ESTUDIO CLIMÁTICO
5. ANÁLISIS DE LS TEMPERATURAS 1 Resumen de temperaturas
An ITS “aims to support and improve the learning and teaching process in certain domain knowledge, considering individuality of a student like in traditional one-to-one instructional process” (Stankov et al., 2007, p.1). Although one-to-one human tutoring is superior to other forms of learning (Bloom 1984), it is difficult to provide due to practical implications such as cost and availability of human tutors. Consequently, this has led to the development of ITS with an ECA as a tutor (Graesser et al., 1999; Heylen et al., 2005; Prendinger et al., 2005) or without an embodiment (Stankov et al., 2007; Keleş et al., 2008; Rau 2013).
A number of studies put forward theories based on the success of one-to-one human tutoring. The aim of these studies is to create a system that models effective behaviours of human tutors and can improve learning more than a normal classroom situation (Zinn et al., 2005). Students seem to learn best when they are active learners. ITS using ECAs seek to emulate the collaborative nature of a human tutor working with a student to correct errors (Wiemer-Hastings et al., 1998; Graesser et al., 1999). Graesser et al. (1999) focus on developing an artificial tutor that imitates the strategies of untrained human tutors, as trained human tutors use complex strategies developed through extensive training which may not be easily implemented within artificial tutoring systems. The success of the pedagogical role of a tutor with an embodiment is discussed in greater detail in the following paragraphs.
26
The adaptation of this strategy with computer to human training has been demonstrated by Graesser et al. (2003; 2004) to have a significant effect on learning gain when compared to reading a text book for an equivalent time to a tutor session with AutoTutor an artificial tutor. This has been demonstrated a number of times in varied domains such as computer literacy and conceptual physics. AutoTutor is a system that uses an ECA to engage in an explanation-centred dialogue using natural language with learners in a specific domain. Figure 2.1 contains an extract from a short tutorial dialogue between AutoTutor and a high ability college student in the domain of computer literacy (Graesser et al., 1999).
Figure 2.1 Extract from AutoTutor Dialogue
AUTOTUTOR: When you turn on the computer, how is the operating system first activated and loaded into RAM?
STUDENT: When the computer is turned on a file is automatically booted up. AUTOTUTOR: Anything else?
STUDENT: The file contains the operating system. AUTOTUTOR: What else?
STUDENT: The operating system is then read into RAM and activated.
AUTOTUTOR: Right the CPU executes instructions in ROM, which help the operating system to start.
27
The results achieved from this research suggest that AutoTutor can be as effective as a human tutor communicating with a student using a computer-based medium (Graesser et al., 2004). Limitations of the system are that: firstly AutoTutor provides cognitive feedback alone and disregards learner emotion; secondly even if the system were to provide affective feedback the use of a talking head would limit the effectiveness of implementing non-verbal communication such as gesture and posture and would rely on facial expression alone. An affective tutoring system would require implementation using a half or full bodied embodiment to fully imitate human emotion.
AutoTutor’s architecture is based on an explanation-centred tutorial interaction that uses state-of-the-art technology such as Latent Semantic Analysis (LSA) and fuzzy rules. LSA is used to compress the domain corpus and then compare these with student answers or contributions (AutoTutor’s effectiveness using LSA is similar to an untrained human tutor and lower when compared to an expert human tutor). Also, the system uses fuzzy rules to choose the next topic. This makes direct comparisons of AutoTutor’s architecture (implementing explanation-centred learning) to the current study (implementing quiz-based learning) difficult to make because of the difference in the type of interaction with students. However the strategies developed for the tutorial dialogue moves and productions rules are used in the current study and are discussed in sections 2.4.2 and 3.3.2 in greater detail.
28
The AutoTutor system architecture includes rules that “specify which dialogue response the tutor will make after a student turn. These are based on the content of the curriculum script, the dialogue history, and the quality of the student’s contribution during the last turn, the cumulative quality of the student’s knowledge, and the cumulative quality of the student-tutor exchange.” (Wiemer-Hastings et al., 1998). For example, the following simple production rule is associated with immediate positive feedback for student contribution C:
IF [scriptComponent = Question(j)
AND max(similarity(C, good-answer(j)) > Threshold]
THEN [produceFeedback: "That’s right."]” (Wiemer-Hastings et al., 1998)
Similarly to AutoTutor, the current study uses an ECA within a tutorial interaction where learning gain is measured. The current approach differs from the work on AutoTutor in that it focuses on tutorial support for students that are completing a multiple choice quiz rather than a fully-fledged explanatory dialogue with learners.
Research in this area is looking at how intelligent tutors can further enhance learning. This includes developing complex feedback strategies, that can easily handle student errors and topic changes to improve learning (Graesser et al., 2005). One strand of research in artificial tutoring builds on the work on ITS, with an ECA that provides cognitive feedback, and seeks to address learner emotion by incorporating affect detection (D'Mello et al., 2006) and affect expression through feedback strategies
29
within dialogue used with learners (Arroyo et al., 2009b). Combined affective and cognitive feedback such as giving hints, summarising learning and corrective and metacognitive tips aim to encourage learners’ effort, focus and time spent on a task. These strategies aim to impact on learner states such as motivation (Keller 1987) and self-efficacy consequently improving learning (McQuiggan et al., 2008; Dennis et al., 2011). The success of these strategies has not yet been conclusively proven as these learner emotional states are difficult to measure (Hussain et al., 2011) in order to enhance learning gain.
Recently, researchers are investigating empathic ECAs that can be extended to share physical spaces with users as empathic robots tutors, for example the EMOTE project (Castellano et al., 2013). However this research is in its infancy as evaluations have not been conducted on the impact on learning gain. Section 2.2 describes how empathy, a type of affect expression strategy can be incorporated into artificial tutorial feedback to enhance learning gains by reducing negative emotion such as frustration during interactions with artificial tutors (Burleson 2006).