Title: How Can Computer Analyses of Language, Text, and
            Conversation Be Woven into Learning and Assessment?
       Speaker: Prof. Arthur C. Graesser  (University of Memphis)
       Date: 9:30-11：00 am. August 28
       Place: Room 201, School of Psychology

Abstract
        Recent advances in computational linguistics and discourse processing have made it possible to analyze naturalistic texts and conversation on multiple    levels of language and discourse.  These advances are influencing the world of assessments of reading, writing, mathematics, science, reasoning, problem solving, and other competencies.  This presentation reports examples of these assessments that analyze natural language.  Coh-Metrix (http://cohmetrix.com) analyzes texts on multiple measures of language and discourse that are aligned with multilevel theoretical frameworks of discourse comprehension and production.  Several dozen measures funnel into five major factors that systematically vary as a function of types of texts (e.g., narrative versus informational) and grade level: narrativity, syntactic simplicity, word concreteness, referential cohesion, and deep (causal) cohesion.  A composite measure called formality increases with low narrativity, syntactic complexity, word abstractness, and high cohesion.Coh-Metrix has also been used to analyze student writing and conversation, even though its central focus is on scaling printed text.
        Learning has occurred over millennia by the learner communicating with the teacher, tutor, master, or mentor in natural language.  Apprenticeship learning has always occurred one-on-one or in small groups with an expert.   Researchers in the discourse and learning sciences have documented the conversation patterns that occur in these interactions.  Researchers in computational linguistics, artificial intelligence, and intelligent tutoring systems have developed computer agents that simulate many of these conversation patterns and help people learn.  This presentation will present recent systems on the internet that help students learn by holding a conversation in natural language.  AutoTutor engages in dialogue with the student on a variety of subject matters in Science, Technology, Engineering, and Mathematics.  Trialogs are conversations between the human students and two computer agents, typically a student agent and a tutor agent.  Students can either observe two agents interact vicariously, interact with a tutor agent as a student agent periodically chimes in, or teach a student agent while a tutor rescues a problematic interaction.  Agents can argue with each other over issues and ask what the human students think about the argument.  Trialogs are being developed for the Internet in serious games with Pearson Education (Operation ARA), in assessments with Educational Testing Service, in training environments with the US Army and Navy, and in a new Center for the Study of Adult Literacy for struggling adult readers.  AutoMentor is being developed for computer mediated communication between a mentor agent and small groups of students in a simulation game on urban planning.  Tests of these systems have shown very encouraging learning gains.