Interpretable Intent Detection in High-Cardinality Scenarios via Dynamical Systems Analysis

• Autores: Eduardo Sánchez Karhunen, José Francisco Quesada Moreno , Miguel Ángel Gutiérrez Naranjo
• Localización: Procesamiento del lenguaje natural, ISSN 1135-5948, Nº. 76, 2026 (Ejemplar dedicado a: Procesamiento del Lenguaje Natural, Revista nº 76, marzo de 2026), págs. 25-37
• Idioma: inglés
• Títulos paralelos:
  • Reconocimiento de Intenciones en Alta Cardinalidad: Interpretación mediante Sistemas Dinámicos
• Enlaces
  • Texto completo
• Resumen
  • español

    La falta de transparencia del aprendizaje profundo limita la confianza en la detección de intenciones. Si bien la teoría de sistema dinámicos ha permitido interpretar las RNNs, su aplicación en escenarios de alta cardinalidad, propios de productos reales, permanece inexplorada. Extendemos este marco analítico a benchmarks con hasta 150 intenciones. Mostramos que las RNNs convergen hacia una solución geométrica interpretable, organizando su espacio de fases en clústeres robustos para cada intención. Observamos que la dimensionalidad intrínseca crece de forma sublineal respecto a la complejidad de la tarea. Basándonos en esto, introducimos la Dimensionalidad Funcional (DF), una métrica novedosa que cuantifica la dimensión mínima necesaria para preservar dicha estructura semántica. Nuestro análisis revela DFs notablemente bajas, lo que sugiere que las RNNs resuelven tareas complejas mediante un subespacio eficiente y organizado, donde los clústeres se alinean con sus vectores de decisión. Todo ello ofrece un marco escalable para auditar e interpretar sistemas de diálogos en entornos de alta cardinalidad.

  • English

    Trustworthy intent detection is limited by deep learning opacity. While dynamical systems theory has emerged as a powerful tool for interpreting Recurrent Neural Networks (RNNs), its application has been unexplored in high-intent, large scale scenarios common to real-world products. We extend this analytical framework to benchmarks with up to 150 intents. We find RNNs trained on these tasks still converge to an interpretable geometric solution, forming robust, intent-specific clusters in their hidden space. We show this space’s intrinsic dimensionality grows sub-linearly with task complexity. Building on this, we introduce Functional Dimensionality (FD), a novel, task-aware metric that quantifies the minimum dimensionality required to preserve this semantic structure. Our analysis reveals FD is remarkably low, suggesting RNNs solve complex tasks via an efficient, highly organized subspace. We show this subspace is structured for inference, with clusters aligning strongly with their corresponding readout vectors. These findings offer a scalable framework for auditing and interpreting high-intent dialogue systems.

• Referencias bibliográficas
  • Abadi, M., P. Barham, J. Chen, Z. Chen, A. Davis, J. Dean, M. Devin, S. Ghemawat, G. Irving, M. Isard, M. Kudlur, J. Levenberg, R. Monga,...
  • Abro, W. A., G. Qi, M. Aamir, and Z. Ali. 2022. Joint Intent Detection and Slot Filling Using Weighted Finite State Transducer and BERT. Applied...
  • Aitken, K., V. V. Ramasesh, A. Garg, Y. Cao, D. Sussillo, and N. Maheswaranathan. 2021. The Geometry of Integration in Text Classification...
  • Arrieta, A. B., N. Díaz-Rodríguez, J. Del Ser, A. Bennetot, S. Tabik, A. Barbado, S. Garcia, S. Gil-Lopez, D. Molina, R. Benjamins, R. Chatila,...
  • Basu, S., A. Banerjee, and R. J. Mooney. 2002. Semi-supervised Clustering by Seeding. In Proceedings of the 19th International Conference...
  • Casanueva, I., T. Temcinas, D. Gerz, M. Henderson, and I. Vulic. 2020. Efficient Intent Detection with Dual Sentence Encoders. In Proceedings...
  • Cho, K., B. van Merrienboer, C. Gulcehre, D. Bahdanau, F. Bougares, H. Schwenk, and Y. Bengio. 2014. Learning Phrase Representations using...
  • Coucke, A., A. Saade, A. Ball, T. Bluche, A. Caulier, D. Leroy, C. Doumouro, T. Gisselbrecht, F. Caltagirone, T. Lavril, M. Primet, and J....
  • Das, A. C., Greg Phalin, I. L. Patidar, M. Gomes, R. Sawhney, and R. Thomas. 2023. The Next Frontier of Customer Engagement: AI-Enabled Customer...
  • Elman, J. L. 1990. Finding Structure in Time. Cognitive Science, 14(2):179–211.
  • Gkinko, L. and A. Elbanna. 2023. The Appropriation of Conversational AI in the Workplace: A Taxonomy of AI Chatbot Users. International Journal...
  • Hemphill, C. T., J. J. Godfrey, and G. R. Doddington. 1990. The ATIS Spoken Language Systems Pilot Corpus. In Proceedings of the Speech and...
  • Hochreiter, S. and J. Schmidhuber. 1997. Long Short-Term Memory. Neural Computation, 9(8):1735–1780.
  • Hubert, L. and P. Arabie. 1985. Comparing partitions. Journal of Classification, 2(1):193–218.
  • Jansen, B. J., D. L. Booth, and A. Spink. 2007. Determining the User Intent of Web Search Engine Queries. In Proceedings of the 16th International...
  • Jolliffe, I. T. 2002. Principal Component Analysis. Springer, 2nd edition.
  • Karpathy, A., J. Johnson, and L. Fei-Fei. 2016. Visualizing and Understanding Recurrent Networks. In Proceedings of the 4th International...
  • Kaufman, L. and P. J. Rousseeuw. 1990. Finding Groups in Data: An Introduction to Cluster Analysis. Wiley-Interscience, 1 edition.
  • Kingma, D. P. and J. Ba. 2015. Adam: A Method for Stochastic Optimization. In Proceedings of the 3rd International Conference on Learning...
  • Larson, S., A. Mahendran, J. J. Peper, C. Clarke, A. Lee, P. Hill, J. K. Kummerfeld, K. Leach, M. A. Laurenzano, L. Tang, and J. Mars. 2019....
  • Levina, E. and P. J. Bickel. 2004. Maximum Likelihood Estimation of Intrinsic Dimension. In Advances in Neural Information Processing Systems...
  • Liu, X., A. Eshghi, P. Swietojanski, and V. Rieser. 2021. Benchmarking Natural Language Understanding Services for building Conversational...
  • Lloyd, S. P. 1982. Least Squares Quantization in PCM. IEEE Transactions on Information Theory, 28(2):129–137.
  • Maheswaranathan, N., A. Williams, M. Golub, S. Ganguli, and D. Sussillo. 2019. Reverse engineering Recurrent Networks for Sentiment Classification...
  • Martelli, M. 1999. Introduction to Discrete Dynamical Systems and Chaos. Wiley- Interscience, 1st edition.
  • Ming, Y., S. Cao, R. Zhang, Z. Li, Y. Chen, Y. Song, and H. Qu. 2017. Understanding Hidden Memories of Recurrent Neural Networks. In Proceedings...
  • Morcos, A. S., D. G. T. Barrett, N. C. Rabinowitz, and M. Botvinick. 2018. On the Importance of Single Directions for Generalization. In Proceedings...
  • Niimi, Y., T. Oku, T. Nishimoto, and M. Araki. 2001. A Rule Based Approach to Extraction of Topics and Dialog Acts in a Spoken Dialog System....
  • Rastogi, A., X. Zang, S. Sunkara, R. Gupta, and P. Khaitan. 2020. Towards Scalable Multi-domain Conversational Agents: The Schema-Guided Dialogue...
  • Ravuri, S. and A. Stolcke. 2015. Recurrent Neural Network and LSTM Models for Lexical Utterance Classification. In Proceedings of the 16th...
  • Rousseeuw, P. J. 1987. Silhouettes: A graphical Aid to the Interpretation and Validation of Cluster Analysis. Journal of Computational and...
  • Sanchez-Karhunen, E., J. F. Quesada-Moreno, and M. A. Gutiérrez-Naranjo. 2024. Interpretation of the Intent Detection Problem as Dynamics...
  • Sanchez-Karhunen, E., J. F. Quesada-Moreno, and M. A. Gutiérrez-Naranjo. 2025. Bias in Intent Detection: A Dynamical Systems Perspective....
  • Sanchez-Karhunen, E., J. F. Quesada-Moreno, and M. A. Gutiérrez-Naranjo. 2026. Interpretability of the Intent Detection Problem: A New Approach....
  • Steinley, D. 2004. Properties of the Hubert-Arable Adjusted Rand Index. Psychological Methods, 9(3):386–396.
  • Strobelt, H., S. Gehrmann, H. Pfister, and A. M. Rush. 2018. LSTMVis: A Tool for Visual Analysis of Hidden State Dynamics in Recurrent Neural...
  • Sussillo, D. and O. Barak. 2013. Opening the Black Box: Low-Dimensional Dynamics in High-Dimensional Recurrent Neural Networks. Neural Computation,...
  • Tenenbaum, J. B., V. de Silva, and J. C. Langford. 2000. A Global Geometric Framework for Nonlinear Dimensionality Reduction. Science, 290(5500):2319–2323.
  • Tomasev, N. and M. Radovanovic. 2016. Clustering Evaluation in High-Dimensional Data. In Unsupervised Learning Algorithms. Springer, pages...
  • Tur, G. and R. De Mori. 2011. Spoken Language Understanding: Systems for Extracting Semantic Information from Speech. John Wiley & Sons.