Rylan Conway

Rylan T. Conway, Evan W. Sangaline

Gesture typing is a method of text entry that is ergonomically well-suited to the form factor of touchscreen devices and allows for much faster input than tapping each letter individually. The QWERTY keyboard was, however, not designed with gesture input in mind and its particular layout results in a high frequency of gesture recognition errors. In this paper, we describe a new approach to quantifying the frequency of gesture input recognition errors through the use of modeling and simulating realistically imperfect user input. We introduce new methodologies for modeling randomized gesture inputs, efficiently reconstructing words from gestures on arbitrary keyboard layouts, and using these in conjunction with a frequency weighted lexicon to perform Monte Carlo evaluations of keyboard error rates or any other arbitrary metric. An open source framework, Dodona, is also provided that allows for these techniques to be easily employed and customized in the evaluation of a wide spectrum of possible keyboards and input methods. Finally, we perform an optimization procedure over permutations of the QWERTY keyboard to demonstrate the effectiveness of this approach and describe ways that future analyses can build upon these results.

Rylan Conway, Lambert Mathias

In a spoken dialogue system, dialogue state tracker (DST) components track the state of the conversation by updating a distribution of values associated with each of the slots being tracked for the current user turn, using the interactions until then. Much of the previous work has relied on modeling the natural order of the conversation, using distance based offsets as an approximation of time. In this work, we hypothesize that leveraging the wall-clock temporal difference between turns is crucial for finer-grained control of dialogue scenarios. We develop a novel approach that applies a {\it time mask}, based on the wall-clock time difference, to the associated slot embeddings and empirically demonstrate that our proposed approach outperforms existing approaches that leverage distance offsets, on both an internal benchmark dataset as well as DSTC2.