Pith. sign in

REVIEW 3 major objections 5 minor 32 references

Structuring Latent Spaces for Stylized Response Generation

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read This paper claims that sharing a structured latent space between a conversation model and an autoencoder over non-parallel style text lets a chatbot produce responses in a target style without losing relevance.

desk verdict A solid, well-scoped extension of SPACEFUSION to non-parallel style data, with a clean method and a credible human evaluation; the automatic metrics are partly circular, so the headline outperformance claim needs an independent check. read the letter →

arxiv 1909.05361 v1 pith:CRVJF65P submitted 2019-09-03 cs.CL cs.AI

Xiang Gao , Yizhe Zhang , Sungjin Lee , Michel Galley , Chris Brockett , Jianfeng Gao , Bill Dolan This is my paper · ORCID
classification cs.CLcs.AI
keywords stylizedresponsegenerationlatentspacealignmentnon-parallelstyletransferconversationmodelingsequence-to-sequenceautoencoderintensitycontrolmulti-tasklearning
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

Generating conversational responses in a chosen style usually requires parallel data that does not exist for most styles. This paper proposes StyleFusion, a way to train one model on ordinary conversations plus unpaired sentences in the target style, and to make the two sources meet in a shared latent space: a compact vector representation of sentences. The model learns to place style sentences near conversation contexts with related meaning, then at inference samples near the conversation model's prediction, with the sampled distance controlling how strong the style is. Experiments on Reddit dialogue with arXiv-style and Sherlock-Holmes-style sentences show responses that stay relevant while becoming measurably more stylized, improving on baselines that blend models at decoding time or share only a decoder. If the geometry of its latent space is trustworthy, the method adds a continuously adjustable style knob to existing conversation systems without retraining on parallel style data.

What carries the argument

The load-bearing object is the structured shared latent space, with Euclidean distance as the assumed measure of semantic and stylistic change. Two regularization families shape it. Fusion objectives minimize cross-latent-space nearest-neighbor distances between conversation predictions $z_{\mathrm{S2S}}(x)$, autoencoded responses $z_{\mathrm{AE}}(y)$, and style-sentence codes $z_{\mathrm{AE}}(s)$, while a spread-out term keeps each space from collapsing to a point. Smoothness objectives interpolate between codes, $z=(1-u)z_1+u z_2+\epsilon$, and require the decoder to generate intermediate text, forcing gradual semantic and stylistic transitions. Together they generalize an earlier paired-data fusion regularizer to non-parallel style data. At inference the model samples $z=z_{\mathrm{S2S}}(x)+r$ with normalized radius $\rho$ and ranks hypotheses by relevance and style probability, making the radius a continuous style-intensity control and the direction a content control.

What would settle it

Take a style corpus whose topics barely overlap the conversation data, for example physics abstracts against casual Reddit chatter, and measure response BLEU against stylized references as $\rho$ grows. If appropriateness collapses at moderate $\rho$ while style intensity rises, the alignment depends on topical proximity rather than a general style structure; if appropriateness holds, the geometric assumption is doing the work.

Watch

Extended reading notes

Core claim

StyleFusion claims that the trade-off between relevance and style in non-parallel stylized response generation is caused by conversation and style data living in separate clusters of the latent space, and that explicit regularization can dissolve it. The training objective combines a sequence-to-sequence conversation model and an autoencoder over style sentences that share a decoder, with new regularizers: a fusion loss that pulls each conversation prediction toward the nearest style-sentence code, and a smoothness loss that requires interpolating between codes to reconstruct intermediate text. At inference the model adds a noise vector of normalized length $\rho$ to the prediction code and ranks candidate outputs by a weighted combination of relevance probability and style-classifier probability. In human and automatic evaluation on arXiv-like and Holmes-like targets, StyleFusion attains the best harmonic mean of appropriateness and style intensity among trainable systems, and the same model can be dialed from ordinary to strongly stylized responses by increasing $\rho$.

Load-bearing premise

The regularizers assume that Euclidean distance and straight-line interpolation in the shared latent space correspond to semantic relatedness and gradual style-content change; if the learned geometry does not respect that, sampling near a conversation prediction will not yield relevant stylized responses.

Editorial extensions

If this is right

  • A single trained StyleFusion model covers the whole style spectrum, from ordinary to strongly stylized, by adjusting $\rho$ at inference.
  • Because style sentences are aligned by nearest-neighbor distance rather than paired labels, any non-conversational style corpus can be injected into a conversation model without parallel data.
  • The ablation results attribute the gain to the fusion and smoothness terms: without them, style intensity drops sharply and diversity collapses, matching the multi-task baseline's bland outputs.
  • Ranking by a blend of relevance probability and style-classifier probability lets the system reject non-stylized samples at larger radii, so style is not purchased at the price of relevance.

Reading between the lines

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

  • Editorial inference: if the nearest-neighbor alignment works because of topical overlap between style and conversation data, applying StyleFusion to a style corpus with little topical overlap would stress-test whether it learns a general style geometry or merely borrows nearby content.
  • Editorial inference: the radius $\rho$ could be calibrated to an external measure of style intensity, such as human ratings of formality, to give users a meaningful style dial rather than an arbitrary distance.
  • Editorial inference: the interpolation smoothness loss suggests a recipe beyond chat: any generator paired with an autoencoder over unlabeled text could acquire a continuous attribute control through the same two regularizers.
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 / 5 minor

Summary. The paper proposes StyleFusion, an approach to stylized response generation that shares a structured latent space between a sequence-to-sequence conversation model and an autoencoder trained on non-parallel style text. The training objective combines standard S2S loss with fusion and smoothness regularizers that pull the S2S prediction, the AE encoding of responses, and the AE encoding of style sentences into a common latent geometry. At inference, the model samples around the S2S prediction in random directions with a tunable radius and reranks 100 candidates by a weighted combination of relevance and a style-classifier score. Experiments on Reddit-to-arXiv and Reddit-to-Holmes compare against MTask, S2S+LM, retrieval, random style sentences, and human references, using BLEU, diversity metrics, classifier-based and keyword-count style metrics, and human appropriateness and style judgments. The paper reports that StyleFusion achieves the best harmonic mean of appropriateness and style intensity among all non-human systems while maintaining relevance across sampling radii.

Significance. Stylized response generation without parallel data is a practical and currently difficult problem, and StyleFusion offers a conceptually clean solution: regularize a shared latent space so that moving from a conversational prediction toward a stylized sentence's encoding yields a controllable, relevant style shift. The strength of the paper is its clear problem formulation, the open-source release of the implementation and data-generation scripts, and the qualitative latent-space analysis that supports the proposed mechanism. The paper also has the virtue of reporting multiple automatic metrics and a human evaluation with two separate tasks. If the empirical claims can be made robust, the contribution is a useful extension of SPACEFUSION to non-parallel data and a solid baseline for future work in controlled dialogue generation.

major comments (3)
  1. [§4.1, §3.3, §4.3, Eq. (15), Table 6] The automatic evaluation is partly circular. The same neural and ngram style classifiers are used to build Dtest by filtering Reddit 2013 (Section 4.1), to define Pstyle in the reranking score of Eq. (15), and to compute two of the three automatic style-intensity metrics in Table 6. A system that generates a diverse candidate pool and reranks by Eq. (15) will therefore score higher on the neural and ngram style columns and on BLEU computed against Dtest references, even if its outputs are not genuinely more style-appropriate. The count metric and the human style-classification task are independent and do show a StyleFusion advantage, so the circularity is not automatically fatal. The paper should verify the classifier-based claims with held-out or externally pretrained classifiers and, ideally, with Dtest references filtered by human annotators or an independent classifier.
  2. [§5.3, Table 5] The headline human-evaluation claim is not fully supported by the reported statistics. The text says that STYLE FUSION and MTask show 'not statistically different' appropriateness, but no test statistic, p-value, confidence interval, or equivalence margin is reported, and no inter-annotator agreement measure is given for the 500-context, 5-annotator setup. The human style-intensity advantage is likewise reported without any uncertainty. Because the central claim is that StyleFusion improves style without sacrificing appropriateness, the authors should report per-system means with bootstrap confidence intervals or paired significance tests and should state the equivalence margin used for the appropriateness comparison.
  3. [§4.5, §5.4, Eq. (15)] The reranking mechanism in Eq. (15) is used by all trainable systems, including the ablation variants, so the automatic ablation results are also entangled with the classifier loop. In Table 6, the difference between +Lconv and +Lstyle in the neural and ngram style columns may reflect not only a genuine style improvement but also a larger or better-ranked candidate pool under the same Pstyle. The authors should confirm the ablation with the independent count metric and human judgments on the same samples, or at least report a sensitivity analysis of the reranking weight λ and show that the conclusion is stable across λ.
minor comments (5)
  1. [§3.2] The sentence beginning 'The iven context may or may not be in the target style' contains a typo: 'iven' should be 'given'.
  2. [Fig. 3, Fig. 4] The style-intensity and fine-grained style plots show aggregate trends without confidence intervals or the number of sampled contexts; adding uncertainty estimates would help readers judge the reliability of the monotonic style increase with ρ.
  3. [§4.3, Table 6] The count metric is normalized by the value of the target style corpus, which makes the Rand baseline's value of 1.00 in Table 6 partly mechanical; a brief explanation of the normalization and its interpretational limits would improve clarity.
  4. [§4.5, §5.1] The inference depends on several fixed hyperparameters, including λ = 0.5 and σ² = 0.12; reporting sensitivity to λ and at least one alternative noise scale would clarify the robustness of the main result.
  5. [§5.2, Fig. 6] The MDS visualization is qualitative; labeling the axes or reporting a quantitative alignment measure, such as nearest-neighbor overlap between z_S2S(x), z_AE(y), and z_AE(s), would strengthen the claim that StyleFusion aligns the three latent spaces.

Circularity Check

2 steps flagged · score 5.0 of 10

Automatic evaluation is partially circular: the Pstyle classifiers used to rerank candidates (Eq 15) are the same classifiers that build Dtest and define the neural/ngram style-intensity metrics; independent count and human evidence partially mitigate.

  1. fitted input called prediction [Section 3.3, Eq. 15; Section 4.3, Table 6]
    "we rank the hypotheses considering both relevance and style intensity. score(hi) = (1−λ)P (hi|zS2S(x)) +λPstyle(hi) (15) ... Pstyle(hi) is the probability of hypothesis hi being targeted style predicted by pretrained classifiers. ... For style intensity evaluation, besides the neural and ngram classifier prediction (Section 3.3), we also use simple word-counting (hereafter count metric) to minimize model-specific effects."

    The neural and ngram style-intensity columns in Table 6 are computed from the same pretrained classifiers that supply Pstyle in the Eq. 15 reranker, which every trainable system uses to select one hypothesis from 100 candidates. A model that maximizes Pstyle at inference is therefore being scored by the very function it was selected to maximize, so its high neural/ngram style-intensity scores are partially forced by construction. The count metric and human style classification are independent and break the loop, which is why this is not fully circular.

  2. fitted input called prediction [Section 4.1, Task and datasets]
    "Dtest is the test set with stylized reference responses, constructed by filtering the Reddit dataset from year 2013 using the trained neural and ngram classifiers. For each context, there are at least 4 reference responses approximately in the targeted style (Pstyle> 0.3)."

    The BLEU references in Dtest are selected by the same Pstyle classifiers that define the Eq. 15 reranking objective and the neural/ngram style-intensity metrics. Consequently, BLEU is computed against responses whose stylizedness is determined by the classifier that the systems are tuned to satisfy, and a reranked hypothesis and a reference can agree because both passed the same decision boundary. This makes the automatic appropriateness comparison less independent, though the count metric and human evaluation are not affected by this particular loop.

full rationale

The main derivation of StyleFusion is not circular: the fusion and smoothness regularizers (Eqs. 1-12) are new objective terms, the latent-space alignment is visualized in Fig. 6, and the ablation in Table 6 shows that adding Lstyle increases style intensity. The self-citations to SPACEFUSION (Gao et al., 2019b) are not load-bearing in a circular way because the paper provides its own MDS visualization and compares empirically against +Lconv (SPACEFUSION) as a baseline. The circularity is in the automatic evaluation pipeline: Eq. 15 selects hypotheses by Pstyle, and the same classifiers provide the neural and ngram style-intensity metrics and construct Dtest for BLEU. Thus the automatic style-intensity advantage and part of the relevance comparison are partially by construction. The independent count metric and the human style-classification task do show a StyleFusion advantage, and human appropriateness is reported as not statistically different from MTask, but no error bars or test statistics are given, so the empirical claim is not fully secured. On balance, the paper has real independent content but its headline automatic results contain a partially circular evaluation loop; score 5.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The method rests on a few unproven assumptions about latent space geometry and the validity of learned style classifiers. The free parameters (ρ, λ, σ^2) are hand-chosen and not fitted, but the style-control claim depends on ρ. No new physical or conceptual entities are introduced.

free parameters (3)
  • Sampling radius ρ
    Controls distance from the conversation prediction in latent space at inference (Eq 14). Varied from 0 to 1 to modulate style intensity; is a user-specified knob rather than learned, and the style-control claim depends on it.
  • Ranking weight λ = 0.5
    Balances relevance and style probability in hypothesis selection (Eq 15). Chosen by hand; affects the reported style-intensity/relevance trade-off.
  • Noise variance σ^2 = 0.12
    Variance of Gaussian noise added in smoothness losses (Eq 8, 10) and used in normalization for ρ (Eq 14). Chosen by hand; expected to affect generation diversity and style smoothness.
assumptions (4)
  • domain assumption Euclidean distance and nearest-neighbor search capture semantic dissimilarity in the learned latent space.
    The fusion and spread-out losses (Eq 1-5) rest on the premise that Euclidean proximity in the latent space corresponds to semantic relatedness. If the geometry is not semantically organized, the alignment will not produce relevant stylized responses.
  • domain assumption Linear interpolation between latent codes yields outputs that gradually change from one content/style to another.
    The smoothness losses (Eq 8-11) assume that moving along a line in latent space produces semantically meaningful interpolations. This is a common but unproven property of neural latent spaces.
  • domain assumption The pretrained style classifiers (neural and ngram) are valid measures of the target style and suitable for filtering test references.
    These classifiers are used to construct Dtest (Section 4.1), to rank hypotheses at inference (Eq 15), and to evaluate style intensity (Section 4.3). If the classifiers are biased, the automatic evaluation and inference selection are biased.
  • ad hoc to paper Pretraining on Dconv then joint training avoids overfitting on small Dstyle.
    Stated in Section 3.2 as a heuristic; the paper does not analyze its necessity or sensitivity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Structuring Latent Spaces for Stylized Response Generation." pith.science (2026). https://pith.science/paper/CRVJF65P

@misc{pith2026190905361,
  author       = {Pith},
  title        = {Pith review of: Structuring Latent Spaces for Stylized Response Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CRVJF65P}},
  note         = {Machine review of arXiv:1909.05361}
}
read the original abstract

Generating responses in a targeted style is a useful yet challenging task, especially in the absence of parallel data. With limited data, existing methods tend to generate responses that are either less stylized or less context-relevant. We propose StyleFusion, which bridges conversation modeling and non-parallel style transfer by sharing a structured latent space. This structure allows the system to generate stylized relevant responses by sampling in the neighborhood of the conversation model prediction, and continuously control the style level. We demonstrate this method using dialogues from Reddit data and two sets of sentences with distinct styles (arXiv and Sherlock Holmes novels). Automatic and human evaluation show that, without sacrificing appropriateness, the system generates responses of the targeted style and outperforms competitive baselines.

Figures

Figures reproduced from arXiv: 1909.05361 by the authors.

Figure 1
Figure 1. STYLEFUSION helps conversational model to distill style from non-conversational, non-parallel sentences by mapping them to points surrounding the related conversations in the structured latent space. Di￾rection and distance from the model prediction (center black dot) roughly correspond to contents and style in￾tensity, respectively, illustrated by examples taken from [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. STYLEFUSION model architecture. ”non-parallel”). Our aim is to train a model jointly on Dstyle and Dconv to generate appropriate re￾sponses in the style similar to sentences from Dstyle, to a given context. The iven context may or may not be in the target style. 3.2 Training In contrast to SPACEFUSION(Gao et al., 2019b), which only fuses context-response pairs, our goal is to additionally map related stylized senten… view at source ↗
Figure 3
Figure 3. Change of the overall style intensity with [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Change of the styles at finer granularity of [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Relevancy of the STYLEFUSION outputs at different ρ as measured by BLEU4 with references of different styles trated in [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: MDS visualization of learned latent spaces. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 25 canonical work pages

  1. [1]

    URL: " 'urlintro :=

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

  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]

    Reina Akama, Kazuaki Inada, Naoya Inoue, Sosuke Kobayashi, and Kentaro Inui. 2017. Generating stylistically consistent dialog responses with transfer learning. In IJCNLP, pages 408--412

    work page 2017

  4. [4]

    Ingwer Borg and P Groenen. 2003. Modern multidimensional scaling: theory and applications. Journal of Educational Measurement, 40(3):277--280

    work page 2003

  5. [5]

    Kyunghyun Cho, Bart van Merrienboer, Dzmitry Bahdanau, and Yoshua Bengio. 2014. On the properties of neural machine translation: Encoder--decoder approaches. In SSST-8, pages 103--111

    work page 2014

  6. [6]

    Zhenxin Fu, Xiaoye Tan, Nanyun Peng, Dongyan Zhao, and Rui Yan. 2018. Style transfer in text: Exploration and evaluation. In Thirty-Second AAAI Conference on Artificial Intelligence

    work page 2018

  7. [7]

    Michel Galley, Chris Brockett, Xiang Gao, Jianfeng Gao, and Bill Dolan. 2019. Grounded response generation task at dstc7. In AAAI Dialog System Technology Challenges Workshop

    work page 2019

  8. [8]

    Jianfeng Gao, Michel Galley, and Lihong Li. 2019 a . Neural approaches to conversational ai. Foundations and Trends in Information Retrieval, 13(2-3):127--298

    work page 2019

Show all 32 references
  1. [9]

    Xiang Gao, Sungjin Lee, Yizhe Zhang, Chris Brockett, Michel Galley, Jianfeng Gao, and Bill Dolan. 2019 b . Jointly optimizing diversity and relevance in neural response generation. NAACL-HLT 2019

    2019

  2. [10]

    Hongyu Gong, Suma Bhat, Lingfei Wu, Jinjun Xiong, and Wen-mei Hwu. 2019. Reinforcement learning based text style transfer without parallel training corpus. arXiv preprint arXiv:1903.10671

    2019 arXiv

  3. [11]

    Zhiting Hu, Zichao Yang, Xiaodan Liang, Ruslan Salakhutdinov, and Eric P Xing. 2017. Toward controlled generation of text. In ICML, pages 1587--1596. JMLR. org

    2017

  4. [12]

    Diederik P Kingma and Jimmy Ba. 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980

    2014 arXiv

  5. [13]

    Jiwei Li, Michel Galley, Chris Brockett, Jianfeng Gao, and Bill Dolan. 2016 a . A diversity-promoting objective function for neural conversation models. In NAACL, pages 110--119

    2016

  6. [14]

    Jiwei Li, Michel Galley, Chris Brockett, Georgios Spithourakis, Jianfeng Gao, and Bill Dolan. 2016 b . A persona-based neural conversation model. In ACL, volume 1, pages 994--1003

    2016

  7. [15]

    Xiaodong Liu, Pengcheng He, Weizhu Chen, and Jianfeng Gao. 2019. Multi-task deep neural networks for natural language understanding. arXiv preprint arXiv:1901.11504

    2019 arXiv

  8. [16]

    Yi Luan, Chris Brockett, Bill Dolan, Jianfeng Gao, and Michel Galley. 2017. Multi-task learning for speaker-role adaptation in neural conversation models. In Proceedings of the Eighth International Joint Conference on Natural Language Processing (Volume 1: Long Papers), volume...

    2017

  9. [17]

    Yi Luan, Yangfeng Ji, Hannaneh Hajishirzi, and Boyang Li. 2016. Multiplicative representations for unsupervised semantic role induction. In ACL

    2016

  10. [18]

    Bryan McCann, Nitish Shirish Keskar, Caiming Xiong, and Richard Socher. 2018. The natural language decathlon: Multitask learning as question answering. arXiv preprint arXiv:1806.08730

    2018 arXiv

  11. [19]

    Kate G Niederhoffer and James W Pennebaker. 2002 a . Linguistic style matching in social interaction. Journal of Language and Social Psychology, 21(4):337--360

    2002

  12. [20]

    Kate G Niederhoffer and James W Pennebaker. 2002 b . Sharing one's story: On the benefits of writing or talking about emotional experience. Handbook of positive psychology

    2002

  13. [21]

    Tong Niu and Mohit Bansal. 2018. Polite dialogue generation without parallel data. Transactions of the Association of Computational Linguistics, 6:373--389

    2018

  14. [22]

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. BLEU : a method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting on association for computational linguistics, pages 311--318. Association for Computational Linguistics

    2002

  15. [23]

    Lianhui Qin, Michel Galley, Chris Brockett, Xiaodong Liu, Xiang Gao, Bill Dolan, Yejin Choi, and Jianfeng Gao. 2019. Conversing by reading: Contentful neural conversation with on-demand machine reading. arXiv preprint arXiv:1906.02738

    2019 arXiv

  16. [24]

    Tianxiao Shen, Tao Lei, Regina Barzilay, and Tommi Jaakkola. 2017. Style transfer from non-parallel text by cross-alignment. In NIPS, pages 6830--6841

    2017

  17. [25]

    Vighnesh Leonardo Shiv, Chris Quirk, Anshuman Suri, Xiang Gao, Khuram Shahid, Nithya Govindarajan, Yizhe Zhang, Jianfeng Gao, Michel Galley, Chris Brockett, et al. 2019. Microsoft icecaps: An open-source toolkit for conversation modeling. In Proceedings of the 57th Conference ...

    2019

  18. [26]

    Sutskever, O

    I. Sutskever, O. Vinyals, and Q. Le. 2014. Sequence to sequence learning with neural networks. In NIPS

    2014

  19. [27]

    Zichao Yang, Zhiting Hu, Chris Dyer, Eric P Xing, and Taylor Berg-Kirkpatrick. 2018. Unsupervised text style transfer using language models as discriminators. In Advances in Neural Information Processing Systems, pages 7287--7298

    2018

  20. [28]

    Yizhe Zhang, Michel Galley, Jianfeng Gao, Zhe Gan, Xiujun Li, Chris Brockett, and Bill Dolan. 2018. Generating informative and diverse conversational responses via adversarial information maximization. In Advances in Neural Information Processing Systems, pages 1813--1823

    2018

  21. [29]

    Yizhe Zhang, Xiang Gao, Sungjin Lee, Chris Brockett, Michel Galley, Jianfeng Gao, and Bill Dolan. 2019. Consistent dialogue generation with self-supervised feature learning. arXiv preprint arXiv:1903.05759

    2019 arXiv

  22. [30]

    Yizhe Zhang, Dinghan Shen, Guoyin Wang, Zhe Gan, Ricardo Henao, and Lawrence Carin. 2017. Deconvolutional paragraph representation learning. In Advances in Neural Information Processing Systems, pages 4169--4179

    2017

  23. [31]

    Tiancheng Zhao. 2019. Learning to Converse With Latent Actions. Ph.D. thesis, Carnegie Mellon

    2019

  24. [32]

    Li Zhou, Jianfeng Gao, Di Li, and Heung-Yeung Shum. 2018. The design and implementation of xiaoice, an empathetic social chatbot. arXiv preprint arXiv:1812.08989

    2018 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.