pith. sign in

arxiv: 2606.07894 · v1 · pith:E2CG362Ynew · submitted 2026-06-05 · 💻 cs.SE

DD-GEPA: Prompt Optimization for Dialogue Disentanglement Focusing on Task Instruction and Utterance Representation

Naoki Takada , Tatsunori Mori This is my paper

Pith reviewed 2026-06-27 20:55 UTC · model grok-4.3

classification 💻 cs.SE
keywords prompt optimizationdialogue disentanglementlarge language modelsGEPAmulti-party chattask instructionutterance representation
0
0 comments X

The pith

Decomposing prompts into task instruction and utterance representation then optimizing them with GEPA raises LLM accuracy on dialogue disentanglement.

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

The paper establishes that large language models can separate interleaved multi-party conversations into coherent dialogues more accurately when their prompts are automatically refined. It splits each prompt into three parts—task instruction, utterance representation, and output instruction—then applies an optimization process to improve the first two parts. Experiments on benchmark datasets demonstrate that the resulting prompts outperform both the starting versions and those written by hand. A sympathetic reader would care because this reduces the need for expert manual tuning when using LLMs for tasks that involve overlapping speaker turns and topics.

Core claim

The central claim is that an automatic prompt optimization method applied after decomposing prompts into task instruction, utterance representation, and output instruction components produces prompts that improve dialogue disentanglement accuracy over the original prompts and can surpass hand-crafted prompts on benchmark datasets.

What carries the argument

GEPA, an optimization method for compound AI systems, applied after decomposing the prompt into task instruction, utterance representation, and output instruction components.

If this is right

  • Optimized prompts achieve higher dialogue disentanglement accuracy than the original prompts on benchmark datasets.
  • The optimized prompts can exceed the performance of hand-crafted prompts.
  • The gains come from targeted optimization of the task instruction and utterance representation components.
  • LLM-based dialogue disentanglement systems become more accurate at separating entangled utterance sequences without model changes.

Where Pith is reading between the lines

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

  • The same decomposition-plus-optimization pattern could be tested on other structured output tasks such as coreference resolution in multi-speaker transcripts.
  • Real-world chat logs with higher noise levels than the benchmarks might show different improvement margins.
  • Replacing GEPA with alternative search methods while keeping the three-component split could isolate whether the decomposition itself drives most of the benefit.

Load-bearing premise

Decomposing the prompt into task instruction, utterance representation, and output instruction components is sufficient to allow GEPA to produce effective optimizations for the dialogue disentanglement task.

What would settle it

Evaluating the GEPA-optimized prompts on the same benchmark datasets and observing no accuracy gain or a drop relative to the original prompts would falsify the central claim.

Figures

Figures reproduced from arXiv: 2606.07894 by Naoki Takada, Tatsunori Mori.

Figure 1
Figure 1. Figure 1: An example of dialogue disentanglement, where the resulting dialogues are shown by color. In multi-party chat, such as Slack and Inter￾net Relay Chat(IRC), multiple dialogues are entan￾gled. Because many users send messages simulta￾neously, multiple dialogues can proceed in paral￾lel and become entangled. Such entanglement can also lead to failures in automatic analysis, because utterances from unrelated d… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of DD-GEPA for automatic optimization of dialogue disentanglement prompts. A program is [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Optimization trajectory of generated pro [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
read the original abstract

Multi-party chat often contains interleaved dialogues because multiple participants can discuss different topics at the same time. Dialogue disentanglement addresses this problem by separating an entangled utterance sequence into coherent dialogues. While large language models (LLMs) are promising for this task, they still struggle with dialogue disentanglement and achieve low accuracy. This paper proposes an automatic prompt optimization for LLM based dialogue disentanglement. We decompose the prompt into three components: task instruction, utterance representation, and output instruction, and optimize them using GEPA, an optimization method for compound AI systems. Experiments on benchmark datasets show that the optimized prompts improve dialogue disentanglement accuracy over the original prompts and can surpass hand crafted prompts.

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

0 major / 3 minor

Summary. The paper proposes DD-GEPA, an automatic prompt optimization approach for LLM-based dialogue disentanglement. The prompt is decomposed into three components (task instruction, utterance representation, and output instruction) that are optimized via the GEPA method for compound AI systems. Experiments on benchmark datasets are reported to show accuracy gains over both the original prompts and hand-crafted prompts.

Significance. If the empirical results hold with appropriate controls, the work demonstrates a practical application of structured prompt decomposition and external optimization to a challenging multi-party dialogue task where standard LLM prompting underperforms. It adds to the literature on automatic prompt engineering by targeting a specific NLP problem without model fine-tuning.

minor comments (3)
  1. [Abstract] Abstract: The claim of accuracy improvements is stated without any numerical results, baselines, dataset names, or statistical details; these must be supplied in the abstract or early results section for the central empirical claim to be evaluable.
  2. The manuscript should include an explicit description of the GEPA optimization procedure, the search space over the three prompt components, and any hyperparameters or stopping criteria used.
  3. Results section: Tables or figures reporting accuracy should include the original prompt, hand-crafted baselines, and the DD-GEPA variant together with error bars or multiple runs if stochasticity is present.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, assessment of significance, and recommendation for minor revision. No specific major comments appear in the provided report.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper describes an empirical prompt-optimization procedure that decomposes prompts into task instruction, utterance representation, and output instruction components, then applies the external GEPA optimizer. The central claim is an experimental accuracy improvement on benchmark datasets. No equations, fitted parameters, self-citations, or derivations are present that reduce the reported result to the inputs by construction. The optimization step and accuracy metric are independent of the claim itself.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No free parameters, axioms, or invented entities are described in the abstract.

pith-pipeline@v0.9.1-grok · 5641 in / 952 out tokens · 23069 ms · 2026-06-27T20:55:22.522167+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

138 extracted references · 20 canonical work pages · 1 internal anchor

  1. [1]

    Title of The Article

    FirstName LastName. Title of The Article. Journal of Natural Language Processing

  2. [2]

    Title of The Book

    FirstNameA LastNameA and FirstNameB LastNameB and FirstNameC LastNameC and FirstNameD LastNameD. Title of The Book

  3. [3]

    著者氏名1 and 著者氏名2 and 著者氏名3. 論文タイトル. プロシーディングスの名前

  4. [4]

    著者氏名1 and 著者氏名2 and 著者氏名3 and 著者氏名4. 技報タイトル

  5. [5]

    ホームページタイトル

    著者氏名. ホームページタイトル. 2017

    2017

  6. [6]

    （2020-11閲覧）

    W3C日本語組版タスクフォース．日本語組版の要件（日本語版）. （2020-11閲覧）

    2020

  7. [7]

    Abril and Robert Plant

    Patricia S. Abril and Robert Plant. The patent holder's dilemma: Buy, sell, or troll?. Communications of the ACM. 2007. doi:10.1145/1188913.1188915

    work page doi:10.1145/1188913.1188915 2007

  8. [8]

    Deciding equivalances among conjunctive aggregate queries

    Sarah Cohen and Werner Nutt and Yehoshua Sagic. Deciding equivalances among conjunctive aggregate queries. 2007. doi:10.1145/1219092.1219093

    work page doi:10.1145/1219092.1219093 2007

  9. [9]

    Special issue: Digital Libraries. 1996

    1996

  10. [10]

    Understanding Policy-Based Networking

    David Kosiur. Understanding Policy-Based Networking. 2001

    2001

  11. [13]

    The title of book two. 2008. doi:10.1007/3-540-09237-4

    work page doi:10.1007/3-540-09237-4 2008

  12. [14]

    Asad Z. Spector. Achieving application requirements. Distributed Systems. 1990. doi:10.1145/90417.90738

    work page doi:10.1145/90417.90738 1990

  13. [15]

    Douglass and David Harel and Mark B

    Bruce P. Douglass and David Harel and Mark B. Trakhtenbrot. Statecarts in use: structured analysis and object-orientation. Lectures on Embedded Systems. 1998. doi:10.1007/3-540-65193-4_29

    work page doi:10.1007/3-540-65193-4_29 1998

  14. [16]

    Donald E. Knuth. The Art of Computer Programming, Vol. 1: Fundamental Algorithms (3rd. ed.). 1997

    1997

  15. [17]

    Donald E. Knuth. The Art of Computer Programming. 1998

    1998

  16. [18]

    Structured Variational Inference Procedures and their Realizations (as incol)

    Dan Geiger and Christopher Meek. Structured Variational Inference Procedures and their Realizations (as incol). Proceedings of Tenth International Workshop on Artificial Intelligence and Statistics, The Barbados

  17. [19]

    Stan W. Smith. An experiment in bibliographic mark-up: Parsing metadata for XML export. Proceedings of the 3rd. annual workshop on Librarians and Computers. 2010. doi:99.9999/woot07-S422

    2010

  18. [20]

    Catch me, if you can: Evading network signatures with web-based polymorphic worms

    Matthew Van Gundy and Davide Balzarotti and Giovanni Vigna. Catch me, if you can: Evading network signatures with web-based polymorphic worms. Proceedings of the first USENIX workshop on Offensive Technologies. 2007

    2007

  19. [21]

    Catch me, if you can: Evading network signatures with web-based polymorphic worms

    Matthew Van Gundy and Davide Balzarotti and Giovanni Vigna. Catch me, if you can: Evading network signatures with web-based polymorphic worms. Proceedings of the first USENIX workshop on Offensive Technologies. 2008

    2008

  20. [22]

    Catch me, if you can: Evading network signatures with web-based polymorphic worms

    Matthew Van Gundy and Davide Balzarotti and Giovanni Vigna. Catch me, if you can: Evading network signatures with web-based polymorphic worms. Proceedings of the first USENIX workshop on Offensive Technologies. 2009

    2009

  21. [23]

    Predicate Path expressions

    Sten Andler. Predicate Path expressions. Proceedings of the 6th. ACM SIGACT-SIGPLAN symposium on Principles of Programming Languages. 1979. doi:10.1145/567752.567774

    work page doi:10.1145/567752.567774 1979

  22. [24]

    LOGICS of Programs: AXIOMATICS and DESCRIPTIVE POWER

    David Harel. LOGICS of Programs: AXIOMATICS and DESCRIPTIVE POWER. 1978

    1978

  23. [25]

    Anisi , title =

    David A. Anisi , title =

  24. [26]

    Clarkson

    Kenneth L. Clarkson. Algorithms for Closest-Point Problems (Computational Geometry). 1985

    1985

  25. [27]

    Introduction to Bayesian Statistics

    Harry Thornburg. Introduction to Bayesian Statistics. 2001

    2001

  26. [28]

    CLIFFORD: a Maple 11 Package for Clifford Algebra Computations, version 11

    Rafal Ablamowicz and Bertfried Fauser. CLIFFORD: a Maple 11 Package for Clifford Algebra Computations, version 11. 2007

    2007

  27. [29]

    Stats and Analysis

    Poker-Edge.Com. Stats and Analysis. 2006

    2006

  28. [30]

    A more perfect union

    Barack Obama. A more perfect union. 2008

    2008

  29. [31]

    The fountain of youth

    Joseph Scientist. The fountain of youth. 2009

    2009

  30. [32]

    Solder man

    Dave Novak. Solder man. ACM SIGGRAPH 2003 Video Review on Animation theater Program: Part I - Vol. 145 (July 27--27, 2003). 2003. doi:99.9999/woot07-S422

    2003

  31. [33]

    Interview with Bill Kinder: January 13, 2005

    Newton Lee. Interview with Bill Kinder: January 13, 2005. Comput. Entertain. 2005. doi:10.1145/1057270.1057278

    work page doi:10.1145/1057270.1057278 2005

  32. [34]

    The Enabling of Digital Libraries

    Bernard Rous. The Enabling of Digital Libraries. Digital Libraries. 2008

    2008

  33. [36]

    (new) Finding minimum congestion spanning trees , journal =

    Werneck, Renato and Setubal, Jo\. (new) Finding minimum congestion spanning trees , journal =. 2000 , issn =. doi:10.1145/351827.384253 , acmid =

    work page doi:10.1145/351827.384253 2000

  34. [38]

    and Mei, Alessandro , title =

    Conti, Mauro and Di Pietro, Roberto and Mancini, Luigi V. and Mei, Alessandro , title =. Inf. Fusion , volume =. 2009 , issn =. doi:10.1016/j.inffus.2009.01.002 , acmid =

    work page doi:10.1016/j.inffus.2009.01.002 2009

  35. [39]

    and Hutchful, David K

    Li, Cheng-Lun and Buyuktur, Ayse G. and Hutchful, David K. and Sant, Natasha B. and Nainwal, Satyendra K. , title =. CHI '08 extended abstracts on Human factors in computing systems , year =. doi:10.1145/1358628.1358946 , acmid =

    work page doi:10.1145/1358628.1358946

  36. [40]

    , title =

    Hollis, Billy S. , title =. 1999 , isbn =

    1999

  37. [41]

    Goossens, Michel and Rahtz, S. P. and Moore, Ross and Sutor, Robert S. , title =. 1999 , isbn =

    1999

  38. [42]

    and Rosenberg, Arnold L

    Buss, Jonathan F. and Rosenberg, Arnold L. and Knott, Judson D. , title =. 1987 , source =

    1987

  39. [43]

    CHI '08: CHI '08 extended abstracts on Human factors in computing systems , year =

    , note =. CHI '08: CHI '08 extended abstracts on Human factors in computing systems , year =

  40. [44]

    Algorithms for Closest-Point Problems (Computational Geometry) , year =

    Clarkson, Kenneth Lee , advisor =. Algorithms for Closest-Point Problems (Computational Geometry) , year =

  41. [45]

    SIGCOMM Comput. Commun. Rev. , year =

  42. [46]

    2004 , isbn =

    IEEE TCSC Executive Committee , booktitle =. 2004 , isbn =. doi:10.1109/ICWS.2004.64 , acmid =

    work page doi:10.1109/icws.2004.64 2004

  43. [47]

    Distributed systems (2nd Ed.) , year =

  44. [48]

    , title =

    Petrie, Charles J. , title =. 1986 , source =

    1986

  45. [49]

    Donald E. Knuth. Seminumerical Algorithms. 1981

    1981

  46. [50]

    E-commerce and cultural values , year =

    Kong, Wei-Chang , Title =. E-commerce and cultural values , year =

  47. [51]

    E-commerce and cultural values , year =

    Kong, Wei-Chang , type =. E-commerce and cultural values , year =

  48. [52]

    Chapter 9 , booktitle =

    Kong, Wei-Chang , editor =. Chapter 9 , booktitle =. 2002 , address =

    2002

  49. [53]

    E-commerce and cultural values , editor =

    Kong, Wei-Chang , title =. E-commerce and cultural values , editor =. 2003 , isbn =

    2003

  50. [54]

    E-commerce and cultural values - (InBook-num-in-chap) , chapter =

    Kong, Wei-Chang , editor =. E-commerce and cultural values - (InBook-num-in-chap) , chapter =. 2004 , address =

    2004

  51. [55]

    E-commerce and cultural values (Inbook-text-in-chap) , chapter =

    Kong, Wei-Chang , editor =. E-commerce and cultural values (Inbook-text-in-chap) , chapter =. 2005 , address =

    2005

  52. [56]

    E-commerce and cultural values (Inbook-num chap) , chapter =

    Kong, Wei-Chang , editor =. E-commerce and cultural values (Inbook-num chap) , chapter =. 2006 , address =

    2006

  53. [57]

    Microelectron

    Mehdi Saeedi and Morteza Saheb Zamani and Mehdi Sedighi , title =. Microelectron. J. , volume =. 2010 , pages =

    2010

  54. [58]

    Mehdi Saeedi and Morteza Saheb Zamani and Mehdi Sedighi and Zahra Sasanian , title =. J. Emerg. Technol. Comput. Syst. , volume =

  55. [59]

    Kirschmer, Markus and Voight, John , title =. SIAM J. Comput. , issue_date =. 2010 , issn =. doi:10.1137/080734467 , acmid =

    work page doi:10.1137/080734467 2010

  56. [60]

    Hoare, C. A. R. , title =. Structured programming (incoll) , editor =. 1972 , isbn =

    1972

  57. [61]

    History of programming languages I (incoll) , editor =

    Lee, Jan , title =. History of programming languages I (incoll) , editor =. 1981 , isbn =. doi:10.1145/800025.1198348 , acmid =

    work page doi:10.1145/800025.1198348 1981

  58. [62]

    , title =

    Dijkstra, E. , title =. Classics in software engineering (incoll) , year =

  59. [63]

    , title =

    Wenzel, Elizabeth M. , title =. Multimedia interface design (incoll) , year =. doi:10.1145/146022.146089 , acmid =

    work page doi:10.1145/146022.146089

  60. [64]

    , title =

    Mumford, E. , title =. Critical issues in information systems research (incoll) , year =

  61. [65]

    and Golden, Donald G

    McCracken, Daniel D. and Golden, Donald G. , title =. 1990 , isbn =

    1990

  62. [66]

    The analysis of linear partial differential operators

    H. The analysis of linear partial differential operators. 1985 , PAGES =

    1985

  63. [67]

    IEEE", address =

    A. Adya and P. Bahl and J. Padhye and A.Wolman and L. Zhou , title =. Proceedings of the IEEE 1st International Conference on Broadnets Networks (BroadNets'04) , publisher = "IEEE", address = "Los Alamitos, CA", year =

  64. [68]

    I. F. Akyildiz and W. Su and Y. Sankarasubramaniam and E. Cayirci , title =. Comm. ACM , volume = 38, number = "4", year =

  65. [69]

    I. F. Akyildiz and T. Melodia and K. R. Chowdhury , title =. Computer Netw. , volume = 51, number = "4", year =

  66. [70]

    ACM", address =

    P. Bahl and R. Chancre and J. Dungeon , title =. Proceeding of the 10th International Conference on Mobile Computing and Networking (MobiCom'04) , publisher = "ACM", address = "New York, NY", year =

  67. [71]

    8 (Special Issue on Sensor Networks)

    D. Culler and D. Estrin and M. Srivastava , title =. IEEE Comput. , volume = 37, number = "8 (Special Issue on Sensor Networks)", publisher = "IEEE", address = "Los Alamitos, CA", year =

  68. [72]

    Natarajan and M

    A. Natarajan and M. Motani and B. de Silva and K. Yap and K. C. Chua , title =. Network Architectures , editor =. 960935712

  69. [73]

    Tzamaloukas and J

    A. Tzamaloukas and J. J. Garcia-Luna-Aceves , title =

  70. [74]

    Zhou and J

    G. Zhou and J. Lu and C.-Y. Wan and M. D. Yarvis and J. A. Stankovic , title =

  71. [75]

    Mapping Powerlists onto Hypercubes

    Jacob Kornerup. Mapping Powerlists onto Hypercubes. 1994

    1994

  72. [76]

    Automatic Parallelization for Distributed-Memory Multiprocessing Systems

    Michael Gerndt. Automatic Parallelization for Distributed-Memory Multiprocessing Systems

  73. [77]

    J. E. Archer, Jr. and R. Conway and F. B. Schneider. User recovery and reversal in interactive systems. ACM Trans. Program. Lang. Syst

  74. [78]

    D. D. Dunlop and V. R. Basili. Generalizing specifications for uniformly implemented loops. ACM Trans. Program. Lang. Syst

  75. [79]

    Heering and P

    J. Heering and P. Klint. Towards monolingual programming environments. ACM Trans. Program. Lang. Syst

  76. [80]

    Donald E. Knuth. The book

  77. [81]

    Korach and D

    E. Korach and D. Rotem and N. Santoro. Distributed algorithms for finding centers and medians in networks. ACM Trans. Program. Lang. Syst

  78. [82]

    : A Document Preparation System

    Leslie Lamport. : A Document Preparation System

  79. [83]

    F. Nielson. Program transformations in a denotational setting. ACM Trans. Program. Lang. Syst

  80. [84]

    Brian K. Reid. A high-level approach to computer document formatting. Proceedings of the 7th Annual Symposium on Principles of Programming Languages

Showing first 80 references.