Pith. sign in

REVIEW 4 major objections 5 minor 31 references

Interpretable Mnemonic Generation for Kanji Learning via Expectation-Maximization

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read An EM algorithm that learns shared mnemonic-construction rules plus learner and kanji traits beats supervised fine-tuning at matching real learner-authored kanji mnemonics, and does so transparently.

desk verdict A genuinely new EM-style framework for learning interpretable mnemonic rules, but the interpretability evidence is weakened by a circular E-step and needs external validation. read the letter →

arxiv 2507.05137 v2 pith:RURFXHAH submitted 2025-07-07 cs.CL cs.AI

classification cs.CLcs.AI
keywords kanjilearningmnemonicgenerationexpectation-maximizationlatentvariablemodelitemresponsetheorycold-startinterpretableNLPlearnermodeling
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

The paper sets out to show that mnemonic generation for kanji can be modeled as a structured, interpretable process rather than a black-box language-model call. It claims that learners draw on a small set of common composition rules, that each learner leans on some rules more than others, and that each kanji makes certain rules more natural. The authors fit this structure with an Expectation-Maximization-type algorithm that alternates between guessing which rules produced each learner's mnemonic and updating the rules, learner affinities, and kanji compatibilities. In a cold-start setting, where a new learner has no history, the resulting mnemonics are preferred over supervised fine-tuning by an LLM judge with a 58.1% win rate, and they score higher on semantic similarity. If right, this gives language-learning tools a way to personalize mnemonics while showing learners the strategy behind each cue.

What carries the argument

The load-bearing object is an Expectation-Maximization loop over a latent rule-activation model, built on a one-parameter logistic item response theory model. It sets $p(z_{ijk}=1)=\sigma(h_{jk}+g_{ik})$, with learner affinity $h_{jk}$ and kanji compatibility $g_{ik}$ as the latent traits. The E-step is the distinctive part: instead of inferring rules by clustering text, it computes per-token likelihoods of the observed learner mnemonic under each candidate rule using a trainable open-weights LM1, and sets $z_{ijk}$ to the top-3 most likely rules. The M-step then fits $h$ and $g$ by binary cross-entropy to those activations, regenerates the rule descriptions by prompting LM2 with the top-8 mnemonics for each rule while enforcing orthogonality, and fine-tunes LM1 so it generates conditioned on the active rules. The loop repeats until validation loss stops improving; at test time, a new learner's activations are computed from population-mean affinities $\bar{h}_k$ plus per-kanji $g_{ik}$.

What would settle it

Have learners mark which of the ten learned rules they actually used for each mnemonic; if the E-step's top-3 likelihood picks agree no better than chance, the interpretable rules are not the ones driving authoring.

Watch

Extended reading notes

Core claim

The central claim is that explicitly modeling the latent rules behind mnemonic construction improves cold-start mnemonic generation over direct fine-tuning. The paper's generative story is that a mnemonic for kanji $i$ by learner $j$ is produced by activating a small set of $K$ shared rules, with activation governed by a learner-rule affinity $h_{jk}$ and a kanji-rule compatibility $g_{ik}$ through $p(z_{ijk}=1)=\sigma(h_{jk}+g_{ik})$. The EM algorithm alternates an E-step that uses the token-level likelihoods of an open-weights generator to assign each learner's mnemonic to its most likely top-3 rules and an M-step that refits the traits, refreshes the rule descriptions through a stronger proprietary LLM, and fine-tunes the generator conditioned on the activated rules. On the Koohii Kanji dataset, this system reaches a 58.1% win rate over supervised fine-tuning under an LLM judge and BERTScore 0.240 versus 0.212, while also returning a readable list of ten rules such as transformation, cause-and-effect, idioms, and anthropomorphism. The claim is therefore not just that generation improves, but that the improvement is carried by structures that can be inspected, clustered, and reused.

Load-bearing premise

The method assumes that the likelihood ranking of an open-weights language model over candidate rules correctly identifies the rules a learner actually used when writing a mnemonic; if that ranking is wrong, the EM loop fits learner and kanji traits to noise and the discovered rules are arbitrary.

Editorial extensions

If this is right

  • In the cold-start setting, EM-generated mnemonics beat supervised fine-tuning in an LLM-judge comparison by 58.1% to 41.9% and improve BERTScore from 0.212 to 0.240.
  • The learned rules can be repurposed as prompts: for example, the common-knowledge rule translates into the question 'What everyday saying does this bring to mind?', which can elicit a learner's own associations.
  • Rule activation for unseen learners is computable from population-average learner affinities plus kanji compatibilities, so the model can generate without retraining per learner.
  • Clustering the learned traits reveals three learner archetypes and four kanji types, giving a route to group-level personalization.
  • The EM loop needs only $I\times J\times K + K\times T$ calls to the large proprietary model rather than full fine-tuning, keeping the iterative procedure tractable.

Reading between the lines

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

  • Editorial extension: the same latent-rule machinery could transfer to other compositional memorization tasks, such as medical terminology or scientific definitions, whenever learners combine fixed components with a small set of rhetorical strategies.
  • Editorial extension: human annotation of which rules were actually used would directly test whether the E-step's top-3 likelihoods recover genuine psychological rules, a claim the paper's automated evaluation does not settle.
  • Editorial extension: the fixed $K=10$ rule count could itself be tuned; a validation-driven choice of $K$ might change both win rate and the interpretability of the rule set.
  • Editorial extension: because the method estimates a per-learner affinity vector, a natural next step is to warm-start $h_j$ from a few newly authored mnemonics and measure how quickly generation quality converges toward the oracle in-context learning performance.
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

4 major / 5 minor

Summary. The paper proposes a generative framework for Japanese kanji mnemonic generation in which mnemonic construction is explicitly modeled through a set of latent rules and per-learner/kanji traits. The authors introduce an EM-type algorithm that alternates between (E-step) assigning rules to learner-authored mnemonics using likelihoods from a fine-tuned small language model (LM1), (M-step) fitting a 1-PL IRT model over the resulting hard assignments and updating the rule set via a proprietary LLM (LM2), and finally fine-tuning LM1 on the updated rule-conditioned data. Experiments on the Koohii Kanji dataset report that the method outperforms an SFT baseline in the cold-start setting (BERTScore 0.240 vs. 0.212, and a 58.1% win rate judged by an LLM), and qualitative analyses of the learned rules and clustering are presented as evidence of interpretability.

Significance. If the central claims hold, the paper would make a useful contribution to personalized language-learning technology and to the interpretability of LLM-driven educational content generation. The idea of learning latent mnemonic-construction rules from learner-authored data, rather than relying on purely black-box prompting, is timely, and the use of an open-weight model for likelihood-based rule assignment is a sensible engineering choice. The manuscript also ships a concrete dataset split and an algorithm description that would allow replication. However, the significance is currently tempered by the lack of external validation of the learned rules, the statistical fragility of the reported performance gains, and the self-referential nature of the interpretability evidence.

major comments (4)
  1. [§3.2.1, Eq. (2)] The E-step computes p_{ijk} = P_LM1(m_{ij} | b_i, r_k), but LM1 is fine-tuned in the previous iteration on exactly these rules {r_k} and the same mnemonics. Therefore p_{ijk} is not independent evidence for which rule generated m_{ij}; the loop can converge to a self-consistent yet arbitrary partition of mnemonics into rule-conditioned groups. The paper's central interpretability claim rests on the identifiability of these z_{ijk} assignments, which is not established. I recommend an external validation, such as human annotations of rule usage on a held-out mnemonic set, or a synthetic-data experiment where the true rule assignments are known and the EM procedure is shown to recover them.
  2. [§5.1, Table 2] The reported gains over SFT are small point estimates: BERTScore improves from 0.212 to 0.240, ROUGE-1 from 0.249 to 0.251, and LUAR barely changes (0.482 vs 0.487). No error bars, multiple seeds, or significance tests are reported for any metric, including the 58.1% LLM-judged win rate. Given that the central claim of 'modest but consistent gains' is built on these numbers, the authors should report confidence intervals, bootstrap tests, or at least multiple random seeds with variance.
  3. [§5.1, compliance rate] The overall compliance rate of 60.58% is computed by prompting GPT-4o to check whether generated mnemonics adhere to rules, but the rules themselves were generated and updated by GPT-4o (LM2). This is a self-referential evaluation: the judge belongs to the same model family that produced the rule definitions. The authors' Limitations section properly notes that automated evaluation may not capture subjective quality, but for the specific claim of rule interpretability, a human evaluation of rule compliance or at least a different judge model trained on human rule annotations is needed to break the circularity.
  4. [§3.2.1, Algorithm 1] The procedure is called an 'EM-type algorithm', but it is heuristic in a way that matters: the E-step performs hard top-3 assignments rather than posterior expectations, and the M-step includes a prompt-based rule update by an external LLM that is not derived from any single objective function. There is no convergence guarantee or error analysis, and early stopping is based on validation loss rather than on the likelihood of the latent-variable model. I do not require a formal EM convergence proof, but the authors should either soften the terminology or provide an objective that the iterative procedure can be interpreted as optimizing, and they should report sensitivity to the initialization and to the top-3/top-8 heuristics.
minor comments (5)
  1. [§4.1, Table 1] The column header 'A vg. M/L' appears to be a typo for 'Avg. M/L'; please correct it.
  2. [§5.3] Several sentences contain grammatical errors that impede readability, e.g., 'we can easily putting leader to either one of the animal' and 'The ground-truth also uses the Star wars references'. A careful proofreading pass is needed.
  3. [§5.2] The cluster analysis reports three learner clusters of sizes 80, 1711, and 72, and four kanji clusters, but the PCA plots (Figures 2 and 3) are not described in enough detail to understand how the clusters relate to the latent dimensions. Please clarify the axes and the number of principal components used.
  4. [§C.2, Table 7] The table header 'Latent Cluster Rules' is ambiguous; it would be clearer to label the rows as 'learner clusters' and 'kanji clusters' explicitly.
  5. [§5.1] The sentence 'According to Prometheus evaluation, using LLM-as-judge, EM is preferred over SFT with a win rate of 58.1% vs. 41.9%' is redundant; the preceding sentence already introduces the Prometheus evaluation. Please streamline.

Circularity Check

2 steps flagged · score 6.0 of 10

The EM loop's rule assignments are self-confirming: LM1 is fine-tuned on the same z it is used to infer, and the interpretability evidence is judged by the same model (GPT-4o) that generates the rules; the cold-start mnemonic predictions themselves are held-out and not circular.

  1. self definitional [Section 3.2.1, Eq. 2 and Algorithm 1 (E-step and M-step fine-tuning)]
    "pijk ← PLM1(mij | bi, rk), ∀k. (2) ... Finally, we fine-tune LM 1 using the rule activations zijk determined in the E-step with the newly updated {rk}, along with the metadata for each kanji, bi. Specifically, we maximize the log-likelihood of all mnemonics according to Eq. 2, where we replace rk with the set of all rules relevant to the mnemonic, {rk}k∈Tij. ... We then loop back to the E-step, recomputing likelihoods pk under the newly updated LM 1."

    The E-step defines the latent rule activation z_ijk as the top-3 rules by LM1's likelihood PLM1(m_ij | b_i, r_k). But in the immediately preceding M-step, LM1 was fine-tuned to maximize the likelihood of the very same mnemonics under exactly those z-assigned rules. Thus high p_ijk for the assigned rules is partly manufactured by training, not independent evidence that rule r_k generated m_ij. The M-step then fits h_jk and g_ik to these self-confirming assignments, and LM2 updates rules from the top-8 examples under p_ijk. The loop can converge to any self-consistent labeling; without external rule annotations, the learned 'interpretable rules' and learner/kanji trait patterns are not identifiable from the data.

  2. other [Section 3.2.1 (Algorithm 1 rule update) and Section 5.1 (compliance evaluation)]
    "rk ← LM2(Ek, {rk′}k′̸=k) ... We prompt GPT-4o to evaluate whether the generated mnemonics adhere to the learned rules by identifying which activated rules are correctly applied. Compliance for each mnemonic is calculated by dividing the number of satisfied rules by the total number of applicable rules. Averaging these values across all examples yields an overall compliance rate of 60.58%, indicating that while the EM-learned rules capture useful and generalizable patterns..."

    The same proprietary model, GPT-4o (called LM2), both generates the rules (rk ← LM2(Ek, ...)) and later judges whether generated mnemonics adhere to those rules. The 60.58% compliance rate is therefore a measure of GPT-4o's consistency with its own rule definitions, not an independent validation that the rules reflect genuine learner-authored mnemonic mechanisms. No human annotations or external rule-identification benchmark are provided, so the paper's interpretability evidence is self-referential. This directly supports the claim that 'interpretable rules' are discovered, making the central interpretability contribution circular even though the cold-start generation metrics are held out.

full rationale

The cold-start mnemonic generation evaluation (BERTScore, ROUGE, LUAR, and the Prometheus win rate) is based on held-out learners and kanji and is not circular: those numbers compare generated mnemonics to actual learner-authored mnemonics. However, the paper's other headline contribution—'interpretable rules' and 'insight into the mechanisms behind effective mnemonic creation'—rests on a self-confirming loop. The E-step assigns each mnemonic to its top-3 rules using LM1 likelihoods, but LM1 is fine-tuned in the same iteration to maximize likelihood under exactly those rule assignments, so the assignment is a fixed point of the algorithm rather than an externally anchored measurement. The subsequent IRT fits (h_jk, g_ik) and rule updates via LM2 inherit this circularity. In addition, the compliance rate used to argue that the learned rules are meaningful is computed by GPT-4o, the same model that generated the rules. These two issues mean the interpretability 'discoveries' reduce by construction to the model's own likelihoods and judgments. The paper's Limitations section does acknowledge that mnemonic quality is evaluated by an LLM rather than human raters, but it does not address the self-referential nature of the rule-compliance check. Because the generation results are independent, the circularity is partial, not total—hence a score of 6 rather than 8 or 10.

Assumptions & free parameters 4 free parameters · 5 assumptions · 2 invented entities

The central claim rests on several assumptions: rule-driven mnemonic construction, the 1-PL IRT activation model, the reliability of LM1 likelihoods for rule assignment, and the validity of LLM-based evaluation. The latent rules are invented constructs generated by GPT-4o without external validation.

free parameters (4)
  • K (number of rules) = 10
    Number of latent rules; chosen empirically (Section 4.3) and ablated in Section 5.4; performance varies little with K, so it is a free hyperparameter.
  • Top-3 rule selection cutoff = 3
    Instead of a likelihood threshold, the top-3 rules per mnemonic are always selected in the E-step (Section 3.2.1). The choice is justified by robustness but is a hand-set hyperparameter.
  • Top-8 mnemonic set size for rule updates = 8
    The rule update step selects the top-8 mnemonics per rule (Section 3.2.1); the choice is justified only by falling within the number of rules used.
  • Learner trait h_jk and kanji compatibility g_ik = learned vectors
    Latent IRT parameters fitted to the top-3 rule assignments during the M-step; the central model depends on them.
assumptions (5)
  • domain assumption Mnemonics are generated by a small set of common rules shared across learners.
    The generative framework assumes rule-driven construction (Sections 1 and 3.1).
  • domain assumption Rule activation for a mnemonic follows the 1-PL IRT model p = sigmoid(h_jk + g_ik).
    Section 3.1 Eq. 1; there is no empirical validation that additive learner and kanji effects adequately describe rule usage.
  • ad hoc to paper GPT-4o can produce accurate and orthogonal rules from example mnemonics.
    Rule initialization and update rely on LM2 (GPT-4o) prompts (Sections 3.3 and Algorithm 1); no external validation of rule quality is provided.
  • domain assumption Learner preferences are stable and can be represented by population-level averages for new learners.
    Section 3.2.2 uses the mean learner trait as a proxy; the paper's own Limitations section notes behavior may evolve over time.
  • domain assumption LLM-based metrics (BERTScore, ROUGE, LUAR, LLM-as-judge) are valid proxies for mnemonic quality.
    Section 4.2 and Limitations admit no human raters are used.
invented entities (2)
  • Latent mnemonic rules r_k
    purpose: To provide an interpretable set of compositional strategies that the generator must follow.
    The rules are generated and refined by GPT-4o from training mnemonics; compliance is evaluated by the same type of model (LM2/GPT-4o), so there is no independent evidence that these rules reflect real cognitive mechanisms.
  • Learner affinity h_jk and kanji compatibility g_ik latent traits
    purpose: To explain and predict rule activation per learner and kanji.
    These are latent variables inferred only from the same dataset and are not validated against external measures of learner style.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Interpretable Mnemonic Generation for Kanji Learning via Expectation-Maximization." pith.science (2026). https://pith.science/paper/RURFXHAH

@misc{pith2026250705137,
  author       = {Pith},
  title        = {Pith review of: Interpretable Mnemonic Generation for Kanji Learning via Expectation-Maximization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RURFXHAH}},
  note         = {Machine review of arXiv:2507.05137}
}
read the original abstract

Learning Japanese vocabulary is a challenge for learners from Roman alphabet backgrounds due to script differences. Japanese combines syllabaries like hiragana with kanji, which are logographic characters of Chinese origin. Kanji are also complicated due to their complexity and volume. Keyword mnemonics are a common strategy to aid memorization, often using the compositional structure of kanji to form vivid associations. Despite recent efforts to use large language models (LLMs) to assist learners, existing methods for LLM-based keyword mnemonic generation function as a black box, offering limited interpretability. We propose a generative framework that explicitly models the mnemonic construction process as driven by a set of common rules, and learn them using a novel Expectation-Maximization-type algorithm. Trained on learner-authored mnemonics from an online platform, our method learns latent structures and compositional rules, enabling interpretable and systematic mnemonics generation. Experiments show that our method performs well in the cold-start setting for new learners while providing insight into the mechanisms behind effective mnemonic creation.

Figures

Figures reproduced from arXiv: 2507.05137 by the authors.

Figure 1
Figure 1. An example of an online platform Koohii Kanji (2024) for sharing mnemonics is kanji 休. A learner can type their mnemonic and share it with others. leverage the compositional structure of kanji; for example, the character 休 (rest) is composed of 人 (person) and 木 (tree), prompting the mnemonic “a person resting by a tree,” and learning plat￾forms support this approach by assigning keywords, sometimes based on visual r… view at source ↗
Figure 2
Figure 2. PCA plot for three learner clusters. To identify a representative learner or kanji for each cluster, we select the member whose rule us￾age vector is closest to the cluster centroid in the Euclidean space. Learners are grouped into clus￾ters, each with its own centroid, and for each clus￾ter, we select a representative learner as the one whose rule vector is nearest to that cluster’s cen￾troid. Similarly, kanji are … view at source ↗
Figure 3
Figure 3. PCA plot for four kanji clusters. by examining the top three rules they most often use. First, the learner lean on common knowledge (Rule 3): for the kanji 團 “group” (keywords pent in + glued), their mnemonic is “A group has to stick together,” and for the kanji 届 “deliver” (keywords flag + sprout), their mnemonic mirrors the familiar quest-flag on game maps. Next, they use casual id￾ioms (Rule 8): for the kanji 怠 “… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Training and test heatmaps: the first two plots show the individual components [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 19 canonical work pages

  1. [1]

    Richard C Atkinson and Michael R Raugh. 1975. An application of the mnemonic keyword method to the acquisition of a russian vocabulary. Journal of experimental psychology: Human learning and memory, 1(2):126

  2. [2]

    A SMART Mnemonic Sounds like "Glue Tonic": Mixing LLMs with Student Feedback to Make Mnemonic Learning Stick

    Nishant Balepur, Matthew Shu, Alexander Hoyle, Alison Robey, Shi Feng, Seraphina Goldfarb-Tarrant, and Jordan Boyd-Graber. 2024. A smart mnemonic sounds like" glue tonic": Mixing llms with student feedback to make mnemonic learning stick. arXiv preprint arXiv:2406.15352

  3. [3]

    Sumanth Doddapaneni, Krishna Sayana, Ambarish Jash, Sukhdeep Sodhi, and Dima Kuzmin. 2024. https://aclanthology.org/2024.personalize-1.12/ User embedding model for personalized language prompting . In Proceedings of the 1st Workshop on Personalization of Generative AI Systems (PERSONALIZE 2024), pages 124--131, St. Julians, Malta. Association for Computat...

  4. [4]

    Michael E Everson. 2011. Best practices in teaching logographic and non-roman writing systems to l2 learners. Annual Review of Applied Linguistics, 31:249--274

  5. [5]

    James W Heisig. 2011. Remembering the kanji 1: A complete course on how not to forget the meaning and writing of Japanese characters. University of Hawaii Press

  6. [6]

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685

  7. [7]

    Kanshudo. 2024. https://www.kanshudo.com Kanshudo: The fastest and most enjoyable way to learn japanese . Accessed: 2024-12-22

  8. [8]

    Seungone Kim, Juyoung Suk, Shayne Longpre, Bill Yuchen Lin, Jamin Shin, Sean Welleck, Graham Neubig, Moontae Lee, Kyungjae Lee, and Minjoon Seo. 2024. https://arxiv.org/abs/2405.01535 Prometheus 2: An open source language model specialized in evaluating other language models . Preprint, arXiv:2405.01535

Show all 31 references
  1. [9]

    Koohii Kanji . 2024. Koohii kanji. https://kanji.koohii.com/

  2. [10]

    Jaewook Lee and Andrew Lan. 2023. Smartphone: Exploring keyword mnemonic with auto-generated verbal and visual cues. In International Conference on Artificial Intelligence in Education, pages 16--27. Springer

  3. [11]

    Jaewook Lee, Hunter McNichols, and Andrew Lan. 2024. Exploring automated keyword mnemonics generation with large language models via overgenerate-and-rank. arXiv preprint arXiv:2409.13952

  4. [12]

    Chin-Yew Lin. 2004. Rouge: A package for automatic evaluation of summaries. In Text summarization branches out, pages 74--81

  5. [13]

    Meta . 2024. Llama-3.2 3b. https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct. Accessed: 2025-05-16

  6. [14]

    Andrew A Neath and Joseph E Cavanaugh. 2012. The bayesian information criterion: background, derivation, and applications. Wiley Interdisciplinary Reviews: Computational Statistics, 4(2):199--203

  7. [15]

    Lin Ning, Luyang Liu, Jiaxing Wu, Neo Wu, Devora Berlowitz, Sushant Prakash, Bradley Green, Shawn O'Banion, and Jun Xie. 2024. User-llm: Efficient llm contextualization with user embeddings. arXiv preprint arXiv:2402.13598

  8. [16]

    OpenAI. 2024. https://openai.com/index/hello-gpt-4o/ Hello gpt-4o . Accessed: 2025-02-19

  9. [17]

    OpenAI. 2025. Openai moderation api. https://platform.openai.com/docs/guides/moderation. Accessed: 2025-05-15

  10. [18]

    Jiaxing Qi, Zhongzhi Luan, Shaohan Huang, Carol Fung, Hailong Yang, and Depei Qian. 2024. Fdlora: personalized federated learning of large language model via dual lora tuning. arXiv preprint arXiv:2406.07925

  11. [19]

    Georg Rasch. 1993. Probabilistic models for some intelligence and attainment tests. ERIC

  12. [20]

    Alireza Salemi, Sheshera Mysore, Michael Bendersky, and Hamed Zamani. 2024. https://doi.org/10.18653/v1/2024.acl-long.399 L a MP : When large language models meet personalization . In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volu...

  13. [21]

    Sumuk Shashidhar, Abhinav Chinta, Vaibhav Sahai, and Dilek Hakkani Tur. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.200 Unsupervised human preference learning . In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 3412--3445, M...

  14. [22]

    Rivera Soto, Olivia Miano, Juanita Ordonez, Barry Chen, Aleem Khan, Marcus Bishop, and Nicholas Andrews

    Rafael A. Rivera Soto, Olivia Miano, Juanita Ordonez, Barry Chen, Aleem Khan, Marcus Bishop, and Nicholas Andrews. 2021. Learning universal authorship representations. In EMNLP

  15. [23]

    Zhaoxuan Tan, Zheyuan Liu, and Meng Jiang. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.371 Personalized pieces: Efficient personalized large language models through collaborative efforts . In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Pro...

  16. [24]

    WaniKani. 2024. https://www.wanikani.com Wanikani . Accessed: 2024-12-22

  17. [25]

    Haiyang Yu, Jingye Chen, Bin Li, and Xiangyang Xue. 2024. Chinese character recognition with radical-structured stroke trees. Machine Learning, 113(6):3807--3827

  18. [26]

    Saizheng Zhang, Emily Dinan, Jack Urbanek, Arthur Szlam, Douwe Kiela, and Jason Weston. 2018. https://doi.org/10.18653/v1/P18-1205 Personalizing dialogue agents: I have a dog, do you have pets too? In Proceedings of the 56th Annual Meeting of the Association for Computational ...

  19. [27]

    Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q Weinberger, and Yoav Artzi. 2019. Bertscore: Evaluating text generation with bert. arXiv preprint arXiv:1904.09675

  20. [28]

    You Zhang, Jin Wang, Liang-Chih Yu, Dan Xu, and Xuejie Zhang. 2024. Personalized lora for human-centered text understanding. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 19588--19596

  21. [29]

    Wanjun Zhong, Duyu Tang, Jiahai Wang, Jian Yin, and Nan Duan. 2021. Useradapter: Few-shot user learning in sentiment analysis. In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 1484--1488

  22. [30]

    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...

  23. [31]

    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 gl...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.