Introduction

This chapter presents the implementation result of a UIA that is based on the DIRECTOR architecture (Figure 8.2) and that accepts user instructions in natural language. The UIA is implemented in Alma/Carne (Section 4.3); therefore, the UIA KB changes based on $\mal$ axioms discussed in Section 5.3. The implemented UIA models the beliefs, desires, intentions, expectations and achievements of the agent and reasons based on a version of the theory of acting and agency (ALFA) presented in Chapter /refalfa. The implemented UIA is the dialog agent ALFRED (Active Logic For Reason Enhanced Dialog) [AJOP02,JAP03,AJP03,JAP04,JAP05b,JAP05a] that can exhibit some of the meta-linguistic skills that are necessary for a truly viable natural language human-computer interface. To reduce the complexity of the dialog problem, ALFRED is always connected to some task-oriented system (TOS) and the communication between ALFRED and a human user is related to the operation of that task-oriented system. The ultimate task of ALFRED is to communicate with a human user in natural language to control different task-oriented systems. The specific parser that Alfred uses is the Link Parser [GLS95],a syntactic parser of English, that is based on Link Grammar [ST93]. The syntactic structure that the Link Parser produces consists of a set of labeled links connecting pairs of words that are associated semantically and syntactically. In general, ALFRED interprets the user requests and creates desires to accomplish the requests. Based on knowledge and availability of time and resources, ALFRED creates intentions to achieve the desires and thus issues the appropriate instruction to the TOS. In order to determine whether an instruction that it issued to a TOS has resulted in the desired action, ALFRED creates an expectation regarding the outcome, whenever it initiates an action. It then compares the actual result that it obtains from either the user or the TOS with the expected outcome.