FASIL Email Summarisation System

FASIL Email Summarisation System Angelo Dalli, Yunqing Xia, Yorick Wilks NLP Research Group Department of Computer Science University of Sheffield {a.dalli, y.xia, y.wilks}@dcs.shef.ac.uk Abstract Email summarisation presents a unique set of requirements that are different from general text summarisation. This work describes the implementation of an email summarisation system for use in a voice-based Virtual Per- sonal Assistant developed for the EU FASiL Project. Evaluation results from the first inte- grated version of the project are presented. 1 Introduction Email is one of the most ubiquitous applications used on a daily basis by millions of people world-wide, tradi- tionally accessed over a fixed terminal or laptop com- puter. In the past years there has been an increasing demand for email access over mobile phones. Our work has focused on creating an email summarisation service that provides quality summaries adaptively and quickly enough to cater for the tight constrains imposed by a real time text-to-speech system. This work has been done as part of the European Union FASiL project, which aims to aims to construct a conversationally intelligent Virtual Personal Assistant (VPA) designed to manage the user’s personal and business information through a voice-based interface accessible over mobile phones. As the quality of life and productivity is to improved in an increasingly information dominated society, peo- ple need access to information anywhere, anytime. The Adaptive Information Management (AIM) service in the FASiL VPA seeks to automatically prioritise and pre- sent information that is most pertinent to the mobile users and adapt to different user preferences. The AIM service is comprised of three main parts: an email sum- mariser, email categoriser, calendar scheduling/PIM interaction and an adaptive prioritisation service that optimizes the sequence in which information is pre- sented, keeping the overall duration of the voice-based dialogue to a minimum. 2 Email Characteristics Email Summarisation techniques share many character- istics with general text summarisation techniques while catering for the unique characteristics of email: 1. short messages usually between 2 to 800 words in length (after thread-filtering) 2. frequently do not obey grammatical or con- ventional stylistic conventions 3. are a cross between informal mobile text or chat styles and traditional writing formats 4. display unique thread characteristics with 87% containing three previous emails or less (Fisher and Moody, 2001) All these four main characteristics combined to- gether mean that most document summarisation tech- niques simply do not work well for email. The voice- based system also required that summaries be produced on demand, with only a short pause allowed for the summariser to output a result – typically a maximum of around 1 second per email. Another main constraint imposed in the FASiL VPA was the presence of two integer parameters – the pre- ferred and maximum length of the summary. The maximum length constraint had to be obeyed strictly, while striving to fit in the summary into the preferred length. These performance and size constraints, coupled with the four characteristics of email largely determined the design of the FASiL Email Summariser. …

Content truncated to 500 words