Pith. sign in

REVIEW 3 major objections 6 minor 35 references

CoMuMDR: Code-mixed Multi-modal Multi-domain corpus for Discourse paRsing in conversations

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper claims that CoMuMDR is the first code-mixed, multi-modal, multi-domain discourse corpus for conversations, and that state-of-the-art parsers perform poorly on it, especially on relation classification.

desk verdict A genuinely new code-mixed discourse corpus with real value as a resource, but the benchmark-based claims about difficulty are not supported by the paper's own numbers. read the letter →

arxiv 2506.08504 v1 pith:OTAZDPSM submitted 2025-06-10 cs.CL cs.AIcs.LG

Divyaksh Shukla , Ritesh Baviskar , Dwijesh Gohil , Aniket Tiwari , Atul Shree , Ashutosh Modi This is my paper
classification cs.CLcs.AIcs.LG
keywords code-mixeddiscourseparsingmulti-modalcorpusconversationrelationsHindi-EnglishcallcentersSDRTannotationrelationclassificationresourcepaper
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This resource paper introduces CoMuMDR, a discourse-parsing corpus collected from real customer call-center conversations in Hindi-English code-mixed speech, with both audio and transcripts and annotations of nine discourse relations across five domains. The authors benchmark five state-of-the-art discourse parsers and GPT-4o on link prediction and relation classification and find that model performance on CoMuMDR is uniformly lower than on existing single-domain English corpora such as STAC and Molweni, with relation classification the weakest component. The paper argues that the poor scores reflect the challenges of multi-domain, code-mixed, conversational audio and that better models are needed for such realistic settings.

What carries the argument

The load-bearing artifact is the CoMuMDR corpus: 799 two-party call-center dialogues, transcribed and diarized into 8,811 utterances, manually segmented into elementary discourse units (clauses), and linked into a directed acyclic discourse graph whose edges carry one of nine discourse relation labels derived from Segmented Discourse Representation Theory (SDRT), a theory of how clauses connect through discourse relations, plus a newly added Question-Answer Complaint Pair. The paper also relies on the standard evaluation machinery of F1 scores for link prediction and link+relation classification, alongside inter-annotator metrics (Kappa, Jaccard, span exact/partial match) to characterize the difficulty of the annotation task.

What would settle it

Re-annotate a sample of CoMuMDR with careful adjudication, then measure agreement between the gold and a fresh expert annotation and compute the resulting human upper-bound F1; if that upper bound is close to the reported model scores, the claim of model underperformance is weakened because the task itself would be under-specified.

Watch

Extended reading notes

Core claim

The central claim is that CoMuMDR is the first corpus to combine code-mixing, multiple modalities, and multiple domains for discourse parsing in two-party dialogues, and that it exposes a real gap: state-of-the-art discourse parsers trained on this corpus achieve markedly lower F1 in relation classification than they do on STAC or Molweni, while link prediction is comparatively less affected. The authors attribute the gap to code-mixed language, imperfect ASR transcription and diarization, and domain variation, and they provide per-relation error analyses showing that rare relation types are rarely or never predicted.

Load-bearing premise

The benchmark conclusion assumes the gold annotations are reliable enough to make low model scores evidence of task difficulty, even though annotators disagree substantially on relation labels (Relationship Kappa is 0.319).

Editorial extensions

If this is right

  • Model scores on CoMuMDR provide a lower-bound benchmark for discourse parsing in realistic code-mixed, speech-derived conversations.
  • The nine-relation scheme, including Question-Answer Complaint Pair, offers a compact adaptation of SDRT that can be reused for other two-party customer-service corpora.
  • Per-relation F1 tables show that rare relations receive zero F1 across models, so progress on this corpus depends on more data or a different learning signal for those classes.
  • Because the released audio features accompany the transcripts, CoMuMDR supports testing whether acoustic information improves discourse parsing over text alone.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Interpreted charitably, the low inter-annotator agreement suggests that the boundary between relations like Acknowledgment and Question-Answer Pair is intrinsically ambiguous; a controlled experiment that asks human annotators to choose among the nine labels for the same spans could reveal how much of the model gap is label noise rather than model limitation.
  • The paper merges several SDRT labels into Continuation, so CoMuMDR cannot evaluate distinctions that matter in other genres; direct F1 comparisons with STAC and Molweni are therefore approximate rather than apples-to-apples.
  • A natural follow-up is to translate CoMuMDR transcripts into English-only text and retrain the same parsers, which would separate the effect of code-mixing from the effect of domain shift.
  • The audio-conditioned direction the paper flags is directly testable: if a parser that consumes the released audio features does not beat the text-only parser, the bottleneck likely lies in the annotation scheme rather than in the missing modality.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. CoMuMDR is presented as a new Hindi-English code-mixed, multi-modal (audio plus text), multi-domain discourse corpus of two-party customer call-center conversations, annotated with nine SDRT-inspired discourse relations at the EDU-span level. The paper describes corpus construction, reports inter-annotator agreement, and evaluates five discourse parsers plus GPT-4o under English-only and multilingual embeddings, comparing against STAC and Molweni. The headline claim is that state-of-the-art parsers underperform on CoMuMDR, demonstrating the difficulty of code-mixed multi-domain discourse parsing.

Significance. The resource fills a genuine gap: existing conversational discourse corpora are monolingual, written, single-domain, and not audio-based. The authors are transparent about the low inter-annotator agreement, provide per-relation breakdowns, and plan to release code, transcriptions, embeddings, and audio features, which is a useful commitment. Running the same five parsers on CoMuMDR, STAC, and Molweni is also a reasonable first benchmark. However, the paper's principal claim that SoTA models underperform on CoMuMDR is currently not supported with sufficient rigor, because the comparison is confounded by annotation reliability, label imbalance, link-distance distributions, and missing significance testing.

major comments (3)
  1. [§3, Table 2] The benchmark interpretation in §4 rests on CoMuMDR being a reliable gold standard, but the reported Relationship Kappa is 0.3190 and Structured Kappa is 0.4044, and the Limitations section itself concedes that label confusion and EDU-span differences produced low agreement. The paper does not report post-adjudication agreement or any human upper bound on link+relation F1, so it is impossible to tell whether the poor model scores reflect task difficulty or label noise. Please add post-adjudication agreement (gold vs. each annotator), per-relation agreement, and a human-annotation upper bound computed on the same evaluation metric.
  2. [§4, Tables 3, 4, 8] The claim that SoTA models underperform on CoMuMDR is confounded by label sparsity and link locality. Table 4 shows that every relation with support at or below 32 receives F1 0.00 from most models, whereas the high-support relations reach F1 0.40-0.70, which is comparable to the overall link+relation scores on STAC and Molweni. Table 8 shows that CoMuMDR links are predominantly local (mean distances 0.95-1.07 versus 0.67-1.86 for STAC/Molweni), which explains the higher link-only F1 on CoMuMDR (e.g., multilingual SSP-BERT 0.9531 vs. 0.7250 and 0.8205). The comparison should be repeated with macro-averaged scores conditioned on relation support, with an evaluation restricted to shared high-frequency relations, and/or with link-distance-matched subsets.
  3. [§4, Results paragraph] The sentence 'For both the settings, CoMuMDR scores are lowest across all the models' is contradicted by the paper's own tables. English-only Hierarchical link+relation F1 is 0.5786 on CoMuMDR versus 0.5704 on STAC, English-only SSP-BERT is 0.5675 versus 0.5623, and GPT-4o link+relation F1 on CoMuMDR (0.2808) exceeds both STAC (0.2729) and Molweni (0.1474) in Table 5. Please qualify the claim and provide confidence intervals or significance tests, since differences below 0.01 on an 81-dialogue test set are within noise.
minor comments (6)
  1. [Abstract and §1] The phrase 'The current discourse parsing datasets ... consists' should be 'consist'; also consider making the relation between the audio modality and the text-only baselines explicit in the abstract.
  2. [Table 1] The row 'Annotator Metrics Kappa Kappa Kappa, Jaccard' is hard to parse; spell out which metrics apply to which corpus and provide the STAC and Molweni Kappa values for direct comparison.
  3. [§3 and Table 6] The relation 'Diarization Continuation' is introduced as an annotation category and appears in Figure 7, but it is not listed in Table 6; make its role and exclusion from evaluation consistent across the main text, tables, and appendix.
  4. [Appendix F] The GPT-4o prompt instructs the model to use 'dia-continuation' for same-speaker split utterances, but this relation is excluded from scoring; clarify how such predictions were handled in the reported numbers.
  5. [Appendix C.5, Algorithm 2] The threshold parameter for span partial match is not given a value; state the threshold so that the reported agreement scores are reproducible.
  6. [Figure 1] The relation arcs and labels (QACPAck, dc) are difficult to follow; please redraw with clear arrows and a legend.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the corpus and its benchmark results are empirically constructed and externally testable, with no fitted parameter renamed as a prediction.

full rationale

This is a resource paper, not a derivation. The central claims are that CoMuMDR is a new code-mixed, multi-modal, multi-domain discourse corpus and that existing discourse parsers and GPT-4o perform poorly on it. Neither claim reduces to its inputs. The corpus was built through ASR transcription, diarization, anonymization, and manual annotation; no gold links or discourse labels were fitted to, or derived from, the downstream model predictions. The baselines are external published systems (Shi and Huang 2019; Liu and Chen 2021; Wang et al. 2021; Yu et al. 2022; Chi and Rudnicky 2022) implemented from their papers and trained from scratch on STAC, Molweni, and CoMuMDR, so the F1 comparisons are external benchmark results rather than constructional identities. The only overlapping-author citation is the ASR system (Verma et al., 2023) used to produce transcripts; this is a pipeline component and is not invoked to justify the underperformance claim, nor does it force any benchmark outcome by definition. The reported low inter-annotator agreement (Structured Kappa 0.4044, Relationship Kappa 0.3190) and the potential confounds in Tables 3 and 4 (rare labels with zero F1, mostly local links) concern the validity or interpretability of the benchmark comparison, i.e., correctness or experimental-design risk, not circularity. Similarly, the statement that 'For both the settings, CoMuMDR scores are lowest across all the models' is contradicted by some rows in Table 3 (English-only Hierarchical 0.5786 and SSP-BERT 0.5675 link+relation F1 on CoMuMDR exceed the corresponding STAC scores), but that is a presentational/statistical issue, not circular reasoning. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the authors' prior work, no ansatz is smuggled in via citation, and no known result is merely renamed. The derivation chain, such as it is, is therefore self-contained and the results are externally testable.

Assumptions & free parameters 1 free parameters · 4 assumptions · 1 invented entities

The corpus construction relies on several domain assumptions about the suitability of SDRT and noisy transcripts, plus one ad hoc annotation rule (head-of-CDU linking). The only unstated numeric parameter is the partial-match threshold in Algorithm 2. The QACP relation is the paper's one invented annotation category, with no independent evidence.

free parameters (1)
  • Span partial match threshold = not reported
    Algorithm 2 uses a Jaccard threshold to define a partial span match, but the value is not stated. This threshold influences the reported inter-annotator agreement in Table 2, and its choice is unverifiable.
assumptions (4)
  • domain assumption SDRT is an appropriate annotation scheme for two-party call-center conversations.
    Section 3 states the labels are based on SDRT. SDRT was developed mainly for written dialogue, and its applicability to code-mixed spoken conversation is assumed without independent validation.
  • domain assumption Transcribed and diarized text is an adequate proxy for audio for discourse annotation.
    Section 3 and the Limitations state that ASR and diarization are imperfect and annotators corrected diarization errors, but the discourse relations are annotated on noisy transcripts rather than directly on audio. This assumption is load-bearing for the text-based benchmarks.
  • ad hoc to paper Linking an EDU only to the head of a CDU preserves the discourse structure.
    Section 3: 'we instructed the annotators to connect an EDU with only the head of a CDU.' This simplification was adopted from prior work but may lose CDU-level relations and reduce annotation consistency.
  • domain assumption The low Kappa values are an acceptable measure of annotation reliability.
    Table 2 and Appendix C.5 report Kappa and other metrics. The paper treats these as meaningful, but the Relationship Kappa of 0.3190 is low by common standards and no post-adjudication reliability is reported.
invented entities (1)
  • Question-Answer Complaint Pair (QACP) relation
    purpose: To annotate a discourse relation where a question and its answer relate to a customer complaint.
    This new label is proposed by the paper based on observed call-center patterns. There is no external validation; in the test set it has support of only 3 examples (Table 4), so its reliability and necessity are not established.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CoMuMDR: Code-mixed Multi-modal Multi-domain corpus for Discourse paRsing in conversations." pith.science (2026). https://pith.science/paper/OTAZDPSM

@misc{pith2026250608504,
  author       = {Pith},
  title        = {Pith review of: CoMuMDR: Code-mixed Multi-modal Multi-domain corpus for Discourse paRsing in conversations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OTAZDPSM}},
  note         = {Machine review of arXiv:2506.08504}
}
read the original abstract

Discourse parsing is an important task useful for NLU applications such as summarization, machine comprehension, and emotion recognition. The current discourse parsing datasets based on conversations consists of written English dialogues restricted to a single domain. In this resource paper, we introduce CoMuMDR: Code-mixed Multi-modal Multi-domain corpus for Discourse paRsing in conversations. The corpus (code-mixed in Hindi and English) has both audio and transcribed text and is annotated with nine discourse relations. We experiment with various SoTA baseline models; the poor performance of SoTA models highlights the challenges of multi-domain code-mixed corpus, pointing towards the need for developing better models for such realistic settings.

Figures

Figures reproduced from arXiv: 2506.08504 by the authors.

Figure 1
Figure 1. An example of a two-party call-center conver [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Distribution of discourse labels in CoMuMDR. based corpora have been proposed for Discourse parsing: STAC (Asher et al., 2016) and Molweni (Li et al., 2020) (check details in App. A.2) [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Comparing SDDP, Struct-Aware, and SA￾DPMD across epochs on CoMuMDR test set. The values on each plot indicate the highest F1 score with respect to each model and discourse parsing method. CoMuMDR has the lowest performance, possibly due to the presence of multiple domains and the chal￾lenge of domain adaptation (Liu and Chen, 2021). Error Analysis: In [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: A sample conversation taken from CoMuMDR. Utterances from the customer are marked in purple, and those of the customer center representative are green. The gold and predicted relations are marked on the left and right sides. You are given a dialogue conversation betwee…
Figure 4
Figure 4. Figure 4: Distribution of the discourse relation labels [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 6
Figure 6. Figure 6: Prompt template used for evaluating GPT-4 [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: Distance between linked EDUs for different corpora [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

35 extracted references · 19 canonical work pages

  1. [1]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Stergos Afantenos, Eric Kow, Nicholas Asher, and J \'e r \'e my Perret. 2015. https://doi.org/10.18653/v1/D15-1109 Discourse parsing for multi-party chat dialogues . In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 928--937, Lisbon, Portugal. Association for Computational Linguistics

    work page doi:10.18653/v1/d15-1109 2015

  4. [4]

    Nicholas Asher, Julie Hunter, Mathieu Morey, Benamara Farah, and Stergos Afantenos. 2016. https://aclanthology.org/L16-1432 Discourse structure and dialogue acts in multiparty dialogue: the STAC corpus . In Proceedings of the Tenth International Conference on Language Resources and Evaluation ( LREC '16) , pages 2721--2727, Portoro z , Slovenia. European ...

    work page 2016

  5. [5]

    Nicholas Asher and Alex Lascarides. 2005. https://www.cambridge.org/de/universitypress/subjects/languages-linguistics/computational-linguistics/logics-conversation Logics of Conversation . Cambridge University Press, Cambridge, England, UK

    work page 2005

  6. [6]

    Jiaao Chen and Diyi Yang. 2023. https://doi.org/10.18653/v1/2023.findings-acl.454 Controllable conversation generation with conversation structures via diffusion models . In Findings of the Association for Computational Linguistics: ACL 2023, pages 7238--7251, Toronto, Canada. Association for Computational Linguistics

    work page doi:10.18653/v1/2023.findings-acl.454 2023

  7. [7]

    Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Xiangzhan Yu, and Furu Wei. 2022. https://doi.org/10.1109/JSTSP.2022.3188113 Wavlm: Large-scale self-supervised pre-training for full stack speech proces...

    arXiv 2022

  8. [8]

    Alexander Chernyavskiy and Dmitry Ilvovsky. 2023. https://doi.org/10.18653/v1/2023.sigdial-1.48 Transformer-based multi-party conversation generation using dialogue discourse acts planning . In Proceedings of the 24th Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 519--529, Prague, Czechia. Association for Computational Linguistics

    work page doi:10.18653/v1/2023.sigdial-1.48 2023

Show all 35 references
  1. [9]

    Ta-Chung Chi and Alexander Rudnicky. 2022. https://doi.org/10.18653/v1/2022.sigdial-1.32 Structured dialogue discourse parsing . In Proceedings of the 23rd Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 325--335, Edinburgh, UK. Association for Co...

    2022 doi

  2. [10]

    Sabit Hassan and Malihe Alikhani. 2023. https://doi.org/10.18653/v1/2023.ijcnlp-main.28 D is CG en: A framework for discourse-informed counterspeech generation . In Proceedings of the 13th International Joint Conference on Natural Language Processing and the 3rd Conference of ...

    2023 doi

  3. [11]

    Kenneth Heafield. 2011. https://aclanthology.org/W11-2123/ K en LM : Faster and smaller language model queries . In Proceedings of the Sixth Workshop on Statistical Machine Translation, pages 187--197, Edinburgh, Scotland. Association for Computational Linguistics

    2011

  4. [12]

    Jan-Christoph Klie, Michael Bugert, Beto Boullosa, Richard Eckart de Castilho, and Iryna Gurevych. 2018. https://aclanthology.org/C18-2002 The INCE p TION platform: Machine-assisted and knowledge-oriented interactive annotation . In Proceedings of the 27th International Confer...

    2018

  5. [13]

    Nithin Rao Koluguri, Taejin Park, and Boris Ginsburg. 2021. https://arxiv.org/abs/2110.04410 Titanet: Neural model for speaker representation with 1d depth-wise separable convolutions and global context . Preprint, arXiv:2110.04410

    2021 arXiv

  6. [14]

    Jiaqi Li, Ming Liu, Min-Yen Kan, Zihao Zheng, Zekun Wang, Wenqiang Lei, Ting Liu, and Bing Qin. 2020. https://doi.org/10.18653/v1/2020.coling-main.238 Molweni: A challenge multiparty dialogues-based machine reading comprehension dataset with discourse structure . In Proceeding...

    2020 doi

  7. [15]

    Jiaqi Li, Ming Liu, Bing Qin, and Ting Liu. 2022. A survey of discourse parsing. Frontiers of Computer Science, 16(5):165329

    2022

  8. [16]

    Jiaqi Li, Ming Liu, Bing Qin, Zihao Zheng, and Ting Liu. 2019. https://arxiv.org/abs/1911.03514 An annotation scheme of a large-scale multi-party dialogues dataset for discourse parsing and machine comprehension . Preprint, arXiv:1911.03514

    2019 arXiv

  9. [17]

    Junyi Jessy Li, Kapil Thadani, and Amanda Stent. 2016. https://doi.org/10.18653/v1/W16-3617 The role of discourse units in near-extractive summarization . In Proceedings of the 17th Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 137--147, Los Ang...

    2016 doi

  10. [18]

    Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. http://arxiv.org/abs/1907.11692 Roberta: A robustly optimized bert pretraining approach . CoRR, abs/1907.11692

    2019 arXiv

  11. [19]

    Zhengyuan Liu and Nancy Chen. 2021. https://doi.org/10.18653/v1/2021.codi-main.11 Improving multi-party dialogue discourse parsing via domain integration . In Proceedings of the 2nd Workshop on Computational Approaches to Discourse, pages 122--127, Punta Cana, Dominican Republ...

    2021 doi

  12. [20]

    Zhengyuan Liu, Ke Shi, and Nancy Chen. 2021. https://doi.org/10.18653/v1/2021.codi-main.15 DMRST : A joint framework for document-level multilingual RST discourse segmentation and parsing . In Proceedings of the 2nd Workshop on Computational Approaches to Discourse, pages 154-...

    2021 doi

  13. [21]

    Michal Lukasik, Boris Dadachev, Kishore Papineni, and Gon c alo Sim \ o es. 2020. https://doi.org/10.18653/v1/2020.emnlp-main.380 Text segmentation by cross segment attention . In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), p...

    2020 doi

  14. [22]

    Mann and Sandra A

    William C. Mann and Sandra A. Thompson. 1988. https://doi.org/doi:10.1515/text.1.1988.8.3.243 Rhetorical structure theory: Toward a functional theory of text organization . Text - Interdisciplinary Journal for the Study of Discourse, 8(3):243--281

    1988 doi

  15. [23]

    Mary L. McHugh. 2012. https://pmc.ncbi.nlm.nih.gov/articles/PMC3900052 Interrater reliability: the kappa statistic . Biochemia Medica, 22(3):276

    2012

  16. [24]

    Philippe Muller, Stergos Afantenos, Pascal Denis, and Nicholas Asher. 2012. https://aclanthology.org/C12-1115 Constrained decoding for text-level discourse parsing . In Proceedings of COLING 2012 , pages 1883--1900, Mumbai, India. The COLING 2012 Organizing Committee

    2012

  17. [25]

    Romain Paulus, Caiming Xiong, and Richard Socher. 2018. https://openreview.net/forum?id=HkAClQgA- A deep reinforced model for abstractive summarization . In International Conference on Learning Representations

    2018

  18. [26]

    Jeffrey Pennington, Richard Socher, and Christopher Manning. 2014. https://doi.org/10.3115/v1/D14-1162 G lo V e: Global vectors for word representation . In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing ( EMNLP ) , pages 1532--1543, Doh...

    2014 doi

  19. [27]

    Alan Lee Rashmi Prasad, Bonnie Webber and Aravind Joshi. 2019. https://catalog.ldc.upenn.edu/LDC2019T05 Penn Discourse Treebank Version 3.0 - Linguistic Data Consortium

    2019

  20. [28]

    Nils Reimers and Iryna Gurevych. 2019. https://doi.org/10.18653/v1/D19-1410 Sentence- BERT : Sentence embeddings using S iamese BERT -networks . In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference...

    2019 doi

  21. [29]

    Zhouxing Shi and Minlie Huang. 2019. https://doi.org/10.1609/aaai.v33i01.33017007 A deep sequential model for discourse parsing on multi-party dialogues . In AAAI 2019, AAAI'19/IAAI'19/EAAI'19. AAAI Press

    2019 doi

  22. [30]

    Tushar Verma, Atul Shree, and Ashutosh Modi. 2023. https://doi.org/10.21437/Interspeech.2023-757 Asr for low resource and multilingual noisy code-mixed speech . In INTERSPEECH 2023, pages 3242--3246

    2023 doi

  23. [31]

    Ante Wang, Linfeng Song, Hui Jiang, Shaopeng Lai, Junfeng Yao, Min Zhang, and Jinsong Su. 2021. https://doi.org/10.24963/ijcai.2021/543 A structure self-aware model for discourse parsing on multi-party dialogues . In Proceedings of the Thirtieth International Joint Conference ...

    2021 doi

  24. [32]

    Yizhong Wang, Sujian Li, and Jingfeng Yang. 2018. https://doi.org/10.18653/v1/D18-1116 Toward fast and accurate neural discourse segmentation . In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 962--967, Brussels, Belgium. Associa...

    2018 doi

  25. [33]

    Chang-Sung Yu. 1986. https://doi.org/10.1002/net.3230160110 Graph theory, by w. t. tutte, encyclopedia of mathematics and its applications, volume 21, addison-wesley publishing company, menlo park, ca., 1984, 333 pp. price: 45.00 . Networks, 16(1):107--108

    1986 doi

  26. [34]

    Nan Yu, Guohong Fu, and Min Zhang. 2022. https://aclanthology.org/2022.coling-1.477 Speaker-aware discourse parsing on multi-party dialogues . In Proceedings of the 29th International Conference on Computational Linguistics, pages 5372--5382, Gyeongju, Republic of Korea. Inter...

    2022

  27. [35]

    Duzhen Zhang, Feilong Chen, and Xiuyi Chen. 2023. https://doi.org/10.18653/v1/2023.acl-long.408 D ual GAT s: Dual graph attention networks for emotion recognition in conversations . In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Vol...

    2023 doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.