pith. machine review for the scientific record. sign in

arxiv: 2604.18539 · v1 · submitted 2026-04-20 · 💻 cs.CL · cs.AI

Recognition: unknown

Transition-Matrix Regularization for Next Dialogue Act Prediction in Counselling Conversations

Eric Rudolph , Philipp Steigerwald , Jens Albrecht
Authors on Pith no claims yet

Pith reviewed 2026-05-10 04:02 UTC · model grok-4.3

classification 💻 cs.CL cs.AI
keywords next dialogue act predictiontransition matrix regularizationKL divergencecounselling conversationsdialogue flowdialogue act taxonomysequence predictionnatural language processing
0
0 comments X

The pith

A KL regularization term based on corpus transition patterns improves next dialogue act prediction in counselling conversations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper aims to show that next dialogue act prediction benefits when models are guided by how acts typically follow one another in real counselling data. It does this by adding a regularization penalty that pulls the model's output distribution toward empirical transition probabilities extracted from the training corpus. A sympathetic reader would care because accurate next-act forecasting supports more coherent automated counselling tools and better analysis of human conversations, especially when labelled data is limited and the act set is large. The approach is tested on a fine-grained 60-class German taxonomy with cross-validation and checked for transfer on an external English dataset. Results indicate the regularization adds consistent value on top of various pretrained encoders and helps weaker models most.

Core claim

The paper establishes that a Kullback-Leibler regularization term aligning a model's predicted dialogue-act distribution with a transition matrix derived from corpus statistics raises macro-F1 by 9 to 42 percent relative to unregularized baselines, depending on the encoder, while also increasing measured alignment between predicted and observed dialogue flows. The gains hold across five-fold cross-validation on the German counselling data and appear to transfer when the same regularized models are evaluated on the HOPE dataset. Systematic ablations show the benefit is largest for weaker encoders and remains positive even when stronger pretrained models are used.

What carries the argument

The KL regularization term that penalizes deviations between the model's softmax output over the 60 acts and a fixed transition matrix precomputed from training-corpus act sequences.

If this is right

  • Macro-F1 gains occur consistently across multiple pretrained encoders and model architectures.
  • Weaker baseline models receive the largest relative improvement from the regularization.
  • Dialogue-flow alignment metrics improve alongside classification accuracy.
  • The same regularized models show positive transfer when tested on an independent counselling dataset in another language.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method could be tested on non-counselling dialogue domains to determine whether the transition priors must be domain-specific or can be drawn from broader conversation data.
  • Because the regularization is lightweight, it offers a route to improve fine-grained dialogue tasks without requiring larger labelled datasets or more expensive model scaling.
  • If the transition matrix is recomputed from each new domain, the approach might serve as a general way to inject discourse structure into any sequence prediction model.

Load-bearing premise

The corpus-derived transition patterns supply useful, unbiased priors that remain valid outside the original training data and counselling domain.

What would settle it

No performance gain or outright degradation when the same regularization is applied to a dialogue corpus whose act-transition statistics differ substantially from those of the counselling training set.

Figures

Figures reproduced from arXiv: 2604.18539 by Eric Rudolph, Jens Albrecht, Philipp Steigerwald.

Figure 1
Figure 1. Figure 1: Comparison of standard NDAP training (a) [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Model architectures for NDAP: BERT + at [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Example 6x6 subset (Fold 0) of the empirical [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Effect of transition loss weight (λtm) on 60-category classification. Left four panels show how Macro-F1, Top-3 Accuracy, Cum70, and JS Divergence change with increasing λtm across encoders. Right panel compares relative macro-F1 gains from TM regularization by encoder. λtm Macro-F1 W-F1 Top-3 Cum70 JS 0.0 .309 ±.011 .495 ±.008 .789 ±.011 .913 ±.019 .299 ±.017 0.2 .314 ±.013 .496 ±.010 .783 ±.014 .915 ±.01… view at source ↗
read the original abstract

This paper studies how empirical dialogue-flow statistics can be incorporated into Next Dialogue Act Prediction (NDAP). A KL regularization term is proposed that aligns predicted act distributions with corpus-derived transition patterns. Evaluated on a 60-class German counselling taxonomy using 5-fold cross-validation, this improves macro-F1 by 9--42% relative depending on encoder and substantially improves dialogue-flow alignment. Cross-dataset validation on HOPE suggests that improvements transfer across languages and counselling domains. In systematic ablations across pretrained encoders and architectures, the findings indicate that transition regularization provides consistent gains and disproportionately benefits weaker baseline models. The results suggest that lightweight discourse-flow priors complement pretrained encoders, especially in fine-grained, data-sparse dialogue tasks.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a KL-divergence regularization term that aligns a model's predicted dialogue-act distribution with empirical transition probabilities derived from a counselling corpus. It evaluates this on a 60-class German counselling taxonomy using 5-fold cross-validation, reporting relative macro-F1 gains of 9-42% over baselines depending on the encoder, improved dialogue-flow alignment, and positive transfer to the HOPE dataset.

Significance. If the gains prove robust under leakage-free cross-validation and include statistical testing, the result would show that lightweight, corpus-derived discourse priors can usefully complement pretrained encoders in fine-grained, data-sparse dialogue tasks. The systematic encoder ablations and cross-dataset experiment are strengths that would support broader applicability to counselling and similar domains.

major comments (2)
  1. [Experimental setup and 5-fold cross-validation description] The experimental setup does not state whether the transition matrix is recomputed exclusively on each training fold or derived once from the full corpus. If the latter, every regularization term during training and validation incorporates test-fold transition statistics, violating the independence assumption required for the claimed prior and rendering the 9-42% macro-F1 improvements an upper bound that may not hold under proper per-fold estimation.
  2. [Results and ablations] Results section reports only relative macro-F1 gains without absolute scores, exact baseline implementations, hyperparameter search details, or statistical significance tests (e.g., paired t-test across folds). This leaves the practical magnitude and reliability of the improvements difficult to assess.
minor comments (2)
  1. [Method] Clarify the exact form of the KL term (including the regularization weight schedule) and how the 60-class taxonomy is mapped to transition counts.
  2. [Results] Add a table of absolute F1 scores per encoder and fold to complement the relative gains.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thorough review and constructive feedback. The two major comments identify important issues of experimental rigor and reporting clarity. We address each below and will revise the manuscript to incorporate the necessary clarifications and additional results.

read point-by-point responses

  1. Referee: [Experimental setup and 5-fold cross-validation description] The experimental setup does not state whether the transition matrix is recomputed exclusively on each training fold or derived once from the full corpus. If the latter, every regularization term during training and validation incorporates test-fold transition statistics, violating the independence assumption required for the claimed prior and rendering the 9-42% macro-F1 improvements an upper bound that may not hold under proper per-fold estimation.

    Authors: We agree that the manuscript does not explicitly describe the computation of the transition matrix within the 5-fold cross-validation protocol. To ensure a leakage-free setup, the transition matrix must be derived exclusively from the training folds. We will revise the Experimental Setup section to state this procedure clearly and will re-run the experiments (if the original runs used the full corpus) to confirm that the reported relative gains hold under per-fold estimation. The revised results will be presented with the same encoder ablations. revision: yes

  2. Referee: [Results and ablations] Results section reports only relative macro-F1 gains without absolute scores, exact baseline implementations, hyperparameter search details, or statistical significance tests (e.g., paired t-test across folds). This leaves the practical magnitude and reliability of the improvements difficult to assess.

    Authors: We accept that absolute performance numbers, precise baseline descriptions, hyperparameter details, and statistical tests are needed for full assessment. The revised manuscript will report absolute macro-F1 scores for all models, provide exact specifications of the baseline encoders and training procedures, document the hyperparameter search ranges and selection criteria, and include paired t-test results across the five folds to establish statistical significance of the observed gains. revision: yes

Circularity Check

0 steps flagged

No circularity: transition matrix is an independent empirical prior

full rationale

The paper proposes a KL regularization term that aligns predicted dialogue-act distributions to transition patterns precomputed from the corpus. This prior is derived externally from aggregate statistics and does not reduce to any model parameter, fitted quantity, or self-referential definition within the training objective. No equations or steps equate the regularization target to the model's own outputs by construction, nor do they rely on self-citation chains or imported uniqueness results. The 5-fold CV evaluation and cross-dataset validation on HOPE are presented as external checks; the method remains self-contained against these benchmarks without any load-bearing step that collapses the claimed improvement into a tautology.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on empirical corpus statistics as priors and the validity of the dialogue taxonomy; the regularization weight is a tunable hyperparameter.

free parameters (1)
  • KL regularization weight
    The strength of the added KL term is a hyperparameter that must be selected or tuned on validation data.
axioms (1)
  • domain assumption The 60-class German counselling taxonomy provides a meaningful and consistent categorization of dialogue acts.
    All evaluations and transition matrix construction depend on this taxonomy being appropriate.

pith-pipeline@v0.9.0 · 5412 in / 1308 out tokens · 38020 ms · 2026-05-10T04:02:16.365283+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

112 extracted references · 65 canonical work pages · 2 internal anchors

  1. [1]

    Proceedings of the AAAI Conference on Artificial Intelligence , author =

    Guiding. Proceedings of the AAAI Conference on Artificial Intelligence , author =. 2020 , pages =. doi:10.1609/aaai.v34i05.6259 , abstract =

    work page doi:10.1609/aaai.v34i05.6259 2020

  2. [2]

    Speaker and time-aware joint contextual learning for dialogue-act classification in counselling conversations , isbn =

    Malhotra, Ganeshan and Waheed, Abdul and Srivastava, Aseem and Akhtar, Md Shad and Chakraborty, Tanmoy , year =. Speaker and time-aware joint contextual learning for dialogue-act classification in counselling conversations , isbn =. Proceedings of the fifteenth. doi:10.1145/3488560.3498509 , abstract =

    work page doi:10.1145/3488560.3498509

  3. [3]

    and Guu, Kelvin and Yu, Adams Wei and Lester, Brian and Du, Nan and Dai, Andrew M

    Wei, Jason and Bosma, Maarten and Zhao, Vincent Y. and Guu, Kelvin and Yu, Adams Wei and Lester, Brian and Du, Nan and Dai, Andrew M. and Le, Quoc V. , year =. Finetuned language models are zero-shot learners , url =. International conference on learning representations (

  4. [4]

    Comparing human roleplayers and

    Rudolph, Eric and Steigerwald, Philipp and Albrecht, Jens , editor =. Comparing human roleplayers and. Proceedings of the 18th international conference on educational data mining , publisher =. 2025 , pages =. doi:10.5281/zenodo.15870260 , abstract =

    work page doi:10.5281/zenodo.15870260 2025

  5. [5]

    Language models are few-shot learners , volume =

    Brown, Tom and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie and Kaplan, Jared D and Dhariwal, Prafulla and Neelakantan, Arvind and Shyam, Pranav and Sastry, Girish and Askell, Amanda and Agarwal, Sandhini and Herbert-Voss, Ariel and Krueger, Gretchen and Henighan, Tom and Child, Rewon and Ramesh, Aditya and Ziegler, Daniel and Wu, Jeffrey and Winte...

    2020

  6. [6]

    Brown, Tom B. and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie and Kaplan, Jared and Dhariwal, Prafulla and Neelakantan, Arvind and Shyam, Pranav and Sastry, Girish and Askell, Amanda and Agarwal, Sandhini and Herbert-Voss, Ariel and Krueger, Gretchen and Henighan, Tom and Child, Rewon and Ramesh, Aditya and Ziegler, Daniel M. and Wu, Jeffrey and W...

    work page internal anchor Pith review doi:10.48550/arxiv.2005.14165 2005

  7. [7]

    InThe 41st International ACM SIGIR Conference on Research & Development in Information Retrieval(Ann Arbor, MI, USA)(SIGIR ’18)

    Chen, Zheqian and Yang, Rongqin and Zhao, Zhou and Cai, Deng and He, Xiaofei , year =. Dialogue act recognition via. The 41st international. doi:10.1145/3209978.3209997 , abstract =

    work page doi:10.1145/3209978.3209997

  8. [8]

    Training language models to follow instructions with human feedback , volume =

    Ouyang, Long and Wu, Jeffrey and Jiang, Xu and Almeida, Diogo and Wainwright, Carroll and Mishkin, Pamela and Zhang, Chong and Agarwal, Sandhini and Slama, Katarina and Ray, Alex and Schulman, John and Hilton, Jacob and Kelton, Fraser and Miller, Luke and Simens, Maddie and Askell, Amanda and Welinder, Peter and Christiano, Paul F and Leike, Jan and Lowe,...

    2022

  9. [9]

    Demasi, Orianna and Li, Yu and Yu, Zhou , editor =. A. Findings of the. 2020 , pages =. doi:10.18653/v1/2020.findings-emnlp.324 , abstract =

    work page doi:10.18653/v1/2020.findings-emnlp.324 2020

  10. [10]

    Journal of the Royal Statistical Society

    Controlling the. Journal of the Royal Statistical Society. Series B (Methodological) , author =. 1995 , note =

    1995

  11. [11]

    Computational Linguistics , author =

    Dialogue act modeling for automatic tagging and recognition of conversational speech , volume =. Computational Linguistics , author =. 2000 , pages =

    2000

  12. [12]

    Language and Speech , author =

    Can prosody aid the automatic classification of dialog acts in conversational speech? , volume =. Language and Speech , author =. 1998 , pages =

    1998

  13. [13]

    Switchboard

    Jurafsky, Daniel and Shriberg, Elizabeth and Biasca, Debra , year =. Switchboard

  14. [14]

    Meta-control of

    Shukuri, Kotaro and Ishigaki, Ryoma and Suzuki, Jundai and Naganuma, Tsubasa and Fujimoto, Takuma and Kawakubo, Daisuke and Shuzo, Masaki and Maeda, Eisaku , month = dec, year =. Meta-control of. doi:10.48550/arXiv.2312.13715 , abstract =

    work page doi:10.48550/arxiv.2312.13715

  15. [15]

    Wagner, Nicolas and Ultes, Stefan , editor =. On the. Proceedings of the 25th. 2024 , pages =. doi:10.18653/v1/2024.sigdial-1.19 , abstract =

    work page doi:10.18653/v1/2024.sigdial-1.19 2024

  16. [16]

    doi:10.48550/arXiv.2512.09804 , abstract =

    Albrecht, Jens and Lehmann, Robert and Poltermann, Aleksandra and Rudolph, Eric and Steigerwald, Philipp and Stieler, Mara , month = dec, year =. doi:10.48550/arXiv.2512.09804 , abstract =

    work page doi:10.48550/arxiv.2512.09804

  17. [17]

    Journal of Machine Learning Research , author =

    Posterior regularization for structured latent variable models , volume =. Journal of Machine Learning Research , author =. 2010 , pages =

    2010

  18. [18]

    Rethinking the Inception Architecture for Computer Vision

    Szegedy, Christian and Vanhoucke, Vincent and Ioffe, Sergey and Shlens, Jonathon and Wojna, Zbigniew , month = dec, year =. Rethinking the. doi:10.48550/arXiv.1512.00567 , abstract =

    work page Pith review doi:10.48550/arxiv.1512.00567

  19. [19]

    Speaker-change

    Shang, Guokan and Tixier, Antoine and Vazirgiannis, Michalis and Lorré, Jean-Pierre , editor =. Speaker-change. Proceedings of the 28th. 2020 , pages =. doi:10.18653/v1/2020.coling-main.40 , abstract =

    work page doi:10.18653/v1/2020.coling-main.40 2020

  20. [20]

    Precognition in

    Su, Nan and Zhang, Yuchi and Liu, Chao and Du, Bingzhu and Wang, Yongliang , month = mar, year =. Precognition in. doi:10.48550/arXiv.2203.03244 , abstract =

    work page doi:10.48550/arxiv.2203.03244

  21. [21]

    Explicit

    Jin, Xisen and Lei, Wenqiang and Ren, Zhaochun and Chen, Hongshen and Liang, Shangsong and Zhao, Yihong and Yin, Dawei , month = oct, year =. Explicit. Proceedings of the 27th. doi:10.1145/3269206.3271683 , abstract =

    work page doi:10.1145/3269206.3271683

  22. [22]

    Regularizing

    Feng, Shaoxiong and Ren, Xuancheng and Chen, Hongshen and Sun, Bin and Li, Kan and Sun, Xu , editor =. Regularizing. Proceedings of the 2020. 2020 , pages =. doi:10.18653/v1/2020.emnlp-main.534 , abstract =

    work page doi:10.18653/v1/2020.emnlp-main.534 2020

  23. [23]

    Dialogue

    Kumar, Harshit and Agarwal, Arvind and Dasgupta, Riddhiman and Joshi, Sachindra and Kumar, Arun , month = sep, year =. Dialogue. doi:10.48550/arXiv.1709.04250 , abstract =

    work page doi:10.48550/arxiv.1709.04250

  24. [24]

    Dialogue-

    Tanaka, Koji and Takayama, Junya and Arase, Yuki , editor =. Dialogue-. Proceedings of the 57th. 2019 , pages =. doi:10.18653/v1/P19-2027 , abstract =

    work page doi:10.18653/v1/p19-2027 2019

  25. [25]

    Findings of the

    Zhang, Jusheng and Fan, Yijia and Cai, Kaitong and Sun, Xiaofei and Wang, Keze , editor =. Findings of the. 2025 , pages =. doi:10.18653/v1/2025.findings-emnlp.335 , abstract =

    work page doi:10.18653/v1/2025.findings-emnlp.335 2025

  26. [26]

    Yu, Xiao and Chen, Maximillian and Yu, Zhou , editor =. Prompt-. Proceedings of the 2023. 2023 , pages =. doi:10.18653/v1/2023.emnlp-main.439 , abstract =

    work page doi:10.18653/v1/2023.emnlp-main.439 2023

  27. [27]

    Response-act

    Srivastava, Aseem and Pandey, Ishan and Akhtar, Md Shad and Chakraborty, Tanmoy , month = apr, year =. Response-act. Proceedings of the. doi:10.1145/3543507.3583380 , abstract =

    work page doi:10.1145/3543507.3583380

  28. [28]

    Ganesh, Ananya and Palmer, Martha and Kann, Katharina , editor =. What. Findings of the. 2021 , pages =. doi:10.18653/v1/2021.findings-acl.418 , urldate =

    work page doi:10.18653/v1/2021.findings-acl.418 2021

  29. [29]

    Yu, Xiao and Chen, Maximillian and Yu, Zhou , month = oct, year =. Prompt-. doi:10.48550/arXiv.2305.13660 , abstract =

    work page doi:10.48550/arxiv.2305.13660

  30. [30]

    Detecting

    Shreevastava, Sagarika and Foltz, Peter , editor =. Detecting. Proceedings of the. 2021 , pages =. doi:10.18653/v1/2021.clpsych-1.17 , abstract =

    work page doi:10.18653/v1/2021.clpsych-1.17 2021

  31. [31]

    Harold Ngabo-Woods, Larisa Dunai, and Isabel Seguí Verdú

    The. Journal of Substance Abuse Treatment , author =. 2016 , pages =. doi:10.1016/j.jsat.2016.01.001 , language =

    work page doi:10.1016/j.jsat.2016.01.001 2016

  32. [33]

    and Moyers, Theresa B

    Miller, William R. and Moyers, Theresa B. and Ernst, Denise and Amrhein, Paul , year =. Manual for the

  33. [34]

    Speech Communication , author =

    First steps towards statistical modeling of dialogue to predict the speech act type of the next utterance , volume =. Speech Communication , author =. 1994 , keywords =. doi:10.1016/0167-6393(94)90071-X , abstract =

    work page doi:10.1016/0167-6393(94)90071-x 1994

  34. [35]

    Ji, Yangfeng and Haffari, Gholamreza and Eisenstein, Jacob , month = apr, year =. A. doi:10.48550/arXiv.1603.01913 , abstract =

    work page doi:10.48550/arxiv.1603.01913

  35. [36]

    and Heidarysafa, Mojtaba and Meimandi, Kiana Jafari and Gerber, Matthew S

    Kowsari, Kamran and Brown, Donald E. and Heidarysafa, Mojtaba and Meimandi, Kiana Jafari and Gerber, Matthew S. and Barnes, Laura E. , month = dec, year =. 2017 16th. doi:10.1109/ICMLA.2017.0-134 , abstract =

    work page doi:10.1109/icmla.2017.0-134 2017

  36. [37]

    Wu, Zixiu and Helaoui, Rim and Reforgiato Recupero, Diego and Riboni, Daniele , month = sep, year =. Towards. Interspeech 2022 , publisher =. doi:10.21437/Interspeech.2022-506 , language =

    work page doi:10.21437/interspeech.2022-506 2022

  37. [38]

    Wu, Zixiu and Balloccu, Simone and Kumar, Vivek and Helaoui, Rim and Reiter, Ehud and Reforgiato Recupero, Diego and Riboni, Daniele , month = may, year =. Anno-. doi:10.1109/ICASSP43922.2022.9746035 , abstract =

    work page doi:10.1109/icassp43922.2022.9746035 2022

  38. [39]

    Learning

    Zhao, Tiancheng and Zhao, Ran and Eskenazi, Maxine , editor =. Learning. Proceedings of the 55th. 2017 , pages =. doi:10.18653/v1/P17-1061 , abstract =

    work page doi:10.18653/v1/p17-1061 2017

  39. [40]

    Ultes, Stefan and Rojas-Barahona, Lina M. and Su, Pei-Hao and Vandyke, David and Kim, Dongho and Casanueva, Iñigo and Budzianowski, Paweł and Mrkšić, Nikola and Wen, Tsung-Hsien and Gašić, Milica and Young, Steve , editor =. Proceedings of. 2017 , pages =

    2017

  40. [41]

    Ravuru, Lohith and Ryu, Seonghan and Choi, Hyungtak and Yang, Haehun and Ko, Hyeonmok , editor =. Multi-. Findings of the. 2022 , pages =. doi:10.18653/v1/2022.findings-aacl.16 , abstract =

    work page doi:10.18653/v1/2022.findings-aacl.16 2022

  41. [42]

    Joint segmentation and classification of dialog acts using conditional random fields , url =

    Zimmermann, Matthias , month = sep, year =. Joint segmentation and classification of dialog acts using conditional random fields , url =. Interspeech 2009 , publisher =. doi:10.21437/Interspeech.2009-262 , language =

    work page doi:10.21437/interspeech.2009-262 2009

  42. [43]

    Patient Education and Counseling , author =

    Automated interventions for multiple health behaviors using conversational agents , volume =. Patient Education and Counseling , author =. 2013 , pages =. doi:10.1016/j.pec.2013.05.011 , language =

    work page doi:10.1016/j.pec.2013.05.011 2013

  43. [44]

    Transactions of the Association for Computational Linguistics , author =

    Large-scale. Transactions of the Association for Computational Linguistics , author =. 2016 , pages =. doi:10.1162/tacl_a_00111 , abstract =

    work page doi:10.1162/tacl_a_00111 2016

  44. [45]

    Codegeex: A pre-trained model for code generation with multilingual benchmarking on humaneval-x

    Chen, Sirui and Wang, Yuan and Wen, Zijing and Li, Zhiyu and Zhang, Changshuo and Zhang, Xiao and Lin, Quan and Zhu, Cheng and Xu, Jun , month = aug, year =. Controllable. Proceedings of the 29th. doi:10.1145/3580305.3599796 , abstract =

    work page doi:10.1145/3580305.3599796

  45. [46]

    Finetuned Language Models Are Zero-Shot Learners

    Wei, Jason and Bosma, Maarten and Zhao, Vincent Y. and Guu, Kelvin and Yu, Adams Wei and Lester, Brian and Du, Nan and Dai, Andrew M. and Le, Quoc V. , month = feb, year =. Finetuned. doi:10.48550/arXiv.2109.01652 , abstract =

    work page internal anchor Pith review doi:10.48550/arxiv.2109.01652

  46. [47]

    Computational Linguistics , author =

    Dialogue act modeling for automatic tagging and recognition of conversational speech , volume =. Computational Linguistics , author =. 2000 , note =

    2000

  47. [48]

    and Engel, R

    Reithinger, N. and Engel, R. and Kipp, M. and Klesen, M. , month = oct, year =. Predicting dialogue acts for a speech-to-speech translation system , volume =. Proceeding of. doi:10.1109/ICSLP.1996.607446 , abstract =

    work page doi:10.1109/icslp.1996.607446 1996

  48. [49]

    Controlling dialogue strategy in llms for counseling conversations , booktitle =

    Srivastava, Rhea and Madaan, Aman and Aggarwal, Arushi and Singh, Manish , year =. Controlling dialogue strategy in llms for counseling conversations , booktitle =

  49. [50]

    A latent variable recurrent neural network for discourse relation language models , booktitle =

    Ji, Yangfeng and Haffari, Gholamreza and Eisenstein, Jacob , year =. A latent variable recurrent neural network for discourse relation language models , booktitle =

  50. [51]

    Meta dialogue policy learning , journal =

    Xu, Yumo and Zhu, Chenguang and Peng, Baolin and Zeng, Michael , year =. Meta dialogue policy learning , journal =

  51. [52]

    Roncal, Adolfo and Iriondo, Ignasi , year =

  52. [53]

    Dialogue act prediction using stochastic context-free grammar induction , booktitle =

    Geertzen, Jeroen , year =. Dialogue act prediction using stochastic context-free grammar induction , booktitle =

  53. [54]

    2022 , pages =

    Proceedings of the 60th annual meeting of the association for computational linguistics , author =. 2022 , pages =

    2022

  54. [55]

    Heterogeneous graph-based strategy prediction in emotional support dialogue , booktitle =

    Xiang, Rui and Wang, Shu , year =. Heterogeneous graph-based strategy prediction in emotional support dialogue , booktitle =

  55. [56]

    Dialogue-act prediction of future responses based on conversation history , booktitle =

    Tanaka, Koji and Takayama, Junya and Arase, Yuki , year =. Dialogue-act prediction of future responses based on conversation history , booktitle =

  56. [57]

    Contextual

    Kobayashi, Sosuke , editor =. Contextual. Proceedings of the 2018. 2018 , pages =. doi:10.18653/v1/N18-2072 , abstract =

    work page doi:10.18653/v1/n18-2072 2018

  57. [58]

    arXiv preprint arXiv:1511.06709 , year=

    Sennrich, Rico and Haddow, Barry and Birch, Alexandra , month = jun, year =. Improving. doi:10.48550/arXiv.1511.06709 , abstract =

    work page doi:10.48550/arxiv.1511.06709

  58. [59]

    doi:10.48550/arXiv.2404.00930 , abstract =

    Han, Ji-Eun and Koh, Jun-Seok and Seo, Hyeon-Tae and Chang, Du-Seong and Sohn, Kyung-Ah , month = apr, year =. doi:10.48550/arXiv.2404.00930 , abstract =

    work page doi:10.48550/arxiv.2404.00930

  59. [60]

    2025 , note =

    Proceedings of the AAAI Conference on Artificial Intelligence , author =. 2025 , note =. doi:10.1609/aaai.v39i2.32116 , abstract =

    work page doi:10.1609/aaai.v39i2.32116 2025

  60. [61]

    Charles J Wallace, Connie J Nelson, Robert Paul Liber- man, Robert A Aitchison, David Lukoff, John P El- der, and Chris Ferris

    Wei, Jason and Zou, Kai , month = aug, year =. doi:10.48550/arXiv.1901.11196 , abstract =

    work page doi:10.48550/arxiv.1901.11196 1901

  61. [62]

    Proceedings of the 2023

    Sohn, Sungryull and Lyu, Yiwei and Liu, Anthony and Logeswaran, Lajanugen and Kim, Dong-Ki and Shim, Dongsub and Lee, Honglak , editor =. Proceedings of the 2023. 2023 , pages =. doi:10.18653/v1/2023.emnlp-main.204 , abstract =

    work page doi:10.18653/v1/2023.emnlp-main.204 2023

  62. [63]

    doi:10.48550/arXiv.2308.00878 , abstract =

    Wu, Qingyang and Gung, James and Shu, Raphael and Zhang, Yi , month = aug, year =. doi:10.48550/arXiv.2308.00878 , abstract =

    work page doi:10.48550/arxiv.2308.00878

  63. [64]

    Dialogue

    Raheja, Vipul and Tetreault, Joel , editor =. Dialogue. Proceedings of the 2019. 2019 , pages =. doi:10.18653/v1/N19-1373 , abstract =

    work page doi:10.18653/v1/n19-1373 2019

  64. [65]

    Anwar, Usman and Ansari, Deevansh , year =. Dual-. Proceedings of the

  65. [66]

    doi:10.18653/v1/2025.findings-acl.1052 , booktitle =

    Zhangwenbo, Zhangwenbo and Yuhan, Wang , year =. doi:10.18653/v1/2025.findings-acl.1052 , booktitle =

    work page doi:10.18653/v1/2025.findings-acl.1052 2025

  66. [67]

    Wu, Shenghan and Zhu, Yimo and Hsu, Wynne and Lee, Mong-Li and Deng, Yang , month = feb, year =. From

  67. [68]

    Jiang, Jiyue and Chen, Liheng and Wang, Sheng and Kong, Lingpeng and Li, Yu and Wu, Chuan , month = jun, year =. Data

  68. [69]

    Qi, Zhiyang and Inaba, Michimasa , month = aug, year =. Data

  69. [70]

    Cheng, Jiale and Sabour, Sahand and Sun, Hao and Chen, Zhuang and Huang, Minlie , month = dec, year =

  70. [71]

    doi:10.48550/arXiv.2202.13047 , abstract =

    Zheng, Chujie and Sabour, Sahand and Wen, Jiaxin and Zhang, Zheng and Huang, Minlie , month = may, year =. doi:10.48550/arXiv.2202.13047 , abstract =

    work page doi:10.48550/arxiv.2202.13047

  71. [72]

    Enhancing

    Kim, Jun-Woo and Han, Ji-Eun and Koh, Jun-Seok and Seo, Hyeon-Tae and Chang, Du-Seong , month = jun, year =. Enhancing. doi:10.48550/arXiv.2406.08718 , abstract =

    work page doi:10.48550/arxiv.2406.08718

  72. [73]

    Dual-task dialogue understanding:

    Anwar, Sibgha and Wiratunga, Nirmalie and Snaith, Mark , year =. Dual-task dialogue understanding:. Proceedings of the

  73. [74]

    Williamson, Minjeong Kim, and Apu Kapadia

    Witzig, Philine and Constantin, Rares and Kovacevic, Nikola and Wampfler, Rafael , month = jul, year =. Multimodal. doi:10.1145/3640794.3665541 , language =

    work page doi:10.1145/3640794.3665541

  74. [75]

    Wang, Ke and Zhu, Jiahui and Ren, Minjie and Liu, Zeming and Li, Shiwei and Zhang, Zongye and Zhang, Chenkai and Wu, Xiaoyu and Zhan, Qiqi and Liu, Qingjie and Wang, Yunhong , month = oct, year =. A. doi:10.48550/arXiv.2410.12896 , abstract =

    work page doi:10.48550/arxiv.2410.12896

  75. [76]

    doi:10.48550/arXiv.2409.01763 , abstract =

    Ta, Hoang-Thang and Thai, Duy-Quy and Rahman, Abu Bakar Siddiqur and Sidorov, Grigori and Gelbukh, Alexander , month = feb, year =. doi:10.48550/arXiv.2409.01763 , abstract =

    work page doi:10.48550/arxiv.2409.01763

  76. [77]

    Almasian, Satya and Bruseva, Milena and Gertz, Michael , month = jul, year =. Numbers. doi:10.48550/arXiv.2407.10283 , abstract =

    work page doi:10.48550/arxiv.2407.10283

  77. [78]

    doi:10.48550/arXiv.2312.04668 , abstract =

    Sohn, Sungryull and Lyu, Yiwei and Liu, Anthony and Logeswaran, Lajanugen and Kim, Dong-Ki and Shim, Dongsub and Lee, Honglak , month = dec, year =. doi:10.48550/arXiv.2312.04668 , abstract =

    work page doi:10.48550/arxiv.2312.04668

  78. [79]

    Interior derivative estimates and

    Dai, Limei and Bao, Jiguang and Wang, Bo , month = may, year =. Interior derivative estimates and. doi:10.48550/arXiv.2305.17831 , abstract =

    work page doi:10.48550/arxiv.2305.17831

  79. [80]

    , month = feb, year =

    Su, Jinyan and Yu, Peilin and Zhang, Jieyu and Bach, Stephen H. , month = feb, year =. Leveraging. doi:10.48550/arXiv.2402.01867 , abstract =

    work page doi:10.48550/arxiv.2402.01867

  80. [81]

    Physical Review Letters , author =

    Neural-network. Physical Review Letters , author =. 2024 , note =. doi:10.1103/PhysRevLett.133.076401 , abstract =

    work page doi:10.1103/physrevlett.133.076401 2024

Showing first 80 references.