{"id":272250,"date":"2016-08-08T09:01:03","date_gmt":"2016-08-08T16:01:03","guid":{"rendered":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/?p=272250"},"modified":"2016-08-17T15:09:11","modified_gmt":"2016-08-17T22:09:11","slug":"microsoft-nlp-researchers-converge-acl-2016-edging-ever-closer-human-like-conversational-experiences","status":"publish","type":"post","link":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/blog\/microsoft-nlp-researchers-converge-acl-2016-edging-ever-closer-human-like-conversational-experiences\/","title":{"rendered":"Microsoft NLP researchers converge at ACL 2016, edging ever closer to human-like conversational experiences"},"content":{"rendered":"

By Bill Dolan<\/a>, Principal Researcher, Microsoft Research<\/em><\/p>\n

\"ACL\"This year, the annual meeting of the Association for Computational Linguistics (ACL) will be held in Berlin, Germany, August 7-12, 2016, at Humboldt University. ACL is the premier conference on natural language processing (NLP) systems and computational linguistics.<\/p>\n

As a Gold sponsor, Microsoft is proud to have more than 20 researchers attending and presenting at ACL. Along with my colleagues in the Natural Language Processing and Speech group<\/a>, we\u2019ll be\u00a0 presenting our latest research aimed at allowing computers to manipulate human language in order to engage and assist users.<\/p>\n

A major focus of our recent work is language generation: giving a \u201cvoice\u201d to machine state, code and structured data, so that users can begin to use language to productively collaborate with their devices through a natural, flowing dialog in which the machine becomes an active conversational collaborator. One of our papers at ACL this year, for instance, demonstrates how we can generate plausible, commonsensical questions in response to a photograph; another explores how we can begin to \u201ctranslate\u201d between natural language and code.<\/p>\n

We\u2019ll also be talking about the latest work in our long-term effort aimed at training neural conversational models from huge volumes of naturally-occurring human conversations, learning how to generate natural-sounding dialog from scratch at each point in a conversation. As natural language dialog becomes an increasingly important direction in interface design, this data-driven approach \u2014 pioneered by Microsoft Research<\/a> \u2014 \u00a0is emerging as a key research area. One important challenge involves imbuing game\/virtual reality characters and personal agents with distinctive personalities, so that their dynamically generated responses sound as if they were produced by a specific, real intelligence. At ACL 2016, we\u2019ll be presenting the first published work on data-driven persona modeling<\/em>, \u201cA Persona-Based Neural Conversation Model<\/a>.\u201d This work demonstrates how state-of-the-art neural modeling techniques train conversational agents that \u201csound like\u201d a specific character. What\u2019s more, the work shows how these techniques permit the persona to adjust its language use to match the linguistic behavior of the person it is talking with; a subtle yet crucial phenomenon that is characteristic of natural human conversation.<\/p>\n

Our ultimate goal is to be able to tap the profile of an arbitrary person and generate conversations that accurately emulate that individual\u2019s persona in terms of linguistic response behavior and other salient characteristics. As the paper states, \u201cthis would dramatically change the ways in which we interact with dialog agents of all kinds, opening up rich new possibilities for user interfaces. Given a sufficiently large training corpus in which a sufficiently rich variety of speakers is represented, this objective does not seem too far-fetched.\u201d<\/p>\n

In addition, Microsoft Research is pleased to cosponsor the 1st<\/sup> Workshop on Representation Learning for NLP (opens in new tab)<\/span><\/a>, which will discuss recent advances in \u201cvector space models of meaning, compositionality and the application of deep neural networks and spectral methods to NLP\u201d as well as explore future research directions.<\/p>\n

If you are going to ACL 2016, please chat with our researchers and scientists about the projects and opportunities at Microsoft that involve solving interesting AI, ML and NLP problems for billions of users. To learn more about our research being presented at ACL 2016, see the sections below for Tutorials, Workshops, Accepted papers and Microsoft attendees.<\/p>\n

Conference details<\/h2>\n