Using context and sensory data to learn first and second person pronouns

We present a method of grounded word learning that is powerful enough to learn the meanings of first and second person pronouns. The model uses the understood words in an utterance to focus on the agents to which they refer. The method then uses chi-square tests to find significant associations between the remaining words and the attributes of the relevant agents. We show that this model can learn from a transcript of a parent-child interaction taken from the CHILDES database [22] that “I” refers to the person who is speaking. With the additional information that questions about wants refer to the addressee, the system can also learn the meaning of “you” from observed dialogue. We show that an incorrect assumption about the probable referent of “want” questions can lead to pronoun reversal, a linguistic error most commonly found in autistic and congenitally blind children. Finally, we present results from a physical implementation on a robot that runs in real time. Our ...

• Type: Conference
• Year: 2006
• Where: HRI
• Authors: Kevin Gold, Brian Scassellati
• Keywords: Grounded Word, HRI 2006, Human Robot Interaction, Natural Language, Person Pronouns