Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

PERL: Pinyin Enhanced Rephrasing Language Model for Chinese ASR N-best Error Correction

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A Pinyin-aware rephrasing model cuts Chinese ASR errors by up to 70%.

desk verdict Useful engineering on Chinese ASR N-best correction with a genuinely new Pinyin+length rephrasing combination, but the method section has a real dimension mismatch and ambiguity about training-time length supervision that must be fixed. read the letter →

arxiv 2412.03230 v3 pith:KCF5NZIS submitted 2024-12-04 cs.CL

Junhong Liang , Bojun Zhang This is my paper · ORCID
classification cs.CL
keywords ChineseASRerrorcorrectionPinyinembeddingN-besthypotheseslengthpredictionrephrasinglanguagemodelmaskedtokencharacterratedomain-specificbenchmark
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 paper proposes a post-correction pipeline for Chinese ASR called PERL, built for the N-best setting where several recognizer hypotheses are available. The central claim is that two auxiliary modules—a Pinyin encoder that turns each character into its romanized pronunciation, and a length predictor that estimates the true sentence length—let a masked rephrasing model fix phonetic confusions and length drift at the same time. On the Aishell-1 benchmark PERL lowers character error rate by 29.1% relative to the 1-best baseline, and on the authors' new domain-specific DoAD benchmark it lowers CER by roughly 70% across law, medical, and official-document domains, while adding only about 3 ms of inference time. If this holds, error correction for Mandarin ASR can be done with a small encoder-based model rather than slow generative LLMs.

What carries the argument

The load-bearing object is the gated fusion of two frozen or pretrained feature streams feeding a mask-prediction rephrasing model. A Pinyin encoder (token embedding plus a unidirectional GRU over the Pinyin sequence and a small Transformer stack) produces phonetic embeddings; a BERT encoder produces semantic embeddings for the masked N-best text. A token-wise gate $g_{s,i}, g_{p,i}$ computed by two MLPs from $E_s[i]$, $E_p[i]$, and the sentence mean embedding $\bar{e}_s$ blends the streams as $E_f = g_s \odot E_s + g_p \odot E_p$. The length predictor, a BERT encoder plus an MLP trained with cross-entropy against gold sentence lengths, supplies $\hat{l}$ mask tokens so the output length is budgeted rather than freely generated.

What would settle it

Train the full PERL pipeline twice: once with gold lengths and once with the frozen length predictor's predicted lengths for the mask budget, then compare CER on the DoAD test sets. If the predicted-length variant loses most of the reported 70% reduction, the central claim depends on an unavailable oracle signal; if the gap is small, the length predictor is doing the work attributed to it.

Watch

Extended reading notes

Core claim

PERL establishes that ASR N-best correction can be treated as constrained rephrasing: the model first predicts the length of the gold sentence, appends that many mask tokens to the concatenated N-best text, and predicts masked positions from a fused representation of semantics and Pinyin. The Pinyin representation comes from a frozen encoder pre-trained to map Pinyin sequences back to characters, and the fusion weights are computed per token from semantic, phonetic, and sentence-level semantic embeddings. The paper reports that this design outperforms generative LLMs such as GPT-4o, DeepSeek, and Qwen2.5, and also beats both oracle character selection and oracle hypothesis selection on the noisy DoAD benchmarks, which the authors attribute to the model exploring a broader semantic space than the fixed N-best candidates allow.

Load-bearing premise

The paper never says whether the rephrasing model is trained with the gold reference length or with the frozen length predictor's output; if training sees the gold length, the model receives a privileged signal that may not be available at inference time.

Editorial extensions

If this is right

  • Adding the Pinyin stream lowers CER on Aishell-1 from 4.78 to 4.10, showing that phonetic cues recover homophone errors the semantic encoder misses.
  • Removing the length predictor causes the largest degradation, raising Aishell-1 CER to 7.28, so constraining output length is the most load-bearing component of the pipeline.
  • The model maintains its advantage across recognizers of different sizes, cutting CER on DoAD for whisper-large-v3 and whisper-small as well as the main distil-whisper model.
  • The roughly 3 ms added latency means the correction step can sit inside real-time ASR post-processing rather than requiring a slow generative model.
  • On the noisy DoAD domains PERL's CER is lower than the oracle character-selection bound, implying context can supply characters absent from all N-best hypotheses.

Reading between the lines

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

  • A direct test of production risk: train the rephrasing model using the frozen length predictor's output instead of any gold length and compare CER; if the gap is large, the reported gains depend on a training-time oracle length signal.
  • The same frozen-encoder-plus-gated-fusion recipe could transfer to other languages whose writing maps to a romanization or syllabary, such as Cantonese Jyutping or Japanese kana, where phonetic confusions follow a similar pattern.
  • The observation that CER worsens at n=6 even though wrong-character coverage keeps rising points to the length predictor and long-input encoder as the bottleneck; an extension would be to feed ASR confidence scores or acoustic features into the length predictor.
  • Since PERL surpasses oracle character selection on DoAD, the oracle-char upper bound is too weak for noisy domains; reporting oracle-sentence or oracle-edit bounds would calibrate how much of the gain comes from exploring outside the candidate list.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes PERL, a three-stage pipeline for Chinese ASR N-best error correction. A Pinyin encoder is pretrained to produce phonetic embeddings, a length predictor estimates the target output length from the concatenated N-best hypotheses, and a rephrasing model fuses semantic and phonetic representations through token-wise gates while predicting masked tokens under a length budget. The authors introduce DoAD, a synthetic domain-specific N-best benchmark built from ECSpell text via Azure TTS and Whisper decoding. On CHP/Aishell-1 and DoAD, PERL is reported to reduce CER by 29.11% and 69-76% respectively relative to the 1-best baseline, with a 3.09 ms per-input latency. Ablations remove the length predictor, the Pinyin encoder, and the N-best input, and the paper includes additional evaluations across Whisper model sizes and a case study of phonetic correction.

Significance. If the reported results hold, PERL would be a practical and useful contribution: it directly addresses phonetic confusions in Chinese ASR and, more importantly, provides a mechanism for length-constrained correction, which is a known weakness of generative LLM baselines. The paper has several empirical strengths: clean ablations showing each component matters, length-prediction counts across three Whisper models, a multi-ASR evaluation, a latency comparison on identical hardware, and a concrete case study illustrating Pinyin-driven correction. However, the quantitative contribution is currently conditional: DoAD is synthetic and unreleased, no confidence intervals or significance tests accompany the headline CER differences, and the method description leaves a load-bearing dimension mismatch and a training/inference ambiguity unresolved. With precise specification and stronger evaluation reporting, the paper could be a solid systems contribution.

major comments (4)
  1. [Section 2.3] The fusion equations are underspecified because E_s and E_p have different lengths. Smasked is formed by appending l_hat [M] tokens to Sconcat, so Es = Enc_s(Smasked) has |Sconcat| + l_hat positions, while Ep = Enc_p(Sconcat) has |Sconcat| positions. The gates g_s,i = MLP_s(Es[i], Ep[i], ē_s) and g_p,i = MLP_p(Es[i], Ep[i], ē_s) require Ep[i] for every position i, including the appended mask positions that determine the output length. The paper does not state whether Ep is padded, truncated, or aligned via an additional mapping. If Ep is zero-padded, the appended positions receive no phonetic signal through the gate; if fusion is restricted to the first |Sconcat| positions, the appended masks are generated without phonetic information. Please specify the exact alignment and justify that the appended positions jointly use semantic and Pinyin representations, or revise the model so that both encoders operate on the same input sequence.
  2. [Section 2.3 / Section 3.2] The paper does not state whether the rephrasing model is trained with the frozen length predictor's predicted length l_hat or with the gold reference length. Section 2.2 trains the length predictor on gold lengths from the same Aishell-1 and DoAD training sets, so if the rephrasing model receives gold lengths during training and predicted lengths at inference, the reported CER gains are measured under a privileged training signal and may overstate production performance. Since length prediction is a central contribution and the ablation in Table 6 shows that removing it raises CER substantially, please specify the training-time procedure. If gold lengths were used during training, rerun the experiments with predicted lengths at both training and inference, or provide a train/inference mismatch analysis.
  3. [Section 4.1 / Table 2 / Section 3.1] The central CER comparisons are reported without error bars or significance tests, and several decisive differences are small: on CHP/Aishell-1 PERL (4.10) is only 0.12 lower than Qwen2.5 (4.22), and on DoAD-Law PERL (3.41) is within 0.42 of the Ocp oracle (3.83). Because DoAD is synthetic, constructed by the authors, and not released, an independent check of the 'around 70% CER reduction' claim is currently impossible. Please provide repeated-seed results with standard deviations or confidence intervals, and make the DoAD construction scripts and data available, or release a representative public subset, so the benchmark can be reproduced and the statistical reliability of the headline comparisons can be assessed.
  4. [Section 2.3 / Section 3.2] The supervision for appended mask tokens is unspecified. The input processing appends l_hat [M] tokens, and the loss is computed over masked positions M, but the paper never explains what the gold token is for an appended position when the reference sentence is shorter than |Sconcat| + l_hat, nor how positions in Smasked are aligned with the gold sentence. If the appended masks are meant to extend the output to length l_hat, the target at each appended position and the handling of truncation at the stated max_length of 128 need to be defined. Without this, the rephrasing objective at the length-determining positions cannot be implemented from the text alone.
minor comments (6)
  1. [Abstract / Figure 1] The abstract contains 'aPinyinEnhancedRephrasingLanguage' without spacing, and 'analyzes' should be 'analyses'; Figure 1's caption contains 'incoporates' instead of 'incorporates'.
  2. [Section 4.2] The subsection heading 'The choice ofn' should read 'The choice of n', and the sentence 'To assess the significance of the length prediction module's significance' is redundant and should be rephrased.
  3. [Table 4] Table 4 reports counts of equal-length predictions but not denominators; since the test set sizes are given in Table 1, please report rates or include denominators so the length-prediction accuracy can be compared across domains.
  4. [Table 5] The caption says colors encode 'wrong/correct/wrong' but lists 'red/blue/orange'; please align the color legend and make the table readable in grayscale.
  5. [Section 3.1] For the CHP/Aishell-1 experiments, the paper does not specify how the N-best lists are generated (ASR model, beam size, or whether they are taken directly from ChineseHP); include these details for reproducibility.
  6. [References] Several references are informal or incomplete, including [18] (a GitHub repository), [22] (a GitHub tool), and [23] (an arXiv preprint without venue); update them to archival versions where available.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: empirical CER comparisons are self-contained.

full rationale

The paper's central claims are empirical CER measurements on held-out test sets (Aishell-1 and DoAD), not derivations from the model's own inputs. The length predictor is a supervised estimator trained on gold reference lengths, and using it at inference is ordinary curve fitting rather than circular reasoning. The Pinyin encoder is pretrained on the external Wang271k corpus and frozen, so it does not encode the target benchmark result. The dynamic masking strategy is attributed to the external ReLM work [16], and ReLM is also used as a baseline; this is a comparison, not a self-citation chain. The two self-citations ([5] and [9]) appear only in the introduction as background and are not load-bearing for PERL's reported gains. Reproducibility concerns are present but are not circularity: Section 2.3's fusion equations are underspecified for appended mask tokens because E_p is shorter than E_s, and Section 3.2 does not state whether the rephrasing model is trained with gold or predicted lengths. These are clarity/implementation issues that could affect faithful reproduction, but they do not make the reported results equivalent to the training labels by construction.

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

The central claim is an empirical performance comparison. No fitted constants enter a derivation; the only hand-chosen numbers are hyperparameters (mask rate, N-best size, max length). The pipeline relies on three transfer assumptions: BERT for semantics, Pinyin encoder pretraining on an error-free corpus, and length predictor trained on the same domain distribution. DoAD's synthetic construction adds a representativeness assumption.

free parameters (3)
  • mask_rate = 0.2
    Hand-set in Section 3.2; controls the fraction of input tokens replaced with [M] and is not swept across values.
  • n_best_size = 5
    The number of ASR hypotheses concatenated at input (Section 3.2); Figure 3 suggests n=5 is a local optimum but it is not tuned per domain.
  • max_length = 128
    Cap on sequence length including appended mask tokens (Section 3.2); determines how many mask tokens can be budgeted for long outputs.
assumptions (4)
  • domain assumption BERT provides transferable semantic representations for noisy ASR text
    Section 2.3 uses a BERT encoder as the semantic backbone; no fine-tuning or validation of this transfer is reported.
  • domain assumption Pinyin encoder pretrained on Wang271k error-free text transfers to ASR N-best inputs
    Section 2.1 pretrains on a spelling-correction corpus; the domain shift to Whisper hypotheses with phonetic noise is not measured.
  • domain assumption Length predictor trained on gold reference lengths from Aishell-1 and DoAD training sets generalizes to their test sets and to other Whisper models
    Section 2.2 trains on the same domains evaluated; Table 4 shows length accuracy on test but with no confidence bounds.
  • domain assumption Synthetic TTS-plus-noise speech in DoAD is representative of real domain ASR errors
    Section 3.1 constructs DoAD with Azure TTS and injected noise; no real recordings are used.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PERL: Pinyin Enhanced Rephrasing Language Model for Chinese ASR N-best Error Correction." pith.science (2026). https://pith.science/paper/KCF5NZIS

@misc{pith2026241203230,
  author       = {Pith},
  title        = {Pith review of: PERL: Pinyin Enhanced Rephrasing Language Model for Chinese ASR N-best Error Correction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KCF5NZIS}},
  note         = {Machine review of arXiv:2412.03230}
}
abstract

Chinese ASR correction is challenging because errors are often \emph{phonetic} (many characters share similar Pinyin) while the correction model must also obey a \emph{length constraint} under noisy N-best hypotheses. Existing approaches either exploit Pinyin only at the prompt/feature level without integrating it into model representations or rely on generative decoding that can drift in length. We propose \textbf{PERL}, a \textbf{constrained rephrasing pipeline} for Chinese N-best ASR correction that (i) predicts the target length and enforces it via mask budgeting, and (ii) fuses \emph{semantic} and \emph{phonetic} (Pinyin) representations through token-wise gates conditioned on sentence semantics. Experiments on Aishell-1 and our new domain N-best benchmark \textbf{DoAD} show that PERL consistently reduces CER (29.11\% on Aishell-1 and up to $\sim$70\% on DoAD) while maintaining low latency. We also provide analyzes of length generalization and phonetic--semantic interactions, showing when PERL relies on phonetic cues versus semantic constraints.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. RAIR: Retrieval-Augmented Iterative Refinement for Chinese Spelling Correction

    cs.CL 2025-04 reject novelty 6.0 of 10

    RAIR combines a fine-tuned retriever, multi-turn length reflection, and adaptive selection to improve LLM-based Chinese spelling correction in equal-length and variable-length scenarios.

Reference graph

Works this paper leans on

33 extracted references · 32 canonical work pages · cited by 1 Pith paper

  1. [1]

    Despite advances in end-to-end models [1], ASR outputs remain vul- nerable to accents, background noise, and speaker variability

    INTRODUCTION Automatic Speech Recognition (ASR) has been widely adopted in applications such as voice interaction and information retrieval. Despite advances in end-to-end models [1], ASR outputs remain vul- nerable to accents, background noise, and speaker variability. These errors not only reduce recognition quality but also propagate to downstream task...

    work page

  2. [2]

    We propose PERL, which integrates semantic and phonetic embeddings via a neural weighting mechanism, supported by a pre-trained Pinyin encoder and an explicit length predictor

    work page

  3. [3]

    We construct DoAD, a domain-specific N-best ASR dataset covering legal, medical, and official document domains

    work page

  4. [4]

    Through extensive experiments, we demonstrate that PERL achieves the best CER among correction models while main- taining low inference latency, making it practical for real- world ASR applications

    work page

  5. [5]

    PERL: Pinyin Enhanced Rephrasing Language Model for Chinese ASR N-best Error Correction

    METHOD Our method consists of three main stages: 1) pretrain a Pinyin en- coder to capture phonological information (Section 2.1), 2) pretrain a length predictor to handle the variable-length issue in ASR correc- tion (Section 2.2), and 3) train a rephrasing model that integrates se- mantic and phonetic features for final correction (Section 2.3). Fig- ur...

    work page Pith review arXiv 2025

  6. [6]

    Data We conduct experiments on the ChineseHP/Aishell-1 (CHP/Aishell-

    EXPERIMENTAL SETUP 3.1. Data We conduct experiments on the ChineseHP/Aishell-1 (CHP/Aishell-

    work page

  7. [7]

    The con- struction process is summarized as follows: Data Processing:We start from ECSpell data [17], a text do- main error correction dataset

    dataset [6] and constructDomainASRDataset (DoAD). The con- struction process is summarized as follows: Data Processing:We start from ECSpell data [17], a text do- main error correction dataset. Each gold-standard sentence is seg- mented by punctuation and normalized to remove Arabic numerals and punctuation marks. Text-to-Speech:The normalized sentences a...

    work page

  8. [8]

    N-best ASR Correction We utilize the ChineseHP [6] dataset for Aishell-1 and perform our own ASR on distil-whisper[18]

    RESULTS 4.1. N-best ASR Correction We utilize the ChineseHP [6] dataset for Aishell-1 and perform our own ASR on distil-whisper[18]. Table 2 shows the N-best ASR cor- rection result, where CER represents the average character error rate, and CERR represents the percentage decrease in the average charac- ter error rate.o cp represents the best character co...

    work page 2024

Show all 33 references
  1. [9]

    To evaluate performance under chal- lenging conditions, we introduce the highly noisy DoAD dataset

    DISCUSSION & CONCLUSION PERL integrates a Pinyin module to reduce phonetic errors and a length predictor to address sequence inconsistencies in N-best Chi- nese ASR error correction. To evaluate performance under chal- lenging conditions, we introduce the highly noisy DoAD dat...

  2. [10]

    Robust speech recognition via large-scale weak supervision,

    Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever, “Robust speech recognition via large-scale weak supervision,” 2022

    2022

  3. [11]

    Read, listen, and see: Leveraging multimodal information helps chi- nese spell checking,

    Heng-Da Xu, Zhongli Li, Qingyu Zhou, Chao Li, Zizhen Wang, Yunbo Cao, Heyan Huang, and Xian-Ling Mao, “Read, listen, and see: Leveraging multimodal information helps chi- nese spell checking,” 2021

    2021

  4. [12]

    Improving Chinese spelling check by character pronunciation prediction: The effects of adaptivity and granularity,

    Jiahao Li, Quan Wang, Zhendong Mao, Junbo Guo, Yanyan Yang, and Yongdong Zhang, “Improving Chinese spelling check by character pronunciation prediction: The effects of adaptivity and granularity,” inProceedings of the 2022 Confer- ence on Empirical Methods in Natural Language ...

    2022

  5. [13]

    From spelling to grammar: A new framework for Chinese grammatical error correction,

    Xiuyu Wu and Yunfang Wu, “From spelling to grammar: A new framework for Chinese grammatical error correction,” inFindings of the Association for Computational Linguistics: EMNLP 2022, Yoav Goldberg, Zornitsa Kozareva, and Yue Zhang, Eds., Abu Dhabi, United Arab Emirates, Dec. ...

    2022

  6. [14]

    3883– 3890, ECAI, 10 2024

    Junhong Liang, Zhu Junnan, Feifei Zhai, Nanchang Cheng, Chengqing Zong, and Yu Zhou,A Hybrid Approach towards Chinese Spelling and Splitting Error Correction, pp. 3883– 3890, ECAI, 10 2024

    2024

  7. [15]

    Pinyin regularization in error correction for chinese speech recognition with large language models,

    Zhiyuan Tang, Dong Wang, Shen Huang, and Shidong Shang, “Pinyin regularization in error correction for chinese speech recognition with large language models,” inInterspeech 2024. pp. 1910–1914, ISCA

    2024

  8. [16]

    Mitigating catastrophic forgetting in multi-domain chinese spelling correction by multi-stage knowledge transfer framework,

    Peng Xing, Yinghui Li, Shirong Ma, Xinnian Liang, Haojing Huang, Yangning Li, Hai-Tao Zheng, Wenhao Jiang, and Ying Shen, “Mitigating catastrophic forgetting in multi-domain chinese spelling correction by multi-stage knowledge transfer framework,” 2024

    2024

  9. [17]

    C-llm: Learn to check chinese spelling errors character by character,

    Kunting Li, Yong Hu, Liang He, Fandong Meng, and Jie Zhou, “C-llm: Learn to check chinese spelling errors character by character,” 2024

    2024

  10. [18]

    Rair: Retrieval-augmented iter- ative refinement for chinese spelling correction,

    Junhong Liang and Yu Zhou, “Rair: Retrieval-augmented iter- ative refinement for chinese spelling correction,” 2025

    2025

  11. [19]

    On the (in)effectiveness of large language models for chinese text correction,

    Yinghui Li, Haojing Huang, Shirong Ma, Yong Jiang, Yangn- ing Li, Feng Zhou, Hai-Tao Zheng, and Qingyu Zhou, “On the (in)effectiveness of large language models for chinese text correction,” 2023

    2023

  12. [20]

    MuCGEC: a multi- reference multi-source evaluation dataset for chinese grammat- ical error correction,

    Yue Zhang, Zhenghua Li, Zuyi Bao, Jiacheng Li, Bo Zhang, Chen Li, Fei Huang, and Min Zhang, “MuCGEC: a multi- reference multi-source evaluation dataset for chinese grammat- ical error correction,” inProceedings of the 2022 Conference of the North American Chapter of the Associ...

    2022

  13. [21]

    Spelling error correction with soft-masked bert,

    Shaohua Zhang, Haoran Huang, Jicong Liu, and Hang Li, “Spelling error correction with soft-masked bert,” 2020

    2020

  14. [22]

    N-best t5: Robust asr error correction using multiple input hy- potheses and constrained decoding space,

    Rao Ma, Mark J. F. Gales, Kate M. Knill, and Mengjie Qian, “N-best t5: Robust asr error correction using multiple input hy- potheses and constrained decoding space,” inINTERSPEECH

  15. [23]

    Exploring the limits of transfer learning with a unified text-to-text transformer,

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Pe- ter J. Liu, “Exploring the limits of transfer learning with a unified text-to-text transformer,” 2023

    2023

  16. [24]

    FastCorrect 2: Fast error correction on mul- tiple candidates for automatic speech recognition,

    Yichong Leng, Xu Tan, Rui Wang, Linchen Zhu, Jin Xu, Wen- jie Liu, Linquan Liu, Xiang-Yang Li, Tao Qin, Edward Lin, and Tie-Yan Liu, “FastCorrect 2: Fast error correction on mul- tiple candidates for automatic speech recognition,” inFind- ings of the Association for Computatio...

    2021

  17. [25]

    Multi-stage large language model correction for speech recognition,

    Jie Pu, Thai-Son Nguyen, and Sebastian Stuker, “Multi-stage large language model correction for speech recognition,” 2024

    2024

  18. [26]

    Chinese spelling correction as rephrasing language model,

    Linfeng Liu, Hongqiu Wu, and Hai Zhao, “Chinese spelling correction as rephrasing language model,” 2024

    2024

  19. [27]

    General and domain adaptive chinese spelling check with error consistent pretraining,

    Qi Lv, Ziqiang Cao, Lei Geng, Chunhui Ai, Xu Yan, and Guo- hong Fu, “General and domain adaptive chinese spelling check with error consistent pretraining,” vol. 22, no. 5, pp. 1–18

  20. [28]

    Belle: Be everyone’s large language model engine,

    BELLEGroup, “Belle: Be everyone’s large language model engine,”https://github.com/LianjiaTech/ BELLE, 2023

    2023

  21. [29]

    A hybrid approach to automatic corpus generation for Chinese spelling check,

    Dingmin Wang, Yan Song, Jing Li, Jialong Han, and Haisong Zhang, “A hybrid approach to automatic corpus generation for Chinese spelling check,” inProceedings of the 2018 Confer- ence on Empirical Methods in Natural Language Processing, Ellen Riloff, David Chiang, Julia Hockenm...

    2018

  22. [30]

    Bert: Pre-training of deep bidirectional transformers for language understanding,

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova, “Bert: Pre-training of deep bidirectional transformers for language understanding,”arXiv preprint arXiv:1810.04805, 2018

    2018 arXiv

  23. [31]

    Revisiting pre-trained models for Chinese natural language processing,

    Yiming Cui, Wanxiang Che, Ting Liu, Bing Qin, Shijin Wang, and Guoping Hu, “Revisiting pre-trained models for Chinese natural language processing,” inProceedings of the 2020 Con- ference on Empirical Methods in Natural Language Process- ing: Findings, Online, Nov. 2020, pp. 65...

    2020

  24. [32]

    pycorrector: Text error correction tool,

    Xu Ming, “pycorrector: Text error correction tool,” 2021

    2021

  25. [2023]

    3267–3271, ISCA

    August 2023, Interspeech 2023, pp. 3267–3271, ISCA

    2023

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.