<?xml version="1.0" standalone="yes"?>
<Paper uid="H93-1074">
  <Title>Mode preference in a simple data-retrieval task</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
INTRODUCTION
</SectionTitle>
    <Paragraph position="0"> Multi-modal systems allow users to both tailor their input style to the task at hand and to use input strategies that combine several modes in a single transaction. As yet no consistent body of knowledge is available for predicting user behavior in multi-modal environments or to guide the design of multi-modal systems. This is particularly true when interfaces incorporate new technologies such as speech recognition. null For activities in a workstation environment, formal comparisons of speech with other input modes have failed to demonstrate a clear advantage for speech on conventional aggregate measures of performance such as time-to-completion \[1, 8, 4\], despite a consistent advantage displayed by speech at the level of single input operations. The difference can actually be attributed to the additional incurred costs of non-real-time recognition and error correction. While real-time performance can he achieved, it is unlikely that error-free recognition will be available in the near future. Given these shortcomings, we might ask if speech can provide advantages to the user along dimensions other than task speed, for example by reducing the effort needed to generate an input.</Paragraph>
    <Paragraph position="1"> There is reason to believe that users are quite good at estimating the response characteristics of an interface and can choose an input strategy that optimizes salient aspects of performance, for example decreasing time-to-completion or minimizing task error \[5, 9\]. By observing the behavior of users in a situation in which they can freely choose between different strategies, we can gain insight into the factors that govern their preference for different input styles.</Paragraph>
    <Paragraph position="2"> A simple data retrieval task was chosen for this study, as the task was one amenable to execution in each of the three modalities that were examined: speech, keyboard and scroller. The database contained information about individuals, such as address, telephone, etc selected from a list of conference attendees. The task consisted of retrieving the record for an individual and recording the last group of digits in their work telephone number (typically of length four). The database contained 225 names for the first experiment and was expanded to 240 names for the second experiment.</Paragraph>
  </Section>
class="xml-element"></Paper>