An Hybrid Approach for Spoken Natural Language Understanding Applied to a Mobile Intelligent Robot Mário Rodrigues 1 , António Teixeira 2 , and Luís Seabra Lopes 2 1 Escola Superior de Tecnologia e Gestão de Águeda, Universidade de Aveiro, 3754-909 Águeda, Portugal 2 Departamento de Electrónica e Telecomunicações/IEETA, Universidade de Aveiro, 3810-193 Aveiro, Portugal Abstract. The greater sophistication and complexity of machines increases the necessity to equip them with human friendly interfaces. As we know, voice is the main support for human-human communication, so it is desirable to interact with machines, namely robots, using voice. In this paper we present the recent evo- lution of the Natural Language Understanding capabilities of Carl, our mobile intelligent robot capable of interacting with humans using spoken natural lan- guage. The new design is based on a hybrid approach, combining a robust parser with Memory Based Learning. This hybrid architecture is capable of performing deep analysis if the sentence is (almost) completely accepted by the grammar, and capable of performing a shallow analysis if the sentence has severe errors. 1 Introduction Recent developments For these robots to emerge it’s essential the development of nat- ural language interfaces, regarded as the only acceptable for a high level of interac- tion [1]. Voice allows hands free communication even without visual contact, great advantages if the machine is a mobile robot. In this line of research, we are develop- ing a mobile intelligent robot named Carl. Currently the development of such robots is still a challenge due to limitations of current technologies and the nature of input information: speaker independent speech recognition is not very reliable, even in quiet environments; performance degrades considerably with background noise; spontaneous spoken language is often highly disfluent. The use of spoken language interfaces in robots requires analysis components robust to various types of disfluencies that can extract the most complete interpretation possi- ble from a given input, grammatically correct or not [2]. It would be of little application a robot that in all the cases the speech recognizer makes an error doesn’t react. In this paper we present the current status of the spoken natural language interface of robot Carl focusing in our implementation of a robust Natural Language Understanding (NLU) module. We start by a brief presentation of the robot, in Section 2. Next, in Section 3 we describe the previous NLU modules, for an easier understanding of our new approach. The new approach is presented in Section 4, and results obtained with these new developments are the object of Section 5. Paper ends with results discussion and indications of ongoing and future work. Rodrigues M., Teixeira A. and Seabra Lopes L. (2004). An Hybrid Approach for Spoken Natural Language Understanding Applied to a Mobile Intelligent Robot. In Proceedings of the 1st International Workshop on Natural Language Understanding and Cognitive Science, pages 145-150 DOI: 10.5220/0002667001450150 Copyright c  SciTePress