There are problems in life that Microsoft can't solve for us, and one is spelling out of context. Type "My aidl holiday wood bee in nue zeelend, were I can go hickin" into Word and the software will be completely befuddled. If you run the Microsoft Word spell checker and choose the first suggested alternative for each error, you'll get: "My aid holiday wood bee in nude zee lend, were I can go hick in." Great.
Ginger Software's unique algorithm would look not at each word but at the sentence as a whole and come up with: "My ideal holiday would be in New Zealand, where I can go hiking." The software can even identify correctly spelled words that are used in the wrong context, such as: "wood bee" instead of "would be." Such words are not identified as mistakes by MS-Word's spell checker.
Ginger Software founder and chief executive, Yael Karov, knows all about spell checkers' helplessness to help. From conversations she's had with professionals, dyslexia organizations in the U.S. and Britain, at conferences abroad and with dyslectics themselves, Karov has learned that many people stopped using spell checkers altogether. "They can't fix the sort of mistakes that dyslectics make," she explains.
The problem isn't rare: 15% of the world's population is dyslectic at some level, she says. About 5% have hard-core dyslexia: They can barely write, which is deeply frustrating. Most simply resort to having somebody else work with them when writing, for lack of alternatives.
Naturally, Karov won't discuss the mathematical algorithm underlying the software she developed. All she will say is that the program studies a vast pool of proper sentences in English and builds a model of proper language: the characteristics of proper sentences - semantically and grammatically - compared with wrong sentences. It wouldn't have corrected "artside" in the photograph to "art" because of the logic of the context. It isn't a question of memory, but of the logic of sentence structure, Karov says.
In other words, the Ginger software doesn't analyze the text at the level of the word, but of the whole sentence. Dyslectics can have trouble choosing the right word - hence the attention to the sentence as a whole.
Take "I lik two play artside" - lik could be any of dozens of words, from lick to look to like, Karov points out. Moreover, there are infinite numbers of sentences in which it could appear. "Our program's uniqueness lies in two elements. Based on its learning, it can create the reasonable alternatives of what the writer might have meant, in correct language, and then it chooses the single most appropriate correction for the whole sentence from that vast number of alternatives," she says.
For instance, the software knows the proper sentence "It was a pleasure to meet you," and can correct a sentence written as "It was a pleshur to meat yo." "But that's simple and life is a lot more complex, especially when the mistakes in a single sentence are right next to each other."
Nor is Ginger foxed by rare sentences never seen by the software. "Students should eksersiv daily to inprove there strenth" will be automatically corrected, in a single click, to "Students should exercise daily to improve their strength." Corrections are chosen based on the context of the error, even when the context itself contains errors; the software knows how to deal with multiple consecutive mistakes.
How does it work? The entire problem sentence is presented in an upper window on the user's monitor. The program highlights the words that it thinks are wrong in that upper window. In a lower window, the program proffers suggestions, highlighting the relevant words. The user can click on individual words to choose the best one or overwrite the automatic correction.
What actually is the product? A software component that the user downloads, which is integrated into Microsoft Word. Later Ginger Software plans to develop components for e-mail and browsers.
Karov established Ginger with her husband, Avner Zangvil, the chief software architect and product strategist. The company's business development is handled by Amit Gilon, son of Zohar Gilon of Tamar Ventures.
Karov herself, a serial entrepreneur, has been working with processing natural language for 20 years. "I studied mathematics and computers, and majored in processing natural language at the Weizmann Institute of Science," she says. She then founded AGENTics, which made electronic cataloging software and which was sold to Mercado Software. "After one year of working at Mercado, I spearheaded the technology's sale to Microsoft, which uses it for MSN Shopping. Later I worked as chief technology officer and R&D vice president at Rosetta Genomics." She stayed at Rosetta from its first stages until its flotation.
At first Rosetta worked at locating new kinds of noncoding microRNA genes that regulate DNA expression into proteins. It was a research mission of the most challenging nature, she says, yet the company managed to track down the microRNA molecules it sought. At the next stage, Rosetta set out to develop diagnostic and therapy products for cancer using microRNA genes.
Karov stayed at Rosetta for about two years during this diagnostics development stage, until the initial public offering.
That for her served as a natural point to move on, given that the company was moving from mathematical theory, her specialty, to biological and medical applications.
Zangvil also has a history as a startup entrepreneur. His was Menta Software, which was sold to GraphOn for $6.5 million worth of stock. Menta designed software that enabled Internet devices to access Windows applications.
Neither spouse made a fortune from their startups, but they're comfortable, Karov says. Meanwhile, Ginger Software will be launching its pilot version of software for dyslectics in about a month. At first it will supply the product for free, but from around year-end it intends to charge an annual licensing fee that will range from $99 in the U.S. to 79 pounds in Britain. (Marketing costs in each country are different, Karov points out.) It will be selling via the Internet as well as education institutions, schools and dyslexia organizations.
For the time being the software works only on English, though the algorithm is generic and could work in any language - but it has to be adjusted by language. Hebrew is a small market and Ginger Software is young, so naturally it would start by focusing on the gigantic English language market, Karov says.
The company plans to launch a browser-based version that would vet chats and e-mails, and is designing another version for teachers and schools to help them diagnose difficulties students may be having. The software will produce a report on each student's particular mistakes, track progress over time and analyze the type of mistakes to help the teacher design a study program for each pupil.
In principle, Ginger's developments are designed to help the community of dyslexics, but Karov doesn't see why it can't serve the larger population of English learners as well. That can wait for 2009, though. To start with, the company is focusing on the people that need its products the most.