IEEE Robotics & Automation Magazine - June 2019 - 84

prompts are arranged from most to least intrusive, errors
may be less frequent.
CTD
In CTD, prompts are delivered after a time delay following a
task direction, i.e., a cue or question for the student. Unlike
SMP or SLP, there is only one prompt: the controlling prompt.
Only the time between the cue and the prompt, known as the
prompt delay interval, is varied. Initially, the delay between the
task direction and the controlling prompt is zero, in what is
termed the zero-second delay trial. The prompt delay interval
is constant for a set of instruction trials until the criterion is
met; it is then systematically increased. The goal of CTD is a
consistently correct response before the prompt.
Chained and Discrete Tasks
In addition to prompting strategies, the manner in which the steps
of the task can be taught is also a consideration. Discrete tasks are
tasks for which a single correct response is expected, such as sight
words (commonly used words that students are taught to memorize as a whole by sight). Some discrete tasks can be subdivided
into smaller sequences of tasks as necessary for instruction.
Chained tasks are sequential in nature. Instruction on
chained tasks is conducted step by step within the sequence.
Examples of a chained task include most building tasks, such
as building a structure (e.g., from the ground up), assembling
an object or puzzle, and so on. Because of their sequential
nature, chained tasks can be taught from the beginning of the
sequence in what is known as forward chaining or, by iterating
from the end of the sequence, backward chaining. Tradeoffs
exist for both. This work uses forward and backward chaining
as appropriate to the task and examines how they impact successful learning from intelligent systems.
Creation of Prompting Hierarchies
The process of creating a prompt hierarchy involves devising
a series of prompts and arranging them in a hierarchical
structure appropriate for the instruction strategy (e.g., SLP or
SMP). The dimensions of this structure are dictated by the
student's response space (e.g., correct, partially correct, incorrect, or no response), the discretization of task steps, and the
modalities and intrusiveness of the response prompts.
To ensure successful instruction, the prompt content and
modalities, as well as the level of intrusiveness, should be
appropriate for the student's capabilities and diagnoses. For
these reasons, we strongly recommend collaboration with an
education domain expert for this process. For the following
experiments, our interdisciplinary team of authors collaborated closely to ensure a successful and productive experience
for the students involved.
Experiments
Single-Case Experimental Design
Experiments were developed for this research using a singlecase experimental design (SCED), a common design method
84

IEEE ROBOTICS & AUTOMATION MAGAZINE

JUNE 2019

in special education research. Rather than comparing groups
or subjects, in SCED participants serve as their own control,
with performance in at least two experimental phases being
compared [20]. SCED methods are used in place of statistical
methods for large groups. This is not only because recruiting
a large number of participants with I/DDs is infeasible but
also because, even if it were possible, the participants' capabilities and diagnoses could be so diverse that drawing even a
coarse statistical inference would be challenging. The objective of the SCED is to determine whether a causal or functional relationship exists between the delivery of the
independent variable (the intelligent instruction system) and
significant increases in the dependent variable (the acquisition and maintenance of the skills taught).
Two types of SCED designs were used for this research: a
combined multiple baseline across participants and a combined multiple baseline across skills. These designs allow for
the evaluation of intervention effects while controlling for
threats to internal validity (i.e., that the learning is due to the
instructional intervention) in situations where alternate
designs are not feasible, such as those that would require
withdrawal of skill knowledge.
In experiments using a multiple baseline across skills, skills
are taught one at a time, and instruction is introduced for
each skill sequentially after the previous one has been learned.
In experiments using a multiple baseline across participants,
each student is taught one at a time, and instruction is introduced to each successive student after the previous student
finishes learning the skill. In all cases, baselines are taken
before instruction, up to the point where the instruction
begins, and probes are made after successful demonstration of
the skills to measure retention.
By introducing the intervention subsequently across a
minimum of three replications, the possibility that any
observed change occurrs due to extraneous factors (e.g., practice or history effects) is eliminated. This allows for experimental control and establishing a causal relationship [21].
Student Participant Population
All studies for this research were performed in accordance
with institutional review board protocols and approval. The
students who participated in these studies were college age
and attendees of a postsecondary education (PSE) program
designed for young adults with I/DDs at the University of
Tennessee, Knoxville, named FUTURE (more information
on the FUTURE program is available at http://futureut.utk
.edu). All students 1) were ages 18-34 with an intelligence
quotient between 57 and 67, 2) received special education services throughout school, and 3) earned modified
high school diplomas before participating in the
FUTURE program.
To meet the three replication minimum requirements for
evaluating the intervention effects, three students per experiment were taught until our criteria were reached, i.e., they
mastered the skill being taught. Seven students participated
overall, four of whom participated in multiple experiments.

https://futureut.utk.edu/ https://futureut.utk.edu/

IEEE Robotics & Automation Magazine - June 2019

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2019