AIML+: Enhancing AIML for the Educational Domain Through Frames and Large Language Models

In educational applications, the Dialogue Systems (DSs) have a disruptive impact. Artificial Intelligence Markup Language (AIML) is a standard DS framework adopted by many educators since it requires minimal programming skills and allows for fast development. However, a major drawback of AIML is its limited natural language understanding (NLU) capability. AIML relies on regular expressions applied to the surface form of the input sentence for computing meaning and this mechanism is often brittle. In contrast, industrial frameworks for developing DSs (e.g., for customer care) are generally frame-based, where the meaning of the input sentence is conveyed through a set of fillers assigned to predefined slots. In this paper we propose AIML+, a framework that extends the AIML with frame-based NLU mechanism. Our aim is to develop an advanced, more robust, version of AIML that remains easy to use, ensuring it is appealing to educators. The contributions of this paper are threefold. First, we define AIML+ by incorporating into AIML several NLU features typical of frame-based frameworks. Second, we propose and release an implementation of the AIML+ engine, which employs Large Language Models to assess the effectiveness of prompting and fine-tuning for slot filling in NLU. Third, we release an annotated corpus, in the specific educational domain of Finite State Automata, used during fine-tuning as a case study for AIML+.