COLING 2012 24 th International Conference on Computational

At the heart of SwiftKey's success are well motivated Machine Learning and Natural Language Processing principles. But that foundation is only the start, it's also required relentless focus on User Experience, solving endless real world issues and building and connecting with the vibrant community of SwiftKey users worldwide. This talk will take you through the story of how we turned a great IME into the most successful paid Android application in the world. Speaker’s Bio: Paul is Chief Software Architect of NLP company SwiftKey, creators of the market-leading input method by the same name. Quillpad multilingual predictive transliteration system Speaker: Ram Prakash H, Tachyon Technologies Abstract: Transliteration has been one of the common methods for multilingual text input. Many earlier methods employed transliteration schemes for defining one to one mapping of input alphabet combinations to output alphabet combinations. Though such well-defined mappings made it easier to write a transliteration program, the end user was burdened with learning the mappings. Further, though transliteration schemes try to map the alphabet combinations phonetically, it is unavoidable to introduce non intuitive combinations into the scheme. An alternative is to use predictive transliteration, where user could input a word, by intuitively combining the input alphabet phonetically and the predictive transliteration system should correctly convert it to the target language. In this talk, I will present the challenges that must be addressed by such a system, and describe how Quillpad can be trained for performing predictive transliteration between any two alphabets. Transliteration has been one of the common methods for multilingual text input. Many earlier methods employed transliteration schemes for defining one to one mapping of input alphabet combinations to output alphabet combinations. Though such well-defined mappings made it easier to write a transliteration program, the end user was burdened with learning the mappings. Further, though transliteration schemes try to map the alphabet combinations phonetically, it is unavoidable to introduce non intuitive combinations into the scheme. An alternative is to use predictive transliteration, where user could input a word, by intuitively combining the input alphabet phonetically and the predictive transliteration system should correctly convert it to the target language. In this talk, I will present the challenges that must be addressed by such a system, and describe how Quillpad can be trained for performing predictive transliteration between any two alphabets. Speaker’s Bio: Ram Prakash is the founder of Tachyon Technologies, and developer of Quillpad, the first online Indian language phonetic transliteration based input system. He was listed as one of the twenty MIT TR-35 2010 Young Innovators from India for Quillpad.