Learning from natural instructions (Q2251459)

In this interesting paper, the authors suggest a machine learning approach based on parsing the inputs as natural language commands and converting them into the logical fragments. They use combined feedback perceptrons learning from both structural [\textit{M. Collins}, ``Discriminative training methods for hidden Markov models: theory and experiments with perceptron algorithms'', in: Proceedings of the ACL-02 conference on empirical methods in natural language processing, EMNLP '02. Stroudsburg, PA: Association for Computational Linguistics. 1--8 (2002; \url{doi:10.3115/1118693.1118694})] and binary perceptron learning rules [\textit{F. Rosenblatt}, Principles of neurodynamics. Perceptrons and the theory of brain mechanisms. Washington, D.C.: Spartan Books (1962; Zbl 0143.43504)]. The proposed learning approach differs from the well-known reinforcement learning methods in the sense that it integrates the interpretation of natural language advice into a learner. The work originates from [\textit{G. Kuhlmann} et al., ``Guiding a reinforcement learner with natural language advice: initial results in RoboCup soccer'', in: The AAAI-2004 workshop on supervisory control of learning and adaptive systems. 30--35 (2004)]. In addition, the semantic interpreter generalizes to unseen instructions in the learning process. The authors also examine the performance of the proposed learning approach on instructional text for some variations of card games.