beta testing phase, is expected to be made widely available by late October (at http://www.gingersoftware.com), as an add-on to Microsoft Word, with a generic version that will be able to spell check documents written on any Web program to be developed in 2009. Although Ginger's spell checker will initially be offered as a free Web service, the company is planning at some point to require users to pay for a subscription. Karov has been working on topics related to the statistical study of language processing for 20 years, and is the author or co-inventor of over 20 patents in the field. After completing an MSc in mathematics and computer science at the Weizmann Institute of Science in Rehovot, focusing on natural language processing, Karov served as a research and development team director at Click Software, which produces software for optimizing organizational efficiency in service companies. She went on to found Agentics, a company that classifies electronic commerce product information based on natural language processing, which was eventually acquired by Mercado Software and Microsoft's MSN division. Prior to founding Ginger Software, Karov was the chief technology officer and vice president of R&D at Rosetta Genomics, a molecular diagnostics company that studied micro-RNA molecules and DNA codes to develop tools for medical diagnoses. "DNA is a 'language,' coded in molecules," explains Karov. "It can be studied using the same statistical models for machine learning that are used for human languages." With Ginger, Karov is returning to working directly with natural language processing, essentially turning her not-yet-completed doctoral dissertation into a business. The basic idea is to use context and the grammatical structure of entire sentences to improve on the performance of standard spell checkers, which work solely at the level of individual words. The Ginger model identifies correct spelling based on surrounding words and most probable intended meaning. Errors are corrected according to the context of the sentence, and only corrections with high semantic and grammatical fit are selected. "When we think and write, we do so in sentences, not in terms of isolated words," says Karov. "A spell checker that focuses only on words is bound to be wrong often." Ginger's context-based approach enables it to achieve 95 percent accuracy rates, according to its internal tests. The company is so confident of its spell-correcting abilities that it often offers users only one suggested re-spelling of a sentence. It can even catch errors that entirely evade standard spell checkers. For example, a sentence such as 'I'd lick too bee by the see' passes through Microsoft Word's spell checker with flying colors because each individual word in isolation is spelled correctly. Ginger's spell checker, considering the entire sentence, has no difficulty suggesting that what the user meant to write is "I'd like to be by the sea." Ginger Software was founded in November 2007 by Karov and her husband, Avner Zangvill, who serves as Ginger's chief architect and is also a serial software entrepreneur with several successful start-ups behind him. Knowing they had the capability of producing a greatly improved spell checker, they researched potential markets for the product, and looked into the possibility of assisting dyslectics. "Studies show that at least 15 percent of the population are diagnosed as suffering from dyslexia, and there are estimates that up to another five percent have some form of dyslexia but have not been diagnosed," according to Dr Gilda Palti, a consultant for Ginger who holds a PhD in the treatment of specific learning disabilities from the University of Bristol and manages a private clinic for dyslexia therapy in England. "That is a surprisingly large number, and the causes can be multi-fold, stemming from problems with hearing, with relating sounds to writing, or aural and visual memory." Karov, who has spent much of the past year working intensely with therapists, dyslexia organizations and dyslectics themselves, says she was "surprised to discover how little understanding there is of how dyslectics process spelling, despite the fact that on average they make three spelling errors per sentence. There isn't even a commonly accepted theory as to what distinguishes dyslectic spelling errors from those made by non-dyslectics." Ginger Software has gathered immense amounts of data on dyslectic-type spelling errors - "the largest such collection in the world, which we will make available for academic research" according to Karov. The company also discovered how keenly dyslectics desired a tool that would help them spell correctly. Extract from an article in Issue 12, September 29, 2008 of The Jerusalem Report. To subscribe to The Jerusalem Report click here.