Smart Selection

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 1524–1533, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Smart Selection Patrick Pantel Microsoft Research One Microsoft Way Redmond, WA 98052, USA ppantel@microsoft.com Michael Gamon Microsoft Research One Microsoft Way Redmond, WA 98052, USA mgamon@microsoft.com Ariel Fuxman Microsoft Research 1065 La Avenida St. Mountain View, CA 94043, USA arielf@microsoft.com Abstract Natural touch interfaces, common now in devices such as tablets and smartphones, make it cumbersome for users to select text. There is a need for a new text selec- tion paradigm that goes beyond the high acuity selection-by-mouse that we have re- lied on for decades. In this paper, we in- troduce such a paradigm, called Smart Se- lection, which aims to recover a user’s in- tended text selection from her touch input. We model the problem using an ensemble learning approach, which leverages mul- tiple linguistic analysis techniques com- bined with information from a knowledge base and a Web graph. We collect a dataset of true intended user selections and simu- lated user touches via a large-scale crowd- sourcing task, which we release to the academic community. We show that our model effectively addresses the smart se- lection task and significantly outperforms various baselines and standalone linguistic analysis techniques. 1 Introduction The process of using a pointing device to select a span of text has a long history dating back to the invention of the mouse. It serves to access functions on text spans, such as copying/pasting, looking up a word in a dictionary, searching the Web, or accessing other accelerators. As con- sumers move from traditional PCs to mobile de- vices (e.g., tablets and smartphones), touch inter- action is replacing the pointing devices of yore. Although more intuitive and arguably a more natu- ral form of interaction, touch offers much less acu- ity (colloquially referred to as the fat finger prob- lem). To select multi-word spans today, mobile devices require an intricate series of gestures that results in cumbersome user experiences1. Conse- quently, there is an opportunity to reinvent the way users select text in such devices. Our task is, given a single user touch, to pre- dict the span that the user likely intended to se- lect. We call this task smart selection. We re- strict our prediction task to cases where a user in- tends to perform research on a text span (dictio- nary/thesaurus lookup, translation, searching). We specifically consider operations on text spans that do not form a single unit (i.e., an entity, a concept, a topic, etc.) to be out of scope. For example, full sentences, paragraph and page fragments are out of scope. Smart selection, as far as we know, is a new re- search problem. Yet there are many threads of re- search in the NLP community that identify multi- word sequences, which have coherent properties. For example, named-entity recognizers identify entities such as people/places/organizations, chun- kers and parsers identify syntactic constituents such as noun phrases, …

Content truncated to 500 words