A Result based Portable Framework for Spoken Language Understanding

Lizhi Cheng, Weijia Jia, Wenmian Yang

Abstract

Researchers have developed RPFSLU, a "plug-and-play" framework that upgrades standard single-turn Spoken Language Understanding (SLU) models to handle complex, multi-turn dialogues. By treating base models as black boxes and feeding back historical prediction results as semantic context, the framework significantly boosts performance across all tested architectures without requiring internal model changes.

Introduction

Spoken Language Understanding (SLU) is the engine behind voice assistants like Siri or Alexa. It involves two main tasks: Intent Detection (ID) (e.g., "The user wants to Navigate") and Slot Filling (SF) (e.g., "The destination is Subway").

While single-turn SLU is well-researched, multi-turn dialogues are much harder. Current approaches face two major hurdles: low portability and wasted information. Most multi-turn models are tightly integrated, making it difficult to swap in new single-turn models. Furthermore, most systems ignore the results (predicted intents and slots) of previous turns, which contain dense semantic clues about the conversation's direction.

The RPFSLU Framework

The authors propose the Result-based Portable Framework for SLU (RPFSLU). Its portability comes from treating the underlying SLU model as a black box. The framework consists of three main pillars:

The RPFSLU architecture, featuring the Dialogue History Representation (DHR) and the Result-based Bi-Feedback Network (RBFN).
  • Result Representation Mechanism: Instead of just looking at labels as categories, the framework maps predicted ID and SF results into a "latent state space" (vectors). This allows the model to understand the meaning of a result (e.g., knowing that a "Navigate" intent is semantically related to "Distance" or "Address" slots).
  • Dialogue History Representation (DHR): This module uses an attention mechanism to weigh previous turns based on how relevant they are to the current utterance, combining both the historical text and the historical predicted results.
  • Result-based Bi-Feedback Network (RBFN): This is a two-round prediction process. In Round 1, the base model makes an initial guess. In Round 2, these initial guesses are fed back into the model. The predicted intent helps refine the slot filling, and the predicted slots help verify the intent.

Experimental Setup

The framework was evaluated on the KVRET (Stanford Multi-turn Dialogue Dataset), which contains 3,031 multi-turn dialogues across calendar scheduling, weather inquiry, and point-of-interest navigation. The authors tested RPFSLU against several established models, including MemNet, SDEN, and state-of-the-art architectures like Stack-Propagation.

Key Results and Insights

The RPFSLU framework provided a visible improvement for every model tested. Results are semantically rich; a predicted label like RestaurantName is a strong signal that the user is in a "booking" mindset. By embedding these results into vectors, the model captures this hidden context.

Performance comparison of various SLU models with and without the RPFSLU framework on the KVRET dataset.
  • State-of-the-Art Gains: When applied to the Stack-Propagation model, it improved Intent Accuracy by 1.2%, Slot Filling (F1) by 4.3%, and Overall Accuracy by 4.1%.
  • Elevating Simple Models: A basic 1-layer BiLSTM wrapped in RPFSLU actually outperformed several complex, specialized multi-turn benchmarks (like SDEN+).
  • Mutual Verification: The Bi-Feedback loop mimics human logic—if a user wants to book a restaurant (Intent), the system is more likely to correctly identify a name as a restaurant (Slot).

Conclusion

RPFSLU demonstrates that historical prediction results are precious information for multi-turn SLU. While the two-round prediction process may increase computational overhead, the framework's portability makes it a future-proof solution for dialogue systems. Future work will explore more complex ways to integrate latent state results with word embeddings to further enhance performance.

Summary updated

BibTeX
@inproceedings{cheng2021result,
  author = {Lizhi Cheng and Weijia Jia and Wenmian Yang},
  title = {{A Result based Portable Framework for Spoken Language Understanding}},
  booktitle = {ICME},
  year = {2021},
  doi = {10.1109/icme51207.2021.9428384},
}
Figure