Full Program »

An Intelligent Tutoring System for Learning Chinese with a Cognitive Model of the Learner

Our poster presents an Intelligent Tutoring System that enables students of Chinese to acquire active knowledge of lexical and grammatical constructions. The system relies on a Bayesian, linguistically motivated cognitive model that represents the estimated knowledge of the learner, and the difficulty of the constructions. This model is dynamically updated given observations about the learner's proficiency in the exercises. The model is then employed at runtime to select the exercises that are expected to maximise the learning outcome.

The tutoring model consists of English-to-Chinese translation exercises. Each exercise is associated with a set of correct solutions, represented in a template format that encodes a large number of alternative translations and synonyms in a compact manner. When faced with an incorrect answer, the system selects the solution with the shortest distance to the input, as measured by the BLEU score [Papineni et al. 2002]. It also provides interactive feedback, using a combination of error-specific and generic rules that provide relevant cues towards the closest correct translation.

The student model contains information about the user's language level. Actual user interactions provide evidence about which linguistic constructions were employed by the user, which ones were not produced (or produced incorrectly), and which constructions were looked up in a provided dictionary. These factors are represented in a Bayesian network and let the system reason about the learner's current knowledge. The model also contains information about the learnability of different constructs, using past interactions to determine how probable it is that a user at a given level fails to remember a given construction. This information is used to determine which constructions are most likely to lie within Vygotsky's Zone of Proximal Development, in order to choose the exercise that is most likely to maximise the learning outcome at the next time step.

In our first experiment, we invited learners of Chinese to use the system and collected measures of their language level. Subjective measures represent the students' self assessments of their Chinese proficiency, while objective measures are automatically estimated from a character recognition test. The data collected in this experiment will be exploited to tune the parameters of the learner's cognitive model.

Subsequent experiments will look into the empirical effects of such a cognitive model on the selection of relevant translation exercises. We will also investigate how to integrate long-term memory effects into the cognitive model, for example the spacing effect, modelled in Pavlik and Anderson's[2003] extension of ACT-R. No constructions are retained forever without repetition, but teaching same words again during every session is definitely suboptimal, therefore such a model is important for any tutoring system that is going to be used over long periods of time.


Papineni, K. et al. (2002). “BLEU: a method for automatic evaluation of machine translation”. Proceedings of ACL, pp. 311–318

Pavlik, P.I., & Anderson, J.R. (2003). “An ACT-R model of the spacing effect”. In F.Detje, D.Doerner & H.Schaub (Eds.), Proceedings of the Fifth International Conference on Cognitive Modeling, pp. 177-182


Michał Kosek    
Department of Informatics
University of Oslo

Michał Kosek is a master's student in Language Technology at the Department of Informatics and a software engineer at the Text Laboratory, Department of Linguistics and Scandinavian Studies, University of Oslo. He has a Master's degree in Computer Science and a Bachelor's degree in Linguistics with specialisation in Chinese language. His research interests have been focused at the intersection of computational linguistics, second language acquisition, intelligent computer-assisted language learning and psycholinguistics.

Pierre Lison    
Department of Informatics
University of Oslo

Pierre Lison is a PhD Research Fellow at the Department of Informatics of the University of Oslo, where he works since 2011. He submitted his PhD dissertation at the end of 2013 on “Structured Probabilistic Modelling for Dialogue Management”. Before coming to Oslo, he worked for 2 years as a researcher at the German Research Centre for Artificial Intelligence (DFKI) in Saarbrücken, where he was involved in several EU-funded research projects. His general research interests lie in the development of rich statistical models for interactive systems.


Powered by OpenConf®
Copyright ©2002-2013 Zakon Group LLC