GazeTheKey: Interactive Keys to Integrate Word Predictions for Gaze-based Text Entry Korok Sengupta Raphael Menges Abstract Institute for Web Science Institute for Web Science In the conventional keyboard interfaces for eye typing, the and Technologies (WeST) and Technologies (WeST) functionalities of the virtual keys are static, i.e., user’s gaze University of Koblenz-Landau University of Koblenz-Landau at a particular key simply translates the associated letter Koblenz, Germany Koblenz, Germany [email protected] [email protected] as user’s input. In this work we argue the keys to be more dynamic and embed intelligent predictions to support gaze- based text entry. In this regard, we demonstrate a novel "GazeTheKey" interface where a key not only signifies the Chandan Kumar Steffen Staab input character, but also predict the relevant words that Institute for Web Science Institute for Web Science and Technologies (WeST) and Technologies (WeST) could be selected by user’s gaze utilizing a two-step dwell University of Koblenz-Landau University of Koblenz-Landau time. Koblenz, Germany Koblenz, Germany [email protected] Web and Internet Science Author Keywords Research Group, University of Gaze input; eye tracking; text entry; eye typing; dwell time; Southampton, UK visual feedback. [email protected] ACM Classification Keywords H.5.2 [Information interfaces and presentation]: User Inter- faces Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed Introduction for profit or commercial advantage and that copies bear this notice and the full citation Gaze-based text entry is a valuable mechanism for the on the first page. Copyrights for third-party components of this work must be honored. For all other uses, contact the Owner/Author. people with severe motor disabilities, who have no or lit- Copyright is held by the owner/author(s). tle control of muscles to operate the physical keyboard. IUI’17 Companion, March 13-16, 2017, Limassol, Cyprus. ACM 978-1-4503-4893-5/17/03. Usually, eye typing is accomplished by onscreen keyboard, DOI: http://dx.doi.org/10.1145/3030024.3038259 where the user selects the letters by looking at them for certain duration called dwell time [4]. Researchers have proposed alternate designs like Dasher [8], or dwell-free Fil- representation basically relies on the next word that the re- teryedping [7] to achieve high performance, which needs spective language model could predict based on the previ- extensive training and expertise for faster and accurate ously entered text by user [9]. Figure1 shows the represen- input. Therefore, to assist novice users, most of the com- tation sketch of the keys containing the letter and the as- mercial and popular gaze-based systems1 follow the con- sociated word suggestion at the bottom. The green framed vectional design of QUERTY key layout, associated with letters on the key are the ones that have been already en- text area and word predictions. In the context, word pre- tered by user. This is succeeded by the letter on the key, dictions becomes a significant aspect, since it can reduce framed in yellow. The red frame signifies the suggestive the number of keystrokes required to write the word and letters predicted based on previous letters. decrease user’s effort while typing. However, for eye typ- ing there is an additional cost of perceptual and cognitive While Figure1 presents the structure of word suggestion load, caused by shifting the focus from the keyboard layout on keys, Figure 2 displays the user interaction methodology to the suggested word list and the repeated scanning of the via different states of a key on eye gaze fixation. a) At first, list. Moreover, users are overloaded by the task of typing the key is in its standard mode to trigger the input of the and might overlook the suggestions from an external area single letter displayed on it. b) This is implemented via a as they are engaged in the visual search task of finding the dwell time that is visualized through an orange circle. When next key to enter the intended letter. Hence, there is a need the user’s gaze is upon the key, the circle grows from the to bring the predicted words within the visual search area of center of the key until the key’s area is completely filled. c) users. Furthermore, in the conventional design the word list The input of the displayed letter is triggered and a visual is typically limited to three or four words, due to the space feedback in appearance of a black pulse as feedback to limitation of eye controlled interfaces [2], e.g., large buttons the user. d) If a suggestion is available by the prediction to tackle eye tracking accuracy. engine, the key now turns into suggestion mode e) Further gaze fixation by the user lets the area of the suggestion fill In this work, we move beyond the conventional design of the key starting from bottom and ending at the top. f) When text suggestions as few predicted words on the top of key- the key is filled after this second dwell time, the currently board layout, and propose a dynamic approach of embed- collected word is replaced by the given suggestion. All key ding the suggestions on the key, which would not only bring actions can be aborted by the user by looking at a different Figure 1: Suggestion per key is the relevant suggestions in the visual attention of users, it position on the screen. estimated with previous letters would also offer the possibility to include much more sug- entered by user, and letter gestions. Figure 3 shows the complete design of GTK keyboard in- associated with the key as input. In terface including above-mentioned functionality (detailed the image above the already GazeTheKey demonstration of GTK usage is available here2). The de- collected part of the word is framed The two major component of GazeTheKey (GTK) design sign includes the principles of eye-controlled interfaces [2] green, the letter on the key as (e.g., enlarged buttons and visual feedback to cope with eye yellow, and the suggestive part by is: how does the keys represent the word suggestions, and prediction engine red. how does the user interact with these suggestions. The 1https://github.com/OptiKey/OptiKey/wiki 2https://www.youtube.com/watch?v=-UDDTJHBPVA tracking accuracy), moreover, it follows the usability heuris- tics of Neilsen [6], keeping the design as close to conven- tional keyboard layout as possible with minor adjustment of key formation and addition of necessary keys for improving efficiency while typing. (a) Standard mode. In the shown example in Figure 3, user typed some letter of a word and the current state of all keys shows relevant suggestions that can be selected via additional dwell time over the key (as explained in Figure 2). One limitation of such a two-step dwell time input is that the user wouldn’t First dwell time. (b) be able to dwell on a letter consecutively to type the letter multiple times. Hence an extra repeat key has been added Figure 3: The new keyboard with characters predicting the next at the lower bottom of the environment. This key serves the possible word if character were chosen. The repeat key on the left purpose to trigger the input of the last selected letter. We bottom row repeats the previously typed in letter. have done some further optimization to minimize the visual search for users, i.e., space and backspace key is used to (c) Letter input. present further information about the currently edited word. ployed. The dictionary of the prediction engine is composed As shown in the example, the space button works as the of a random subset of 50.000 English sentences from the 5 confirmation of the typed word and therefore it is displayed Tatoeba database. The proposed design presents several on the bottom of key. A preview of the edited word (after word predictions as one suggestion per key is to be com- deletion by backspace) is displayed on the backspace key puted. Therefore, an asynchronous calculation has been (d) Suggestion mode. (essentially showing the usage of respected key). These set up to guarantee an interruption-free interaction, other- simple heuristics help the user to stay with the gaze on the wise the rendering may freeze after letter input for some same position without the need to check the intended action seconds and user’s gaze would have no effect on the inter- on the edited word. face. Implementation Evaluation (e) Second dwell time. The proposed keyboard has been developed as experi- We have conducted an initial small-scale study with SMI mental eyeGUI [5] element in C++. eyeGUI is a graphical REDn remote eye tracker to investigate how the proposed user interface framework for eye tracking driven applica- design is able to support the task of eye typing. Ten partici- tions. Rendering is handled by OpenGL function calls and pants (5 male, 5 female) contributed in our study, they were the FreeType23 library is used for font rasterization. For the aged between 21 and 30 years (mean 24.8, SD 2.347) and relevant word predictions, Presage4 library has been em- had no prior experience with eye controlled interfaces. We (f) Suggestion input. asked the participants to type 25 different sentences (in five 3http://www.freetype.org Figure 2: Key interactions. 4http://presage.sourceforge.net 5https://tatoeba.org/eng sessions), which were randomly chosen from the phrase 2. C. Kumar, R.
Details
-
File Typepdf
-
Upload Time-
-
Content LanguagesEnglish
-
Upload UserAnonymous/Not logged-in
-
File Pages4 Page
-
File Size-