SISU: SpeechState meets Information State Update (ISU)

The repo contains an ISU implementation described in: Combining Information State Update, Harel Statecharts and LLMs for controllable and flexible Conversational AI (Maraev et al., CLASP 2025)

It also contains a partial implementation of the dialogue system IBiS1 (Larsson, Staffan (2002): Issue-based Dialogue Management), including a small domain.

Prerequisites:

1. Install NodeJS (LTS version) and Yarn (v2 or higher).
2. Install all dependencies:

   yarn
       

3. Create src/azure.ts, insert your Azure key where marked. This is required for running the system with speech (in browser frontend), but not for automated testing (in backend).

   export const KEY = "<your_key_goes_here>"
       

Run SISU:

1. Run backend in development mode:

   yarn dev
       

2. Open the link that was shown in your browser, e.g. http://localhost:5173/
3. Allow access to you microphone.
4. When you unblock the pop-up window, and reload the page you will see the state inspector; this can be useful during development.

Run automated tests:

yarn test

Cite this work:

@inproceedings{maraev_combining_2025,
	author = {Vladislav Maraev and Alexander Berman and Staffan Larsson},
	title = {Combining Information State Update, Harel Statecharts and {LLM}s for controllable and flexible Conversational {AI}},
	editor = {Ilinykh, Nikolai  and
      Appelgren, Mattias  and
      Lagerstedt, Erik},
	booktitle = {Proceedings of the 2025 CLASP Conference on Language models And RePresentations (LARP)},
	year = {2025},
	address = {Gothenburg, Sweden},
	publisher = {Association for Computational Linguistics},
	url = {https://aclanthology.org/2025.clasp-main.3/},
	pages = {28--37},
	ISBN = {979-8-89176-249-7},
	abstract = {The rise of LLM-based approaches to dialogue systems has created an increased need for controllable dialogue. This paper addresses this need by presenting an implementation of a dialogue system based on information state update approach according to Larsson (2002). This enables the integration of rule-based handling of dialogue, expressed by Harel{'}s statecharts (1987), and Larsson{'}s theoretical account grounded in theories of dialogue, expressed by information state update rules. We demonstrate how our approach applies to dialogue domains involving form-filling. We also propose how LLMs can be employed to inject domain knowledge and be used in various components of a hybrid dialogue system, while maintaining control over the overall dialogue logic.}
}