A comparison of word- frequency based graphical user interface spellers
Keywords
Loading...
Authors
Issue Date
2020-09-28
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Brain-computer interface (BCI) spelling applications can be improved by optimized user
interface design and by incorporating natural language processing (NLP) models.
In this thesis a shot at language model based user interfaces is proposed, aimed at individuals
who are limited to or prefer a few commands. Such interfaces may for example be useful to
people that dislike P300 matrix spellers spellers, and have suffered complete loss of control
over voluntary muscles except eye muscles due to their amyotrophic lateral sclerosis (ALS),
cerebral palsy (CP) or locked-in syndrome (LIS) condition.
The proposed interfaces, also usable outside the field of BCI, are designed with regard to a
word dictionary, as to reduce the number of keystrokes during the typing process. The
interfaces are respectively a Huffman speller, word selection after entering a “bag of letters”,
an adaptation of the Bremen speller, and a six-directional speller. However the Hex-a-grid
yielded invalid results.
For each interface, the dynamic letter ordering is compared to an unchanging (static) version of
the interface. The speed and the number and nature of keystrokes, as well as general
preferences, are central in the comparison both between and within interfaces.
In the absence of learning, there were significantly higher entry rates for our adaptation of the
Bremen (dynamic) interface, both empirically and hypothetically in a BCI-setting. In addition,
there were significantly higher hypothetical entry rates for the Huffman dynamic compared to
Letter-bag dynamic mechanisms.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen