| |August 20189CIOReviewdeployment of Machine Learning (ML) and Deep Learning models. Before we dive deeper, let us exam-ine the macro factors that are lead-ing to an unprecedented interest in text computation:· Traditional computing focus and the computing architecture is designed to process binary infor-mation and translating quantitative information in the decimal system to binary, is but natural. However, humans as a species do not commu-nicate in numbers; well mostly not! Consequently, we have generated exponentially more unstructured text data than we have generated structured quantitative data. Even the most sophisticated quants did not know, until very recently, how to process this information.· The rise of computing power and availability of data has fueled the Intelligence Revolution, whether we call it BI or Big Data or AI or any other new jargon that I am sure we will conjure very shortly. These ad-vances have also contributed to the field of computational linguistics. We can now ingest text and do all sorts of computational `magic' with them. Imagine if you can unleash the data science power by encoding the meaning of a billion words in a 300-dimensional vector space. The good news has only just started to unfold. Such vectors, derived from Neural Networks that have been painstakingly trained on massive datasets such as Google News and Wikipedia are available for our use.The Business of TextWith such exciting developments, the techie in us can very easily get carried away. However, first things first ­ let us look at some of the un-derlying problems that we are trying to solve. These problem statements will lead us to applications in the le-gal tech space and beyond. · Information Explosion ­ does not need an introduction and more information to add to the explo-sion! The business needs to cull out relevant nuggets of informa-tion from a pile of information, of-ten within short time frames. This renders a manual approach infruc-tuous. This has led to the rise of e-discovery tools and solutions in the legal fraternity. · Meaning-based computation. For decades, we have relied exclu-sively upon human intelligence to derive the meaning of words and all further processing has been depend-ent on our grey matter rather than silicon chips. Since such solutions were not around, we are yet to ar-ticulate the problem statements in this domain. Pause for a minute and think about it. You might get more answers specific to your domain than I can explain in the confines of this space. As an example, the sourc-ing of candidate profiles can be-come intelligent since the machines can now `understand' what is being mentioned in the CVs.· Look who's talking. Let's change focus from the input layer to the output layer. Machines can not only ingest and process information, but they can also generate natural language outputs. In case of the le-gal fraternity, this has led to contract creation tools. Chatbots that use some of these principles are becom-ing ubiquitous. There are innumer-able possibilities since these are rela-tively `newer' capabilities.What's the Good Word?So, how do we convert the heap of unprocessed and unstructured text into a gold mine? How can unused words become the good word, in a business and economic sense?In order to get bang for the buck, one needs to approach this field dif-ferently. Borrowing an analogy from Daniel Kahneman's `Slow Think-ing', one needs to by-pass the hard-wired neural circuitry (`Fast Think-ing') and discover more possibilities with these new-found capabilities of processing unstructured text. If we look at in-house legal departments, technology budgets for solving legal problems have traditionally been minimalistic and this segment did not evolve much for several decades Rajiv Maheshwari
< Page 8 | Page 10 >