Homecare in Natural Language


Despite the advances in dialog systems, most of the dialogue-based interfaces implemented in the medical domain are devoted to simple approaches to dialogue management. These applications offer system-driven dialogues that limit the variety of expressions users might use. Thereby, patients might have the impression of being following unnatural conversations. In general, this technology is targeted to patients who might be either affected by different diseases, perhaps chronic, or might belong to a risk group (e.g., elders). Therefore, having unnatural conversation would be for many of them not only annoying but also ineffective when handling their own care at home. This issue clearly produces a negative impact on patients' comfortability.


HomeNL (Homecare assistance in natural language) is a proof-of-concept conversational assistance for hypertensive patients monitoring. It is a laboratory for studying NLP and intelligent dialogues in the medical domain. This proof of concept explores theoretical approaches for building conversational interfaces. The development process of HomeNL has included issues such as selecting an appropriate theory for dialogue management, language understanding and generation as well as knowledge-bases and reasoning.




Several components were arranged into a distributed multiagent architecture: the open agent architecture (OAA). These components are in charge of a specific linguistic task. For instance, the language understanding and generation components were both modeled in the linguistic-driven formalism of multimodal-categorial grammars (MMCCG) and wrapped as an OAA-agent. DIPPER, the dialogue manager, already provides OAA-agents for managing the dialogue. In addition, the Festival speech synthesis was used as text to speech (TTS) and wrapped as an OAA-agent. The other agents are part of the core of HomeNL. The behavioral agent accesses the knowldege representation (KR), which was implemented in LISP and RACER was used as inference engine.

HomeNL principal goals:

  1. to Fully enable mixed-initiative by supporting language understanding and generation.
  2. to use logics-based knowledge representations and reasoners.
  3. to study discourse, intentions recognition as well as phenomena as ellipsis, anaphora and grounding in homecare dialogues.
  4. to explore theoretical approaches to dialogue.
  5. to offer an extensive architecture for the incremental integration of other agents, such as embodied conversational agents ECA's and ASRs.

The domain of HomeNL is the same of the dialogue system HOMEY. So far, the grammar supports expressions commonly used between phisician and patients as collected in this project.


HomeNL is an active research project at the Laboratory for Biomedical Informatics (Università degli Studi di Pavia).

Labmedinfo is looking for partnerships for:
  - training an ASR with medical terminology.

For students:

Extending the coverage of the CCG grammar on the basis of the Italian guidelines for the management of patients with hypertension.

Integrating Risk Management in the Knowledge Representation.

Integrating a ASR with medical terminology.