Keystroke dynamics as signal for shallow syntactic parsing

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 609–619, Osaka, Japan, December 11-17 2016. Keystroke dynamics as signal for shallow syntactic parsing Barbara Plank University of Groningen The Netherlands b.plank@rug.nl Abstract Keystroke dynamics have been extensively used in psycholinguistic and writing research to gain insights into cognitive processing. But do keystroke logs contain actual signal that can be used to learn better natural language processing models? We postulate that keystroke dynamics contain information about syntactic structure that can in- form shallow syntactic parsing. To test this hypothesis, we explore labels derived from keystroke logs as auxiliary task in a multi-task bidirectional Long Short-Term Memory (bi-LSTM). Our results show promising results on two shallow syntactic parsing tasks, chunking and CCG su- pertagging. Our model is simple, has the advantage that data can come from distinct sources, and produces models that are significantly better than models trained on the text annotations alone. 1 Introduction As people produce text, they unconsciously produce loads of cognitive side benefit such as keystroke logs, brain activations or gaze patterns. However, natural language processing (NLP) hitherto almost exclusively relied on the written text itself. We argue that cognitive processing data contains potentially useful information beyond the linguistic signal and propose a novel source of information for shallow syntactic parsing, keystroke logs. Keystroke dynamics concerns a user’s typing pattern. When a person types, the latencies between successive keystrokes and their duration reflect the unique typing behavior of a person. Keystroke logs, the recordings of a user’s typing dynamics, are studied mostly in cognitive writing and translation process research to gain insights into the cognitive load involved in the writing process. However, until now this source has not yet been explored to inform NLP models. Very recent work has shown that cognitive processing data carries valuable signal for NLP. For in- stance, eye tracking data can inform sentence compression (Klerke et al., 2016) and gaze is predictive for part-of-speech (Barrett and Søgaard, 2015; Barrett et al., 2016). Keystroke logs have the distinct advantage over other cognitive modalities like eye tracking or brain scanning, that they are readily available and can be harvested easily, because they do not rely on any special equipment beyond a keyboard. Moreover, they are non-intrusive, inexpensive, and have the potential to offer continuous adaptation to specific users. Imagine integrating keystroke logging into (online) text processing tools. We hypothesize that keystroke logs carry syntactic signal. Writing time between words can be seen as proxy of the planning process involved in writing, and thus represent structural information between words. To test our hypothesis, we evaluate a multi-task bidirectional Long-Short Term Memory (bi- LSTM) model that is—to the best of our knowledge—the first to exploit keystroke logs to improve NLP models. We test our model on two shallow syntactic tasks, chunking and CCG supertagging. The choice of tasks is motivated by the fact that writing research analyzes so-called bursts of writing (i.e., consecutive spans of text, cf. Section 2.2), which are related …

Content truncated to 500 words