Pickup Usability Dominates - Human-Computer Interaction and Innovation in Handheld, Mobile and Wearable Technologies - page 53

model. The position of the target word in this list
is averaged over all words to give the average
ranked-list position for that corpus and keypad. An
ARP value of 1.0 indicates that the correct word
was always in the first position in the ranked list
of suggestions, a value of 2.0 that, on average,
the correct word was second in the ranked list.
We predicted an ARP value of around 1.03 for
a large corpus of English language newsarticle
articles using a standard phone keypad layout.
ARP naturally biases the averaging process so
that words are taken into account proportionally
to their occurrence in the text corpus.
Disambiguation accuracy (e.g. (Gong &
Tarasewich, 2005)) reports the percentage of times
the first word suggested by the disambiguation pro-
cess is the word the user intended—a DA value of
100% implies the disambiguation process always
give the correct word first, while 50% indicates
that it only manages to give the correct word first
half of the time. Gong and Tarasewich reported
DA of 97% for written English corpus and 92%
for SMS messages (both on a phone pad). This
is a more intuitive and direct measure than ARP,
but does not take into account the performance of
words that do not come first in the list.
KSPC (MacKenzie, 2002a) reports the aver-
age number of keystrokes required to enter a
character, for example home followed by a space
on a standard T9 mobile phone requires 6 key-
strokes—4663* where * is the next suggestion
key, giving a KSPC for hello of 5/4=1.25. As
with ARP and DA the value is normally averaged
over a large corpus of appropriate text for the
target language. A KSPC value of 1.0 indicates
perfect disambiguation as the user never needs to
type any additional letters, while a higher figure
reflects the proportional need for the next key
in disambiguation (and a lower level, successful
word completion). Full-sized non-ambiguous
keyboards achieve KSPC=1.00, standard date-
stamp method for entering text on 3 keys achieves
KSPC=6.45, date-stamp like interaction on 5 keys
achieves KSPC=3.13 and multitap on a standard

9-key mobile phone achieves a KSPC of around
2.03 (MacKenzie, 2002a). Hasselgren et al.
(Hasselgren et al., 2003) reported KSPC of 1.01
and 1.08 for T9 using Swedish news and SMS
corpora respectively, improving to 0.88 and 1.01
respectively for their bigram model with word
completion. KSPC does take into account ranked
list position for all words and compares easily with
non-predictive text entry approaches; however, it
is a rather abstract measure being based on letters,
especially for dictionary-based approaches that
are inherently word-based.
To gain an insight into potential expert user
behaviour with different keyboards, different
approaches have been taken to modelling interac-
tion in order to predict expert (trained, error-free)
performance. There are two basic approaches:
physical movement modelling and keystroke
level modelling. We (Dunlop & Crossan, 2000)
proposed a keystroke level model based on Card,
Moran and Newall's work (S. K. Card, Moran, &
Newell, 1980). Our model was based on predict-
ing the time T(P) taken by an expert user to enter
a given phrase. The model calculates this in an
equation that combines a set of small time measure-
ments for elements of the user interaction. In the
case of text entry: the homing time for the user to
settle on the keyboard T h (0.40 seconds); the time
it takes a user to press a key T k (0.28s); the time
it takes the user to mentally respond to a system
action T m (1.35s); the length of an average word
k w (in our study, 4.98); and the number of words
in the phrase w (in our model, 10). In addition,
for predictive text entry where disambiguation
occurs by the user moving through the ranked
list of suggestions, the ARP value is required here
given as a =1.03). The overall time equation for
entering a phrase is then given as follows:

T(P) = T h + w (k w T k + a(T m + T k )) (1)

Equation 1: Dunlop and Crossan's keystroke
model. This model, as reported in (Dunlop &
Crossan, 2000) and corrected by (Pavlovych &