Pith. sign in

REVIEW 5 major objections 5 minor 53 references

QueEn: A Large Language Model for Quechua-English Translation

T0 review · 5 major / 5 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 retrieval-augmented, low-rank fine-tuned LLM lifts Quechua-English translation beyond zero-shot baselines, the authors report.

desk verdict The abstract claims BLEU 17.6 for a RAG+LoRA model, but Table 1 never tests that model and the best BLEU is 0.235; the paper's central result is unsupported. read the letter →

arxiv 2412.05184 v1 pith:OOWHM7WZ submitted 2024-12-06 cs.CL cs.AI

Junhao Chen , Peng Shu , Yiwei Li , Huaqin Zhao , Hanqi Jiang , Yi Pan , Yifan Zhou , Zhengliang Liu
show 2 more authors
Lewis C Howe Tianming Liu
This is my paper · ORCID
classification cs.CLcs.AI
keywords Quechuamachinetranslationretrieval-augmentedgenerationparameter-efficientfine-tuningLoRAlow-resourcelanguagesendangeredBLEU
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 argues that combining retrieval-augmented generation (RAG) with Low-Rank Adaptation (LoRA) fine-tuning turns a large language model into a markedly better Quechua-English translator without full retraining. The authors build QueEn, which retrieves dictionary entries and grammar snippets for each query and feeds them into a fine-tuned LLM, and report that RAG-augmented models beat zero-shot GPT-4o and LLaMA 3.1 405B on BLEU, ROUGE, and BERTScore. The method matters because Quechua has only about 14,000 parallel sentence pairs, so any approach that exploits existing dictionaries and grammar guides could extend machine translation to many endangered languages.

What carries the argument

The load-bearing mechanism is the retrieval-augmented generation pipeline with a dual index. For each query $q$, the system retrieves exact matches $D_q^{\text{keyword}} = \{d \in D \mid q \cap \operatorname{Keywords}(d) \neq \emptyset\}$ and semantic matches $D_q^{\text{vector}}$ from the top $K$ documents by cosine similarity, then combines them as $D_q = D_q^{\text{keyword}} \cup D_q^{\text{vector}}$ and builds a prompt $P$ that includes the retrieved lexical and grammatical content. Low-Rank Adaptation (LoRA) is the parameter-efficient fine-tuning component: it approximates the weight update as $\Delta W = A \cdot B$ with $A \in \mathbb{R}^{d \times r}$, $B \in \mathbb{R}^{r \times k}$, and $r \ll \min(d,k)$, so only the small matrices $A$ and $B$ are trained. This lets a general-purpose LLM access explicit Quechua dictionary and grammar knowledge at inference time while keeping adaptation cheap.

What would settle it

Recompute BLEU, ROUGE, and BERTScore on the Siminchik test set using human-translated English references instead of the machine-generated ones; if the RAG + LoRA models no longer beat zero-shot baselines, the central claim fails. Also verify whether the abstract's BLEU of 17.6 can be reproduced from the released test set, since Table 1 reports 0.235 for the best system.

Watch

Extended reading notes

Core claim

The central claim is that retrieval and low-rank adaptation are complementary for low-resource translation: retrieval supplies lexical and grammatical knowledge the model lacks, while LoRA adapts the model's behaviour with few trainable parameters. The system indexes a Quechua-English dictionary and grammar guide using both keyword matching and embedding similarity, then constructs a prompt $P = \operatorname{ConstructPrompt}(q, D_q)$ from the retrieved set and generates a translation $R = \operatorname{LLM}(P)$. In the paper's experiments, GPT + RAG reaches the highest scores on all three metrics (BLEU 0.235, ROUGE 0.278, BERTScore 0.963), and the authors interpret the gains over plain GPT and LLaMA as evidence that the retrieval mechanism compensates for Quechua's agglutinative morphology and scarce data.

Load-bearing premise

The evaluation assumes that ChatGPT-4's Spanish-to-English translations, generated in Section 4.1, are accurate enough to serve as ground-truth references for Quechua-English translation; if those references are wrong, the reported scores and rankings are not trustworthy.

Editorial extensions

If this is right

  • If the paper's claim holds, the same RAG + LoRA recipe should improve LLM translation for other low-resource languages that have dictionaries and grammar guides but little parallel text.
  • The dual retrieval mechanism (exact keyword plus semantic embedding) should help the model handle Quechua's agglutinative morphology by surfacing relevant morpheme-level dictionary entries.
  • Because BERTScore stays high even when BLEU is low, the results suggest retrieval mainly improves lexical selection rather than overall semantic content, pointing to where future effort should focus.
  • Parameter-efficient fine-tuning keeps adaptation feasible for communities with limited computational resources, which is a concrete step toward language preservation tools.

Reading between the lines

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

  • The evaluation references are machine-translated from Spanish to English by a GPT model; if those translations carry systematic bias, the reported ranking is relative to an artificial target, and a human reference set could change it.
  • The abstract's BLEU of 17.6 is an order of magnitude above the best table value (0.235); reconciling this discrepancy is necessary before the headline number can be taken as the method's true performance.
  • A natural transfer test would apply the same pipeline to another endangered language with a dictionary, such as Aymara or Guaraní, to see whether the gains come from the method or from Quechua-specific properties.
  • The pipeline could be extended with morpheme-aware tokenization, since Quechua suffixes carry grammatical roles that dictionary retrieval alone may not disambiguate.
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

5 major / 5 minor

Summary. The paper proposes QueEn, a system for Quechua-English translation that combines Retrieval-Augmented Generation (RAG) with Low-Rank Adaptation (LoRA) fine-tuning. The authors use the Siminchik corpus, generate English references from Spanish translations with ChatGPT-4, and compare GPT-4o and LLaMA variants with and without RAG and fine-tuning. The abstract claims a BLEU score of 17.6 versus 1.5 for standard GPT models, but Table 1 reports BLEU scores between 0.038 and 0.235, and no experimental row instantiates the combined RAG+LoRA configuration described in Section 3. The manuscript therefore does not currently support its central claim.

Significance. Quechua-English translation is an important and underexplored low-resource task, and the combination of RAG with parameter-efficient fine-tuning is a reasonable direction to investigate. If the claimed results were supported, the paper would be a useful contribution to low-resource MT. However, the current manuscript has no machine-checked proofs, no released code or data, and, critically, does not evaluate the proposed combined model. The significance of the contribution is therefore not yet established.

major comments (5)
  1. [Abstract and Table 1] The abstract's central quantitative claim, "a BLEU score of 17.6 compared to 1.5 for standard GPT models," is not supported by Table 1. The table reports BLEU scores of 0.235 for GPT + RAG and 0.096 for plain GPT, with no double-digit BLEU anywhere. The abstract and the experimental table describe different results, and the paper gives no explanation for the discrepancy.
  2. [Section 3 and Table 1] The proposed QueEn framework is defined in Section 3 as combining LoRA fine-tuning with RAG, yet Table 1 contains no row for the combined RAG+LoRA configuration. The rows are LLaMA, GPT, LLaMA + finetune, LLaMA + RAG, and GPT + RAG. The central claim that the integration of RAG with fine-tuning "substantially exceeds" baselines is therefore never tested by the reported experiments. A row labeled "ours" that instantiates the full method must be added, or the paper must clearly state which existing row corresponds to QueEn.
  3. [Section 4.1] The experimental setup conflates translation directions. The dataset is described as Quechua-Spanish ("que spa clean"), and Section 4.1 states that ChatGPT-4 is used to translate Spanish into English, but the following sentence says this "enables us to evaluate English-to-Quechua translation scenarios." The title and abstract promise Quechua-English translation. The paper must specify the actual source and target language in the evaluation, and the metric computation must match that direction.
  4. [Section 4.1 and Table 1] The reference translations are machine-generated: ChatGPT-4 produced the English references from Spanish, and GPT-based models are then scored against those references. This creates a circularity risk because the reference may share systematic stylistic and lexical biases with the models being evaluated. The authors should validate the ChatGPT-4 references against human translations or an independent reference set, or report inter-annotator agreement.
  5. [Section 4.2 and Section 4.3] Section 4.2 says a "combination of automated and human evaluation metrics" was employed, but no human evaluation setup, protocol, or results appear anywhere in the paper. Additionally, the experimental section omits load-bearing details needed for reproducibility: the specific base model versions (e.g., which LLaMA and GPT variants), LoRA rank r and other LoRA hyperparameters, training hyperparameters, retrieval top-K, embedding model used in RAG, and the train/validation/test split sizes. These omissions prevent verification of the reported results.
minor comments (5)
  1. [Section 4.1] There is a typo in Section 4.1: "e applied ChatGPT-4" should read "We applied ChatGPT-4."
  2. [Section 4.2] The metric name is rendered as "BER TScore" in Section 4.2; it should be "BERTScore."
  3. [Table 1] Table 1 uses inconsistent capitalization, e.g., "LLama" for LLaMA and "Rouge Score" for ROUGE; please standardize the notation.
  4. [Section 2.5] In Section 2.5, the text refers to "the QueEn project" and cites reference [43], but the cited work by Ortega et al. is about neural machine translation with a polysynthetic language, not the QueEn project. This citation should be corrected or removed.
  5. [Figure 1] Figure 1's caption says the retrieved information is "processed by a GPT-4 model," but the proposed method in Section 3 is described as a general LLM enhanced with LoRA; clarify which model the figure depicts.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant derivation circularity; the experimental reporting has consistency problems, but the central claims do not reduce to their inputs by construction.

full rationale

The paper's formal content is limited to standard building blocks: Section 3.1 restates the LoRA low-rank update W' = W + A·B, and Section 3.2 defines a standard RAG pipeline ending in R = LLM(ConstructPrompt(q, D_q)). Neither of these defines or forces any BLEU, ROUGE, or BERTScore outcome, and no parameter is fitted to the test data and then renamed as a prediction. The only self-citation is reference [3] (Shu et al., with overlapping authorship), used to support the background claim that LLMs perform poorly zero-shot on low-resource languages; the paper's own Table 1 independently shows near-zero zero-shot BLEU scores, so this citation is not load-bearing. The abstract's BLEU claim of 17.6 vs 1.5 matches no entry in Table 1, and the combined RAG+LoRA configuration named QueEn is never evaluated as a separate row; those are serious reporting and completeness flaws, but they are not circular reasoning. The ChatGPT-4-generated English references could bias scores in favor of GPT-family systems if the intended task is Quechua-to-English, but this is a reference-construction validity threat rather than a by-construction identity between outputs and references. Accordingly, no specific circular step meets the reduction standard imposed here, and the score reflects only the minor, non-load-bearing self-citation.

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

The central claim rests on several unvalidated assumptions about the evaluation data and the effectiveness of generic RAG and LoRA components. No explicit free parameters with fitted values are reported; the free parameters listed are unknown hyperparameters that the paper does not disclose.

free parameters (4)
  • LoRA rank r
    LoRA rank is a hyperparameter that controls the trade-off between efficiency and capacity; the paper does not report the value used.
  • Top K retrieved documents
    The number of retrieved entries in the vector-based retrieval (Section 3.2) is never specified.
  • Number of ChatGPT-4-generated references per sentence
    Section 4.1 says 'multiple English translations' were generated, but not how many, which directly affects BLEU and ROUGE scores.
  • Siminchik subset split ratio
    The paper says the data is divided into validation and test sets but does not report the size of each split.
assumptions (4)
  • domain assumption ChatGPT-4 produces accurate Spanish-to-English translations that can serve as ground-truth references for Quechua-English translation.
    Section 4.1 relies on this to construct evaluation references; if the translations are biased or inaccurate, all reported metrics are compromised.
  • domain assumption The Siminchik 'que spa clean' subset is representative of Quechua and correctly aligned with the Spanish and English transcriptions.
    The paper uses this subset without describing preprocessing, cleaning, or quality checks on the alignments.
  • domain assumption Retrieving dictionary and grammar entries via keyword and embedding search improves translation quality.
    The mechanism is described generically in Section 3.2 without evidence that the retrieval content is actually used in the reported experiment.
  • domain assumption BLEU, ROUGE, and BERTScore are valid evaluation metrics for Quechua, a polysynthetic low-resource language.
    The paper does not discuss the limitations of these metrics for agglutinative languages with rich morphology.

how reviews work

0 comments
Cite this review

Pith. "Pith review of QueEn: A Large Language Model for Quechua-English Translation." pith.science (2026). https://pith.science/paper/OOWHM7WZ

@misc{pith2026241205184,
  author       = {Pith},
  title        = {Pith review of: QueEn: A Large Language Model for Quechua-English Translation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OOWHM7WZ}},
  note         = {Machine review of arXiv:2412.05184}
}
read the original abstract

Recent studies show that large language models (LLMs) are powerful tools for working with natural language, bringing advances in many areas of computational linguistics. However, these models face challenges when applied to low-resource languages due to limited training data and difficulty in understanding cultural nuances. In this paper, we propose QueEn, a novel approach for Quechua-English translation that combines Retrieval-Augmented Generation (RAG) with parameter-efficient fine-tuning techniques. Our method leverages external linguistic resources through RAG and uses Low-Rank Adaptation (LoRA) for efficient model adaptation. Experimental results show that our approach substantially exceeds baseline models, with a BLEU score of 17.6 compared to 1.5 for standard GPT models. The integration of RAG with fine-tuning allows our system to address the challenges of low-resource language translation while maintaining computational efficiency. This work contributes to the broader goal of preserving endangered languages through advanced language technologies.

Figures

Figures reproduced from arXiv: 2412.05184 by the authors.

Figure 1
Figure 1. Illustrating a retrieval-augmented generation (RAG) architecture: Documents are indexed [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 31 canonical work pages

  1. [1]

    OracleSage: Towards Unified Visual-Linguistic Understanding of Oracle Bone Scripts through Cross-Modal Knowledge Fusion

    Jiang H, Pan Y, Chen J, et al. OracleSage: Towards Unified Visual-Linguistic Understanding of Oracle Bone Scripts through Cross-Modal Knowledge Fusion. arXiv preprint arXiv:2411.17837 2024

    arXiv 2024

  2. [2]

    Towards Next-Generation Medical Agent: How o1 is Reshaping Decision-Making in Medical Scenarios

    Xu S, Zhou Y, Liu Z, et al. Towards Next-Generation Medical Agent: How o1 is Reshaping Decision-Making in Medical Scenarios. arXiv preprint arXiv:2411.14461 2024

    arXiv 2024

  3. [3]

    Transcending Language Boundaries: Harnessing LLMs for Low- Resource Language Translation

    Shu P, Chen J, Liu Z, et al. Transcending Language Boundaries: Harnessing LLMs for Low- Resource Language Translation. arXiv preprint arXiv:2411.11295 2024

    arXiv 2024

  4. [4]

    Legal Evalutions and Challenges of Large Language Models

    Wang J, Zhao H, Yang Z, et al. Legal Evalutions and Challenges of Large Language Models. arXiv preprint arXiv:2411.10137 2024

    arXiv 2024

  5. [5]

    EchoFM: Foundation Model for Generalizable Echocardiogram Analysis

    Kim S, Jin P, Song S, et al. EchoFM: Foundation Model for Generalizable Echocardiogram Analysis. arXiv preprint arXiv:2410.23413 2024

    arXiv 2024

  6. [6]

    Robot Control via Natural Instructions Empowered by Large Language Model

    Wu Z, Shu P, Li Y, Li Q, Liu T, and Li X. Robot Control via Natural Instructions Empowered by Large Language Model. In: Discovering the Frontiers of Human-Robot Interaction: Insights and Innovations in Collaboration, Communication, and Control. Springer, 2024:437–57

    work page 2024

  7. [7]

    3D-CT-GPT: Generating 3D Radiology Reports through Inte- gration of Large Vision-Language Models

    Chen H, Zhao W, Li Y, et al. 3D-CT-GPT: Generating 3D Radiology Reports through Inte- gration of Large Vision-Language Models. arXiv preprint arXiv:2409.19330 2024

    arXiv 2024

  8. [8]

    LLMs for Coding and Robotics Education

    Shu P, Zhao H, Jiang H, et al. LLMs for Coding and Robotics Education. arXiv preprint arXiv:2402.06116 2024

    arXiv 2024

Show all 53 references
  1. [9]

    A complete survey on llm-based ai chatbots

    Dam SK, Hong CS, Qiao Y, and Zhang C. A complete survey on llm-based ai chatbots. arXiv preprint arXiv:2406.16937 2024

    2024 arXiv

  2. [10]

    Improving image captioning descriptiveness by ranking and llm-based fusion

    Bianco S, Celona L, Donzella M, and Napoletano P. Improving image captioning descriptiveness by ranking and llm-based fusion. arXiv preprint arXiv:2306.11593 2023

    2023

  3. [11]

    From images to textual prompts: Zero-shot visual question answering with frozen large language models

    Guo J, Li J, Li D, et al. From images to textual prompts: Zero-shot visual question answering with frozen large language models. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2023:10867–77

    2023

  4. [12]

    Eg-spikeformer: Eye-gaze guided transformer on spiking neural networks for medical image analysis

    Pan Y, Jiang H, Chen J, et al. Eg-spikeformer: Eye-gaze guided transformer on spiking neural networks for medical image analysis. arXiv preprint arXiv:2410.09674 2024

    2024 arXiv

  5. [13]

    Do large language models speak all languages equally? a comparative study in low-resource settings

    Hasan MA, Tarannum P, Dey K, Razzak I, and Naseem U. Do large language models speak all languages equally? a comparative study in low-resource settings. arXiv preprint arXiv:2408.02237 2024

    2024 arXiv

  6. [14]

    adaptmllm: Fine-tuning multilingual language models on low-resource languages with integrated llm playgrounds

    Lankford S, Afli H, and Way A. adaptmllm: Fine-tuning multilingual language models on low-resource languages with integrated llm playgrounds. Information 2023;14:638

    2023

  7. [15]

    LLMs Are Few-Shot In-Context Low-Resource Lan- guage Learners

    Cahyawijaya S, Lovenia H, and Fung P. LLMs Are Few-Shot In-Context Low-Resource Lan- guage Learners. arXiv preprint arXiv:2403.16512 2024

    2024 arXiv

  8. [16]

    Endangered Languages

    Sallabank J and Austin PK. Endangered Languages. The Routledge Handbook of Applied Linguistics 2022:362–73. 14

    2022

  9. [17]

    Rule-based machine translation

    Shiwen Y and Xiaojing B. Rule-based machine translation. In: Routledge encyclopedia of trans- lation technology. Routledge, 2014:186–200

    2014

  10. [18]

    The Georgetown-IBM experiment demonstrated in January 1954

    Hutchins WJ. The Georgetown-IBM experiment demonstrated in January 1954. In: Conference of the Association for Machine Translation in the Americas. Springer. 2004:102–14

    1954

  11. [19]

    A framework for interactive and automatic refinement of transfer-based machine translation

    Llitj´ os AF, Carbonell JG, and Lavie A. A framework for interactive and automatic refinement of transfer-based machine translation. In: Proceedings of the 10th EAMT Conference: Practical applications of machine translation. 2005

    2005

  12. [20]

    Interlingual machine translation

    Richens RH. Interlingual machine translation. The Computer Journal 1958;1:144–7

    1958

  13. [21]

    Statistical machine translation

    Lopez A. Statistical machine translation. ACM Computing Surveys (CSUR) 2008;40:1–49

    2008

  14. [22]

    The mathematics of statistical machine translation: Parameter estimation

    Brown PF, Della Pietra SA, Della Pietra VJ, and Mercer RL. The mathematics of statistical machine translation: Parameter estimation. Computational linguistics 1993;19:263–311

    1993

  15. [23]

    Statistical machine translation: IBM models 1 and 2

    Collins M. Statistical machine translation: IBM models 1 and 2. Columbia Columbia Univ 2011

    2011

  16. [24]

    Introducing a translation dictionary into phrase-based SMT

    Okuma H, Yamamoto H, and Sumita E. Introducing a translation dictionary into phrase-based SMT. IEICE transactions on information and systems 2008;91:2051–7

    2008

  17. [25]

    A hierarchical phrase-based model for statistical machine translation

    Chiang D. A hierarchical phrase-based model for statistical machine translation. In: Proceedings of the 43rd annual meeting of the association for computational linguistics (acl’05). 2005:263– 70

    2005

  18. [26]

    Neural machine translation: A review

    Stahlberg F. Neural machine translation: A review. Journal of Artificial Intelligence Research 2020;69:343–418

    2020

  19. [27]

    A context-aware recurrent encoder for neural machine translation

    Zhang B, Xiong D, Su J, and Duan H. A context-aware recurrent encoder for neural machine translation. IEEE/ACM Transactions on Audio, Speech, and Language Processing 2017;25:2424– 32

    2017

  20. [28]

    Neural machine translation by jointly learning to align and translate

    Bahdanau D. Neural machine translation by jointly learning to align and translate. arXiv preprint arXiv:1409.0473 2014

    2014 arXiv

  21. [29]

    Attention is all you need

    Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need. In: Advances in neural information processing systems. Vol. 30. 2017:5998–6008

    2017

  22. [30]

    Opennmt: Open-source toolkit for neural machine translation

    Klein G, Kim Y, Deng Y, Senellart J, and Rush AM. Opennmt: Open-source toolkit for neural machine translation. arXiv preprint arXiv:1701.02810 2017

    2017 arXiv

  23. [31]

    Marian: Fast neural machine transla- tion in C++

    Junczys-Dowmunt M, Grundkiewicz R, Dwojak T, et al. Marian: Fast neural machine transla- tion in C++. arXiv preprint arXiv:1804.00344 2018

    2018 arXiv

  24. [32]

    Google’s neural machine translation system: Bridging the gap between human and machine translation

    Wu Y, Schuster M, Chen Z, et al. Google’s neural machine translation system: Bridging the gap between human and machine translation. arXiv preprint arXiv:1609.08144 2016

    2016 arXiv

  25. [33]

    Quechua language shift, maintenance, and revitaliza- tion in the Andes: The case for language planning

    Hornberger NH and Coronel-Molina SM. Quechua language shift, maintenance, and revitaliza- tion in the Andes: The case for language planning. 2004

    2004

  26. [34]

    Reversing quechua language shift in South America

    Hornberger NH and King KA. Reversing quechua language shift in South America. Multilingual Matters 2001:166–94. 15

    2001

  27. [35]

    Morphology in Quechuan languages

    Adelaar WF. Morphology in Quechuan languages. In: Oxford Research Encyclopedia of Lin- guistics. 2020

    2020

  28. [36]

    Semantics and pragmatics of evidentials in Cuzco Quechua

    Faller MT. Semantics and pragmatics of evidentials in Cuzco Quechua. stanford university, 2002

    2002

  29. [37]

    Language models are few-shot learners

    Brown TB. Language models are few-shot learners. arXiv preprint arXiv:2005.14165 2020

    2005 arXiv

  30. [38]

    Gpt-4 technical report

    Achiam J, Adler S, Agarwal S, et al. Gpt-4 technical report. arXiv preprint arXiv:2303.08774 2023

    2023 arXiv

  31. [39]

    Llama: Open and efficient foundation language models

    Touvron H, Lavril T, Izacard G, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971 2023

    2023 arXiv

  32. [40]

    Retrieval-augmented generation for knowledge-intensive nlp tasks

    Lewis P, Perez E, Piktus A, et al. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in Neural Information Processing Systems 2020;33:9459–74

    2020

  33. [41]

    Review of large vision models and visual prompt engineering

    Wang J, Liu Z, Zhao L, et al. Review of large vision models and visual prompt engineering. Meta-Radiology 2023:100047

    2023

  34. [42]

    Mergen: The First Manchu-Korean Machine Translation Model Trained on Augmented Data

    Seo J, Byun S, Kang M, and Lee S. Mergen: The First Manchu-Korean Machine Translation Model Trained on Augmented Data. arXiv preprint arXiv:2311.17492 2023

    2023 arXiv

  35. [43]

    Neural machine translation with a polysynthetic low resource language

    Ortega JE, Castro Mamani R, and Cho K. Neural machine translation with a polysynthetic low resource language. Machine Translation 2020;34:325–46

    2020

  36. [44]

    Poincar \’e glove: Hyperbolic word embeddings

    Tifrea A, B´ ecigneul G, and Ganea OE. Poincar \’e glove: Hyperbolic word embeddings. arXiv preprint arXiv:1810.06546 2018

    2018 arXiv

  37. [45]

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

    Kenton JDMWC and Toutanova LK. Bert: Pre-training of deep bidirectional transformers for language understanding. In: Proceedings of naacL-HLT. Vol. 1. Minneapolis, Minnesota. 2019:2

    2019

  38. [46]

    Hire a linguist!: Learning endangered languages in LLMs with in-context linguistic descriptions

    Zhang K, Choi Y, Song Z, He T, Wang WY, and Li L. Hire a linguist!: Learning endangered languages in LLMs with in-context linguistic descriptions. In: Findings of the Association for Computational Linguistics ACL 2024. 2024:15654–69

    2024

  39. [47]

    Full parameter fine-tuning for large language models with limited resources

    Lv K, Yang Y, Liu T, Gao Q, Guo Q, and Qiu X. Full parameter fine-tuning for large language models with limited resources. arXiv preprint arXiv:2306.09782 2023

    2023 arXiv

  40. [48]

    Parameter-efficient fine-tuning of large-scale pre-trained language models

    Ding N, Qin Y, Yang G, et al. Parameter-efficient fine-tuning of large-scale pre-trained language models. Nature Machine Intelligence 2023;5:220–35

    2023

  41. [49]

    Lora: Low-rank adaptation of large language models

    Hu EJ, Shen Y, Wallis P, et al. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685 2021

    2021 arXiv

  42. [50]

    Siminchik: A Speech Corpus for the Quechua Language

    Cardenas R et al. Siminchik: A Speech Corpus for the Quechua Language. In: Proceedings of the 1st Workshop on Collaboration and Computing for Under-Resourced Languages. 2018:28–35. url: http://lrec-conf.org/workshops/lrec2018/W14/pdf/book_of_proceedings.pdf# page=28

    2018

  43. [51]

    Recall-oriented understudy for gisting evaluation (rouge)

    Lin C. Recall-oriented understudy for gisting evaluation (rouge). Retrieved August 2005;20:2005. 16

    2005

  44. [52]

    Bleu: a method for automatic evaluation of machine translation

    Papineni K, Roukos S, Ward T, and Zhu WJ. Bleu: a method for automatic evaluation of machine translation. In: Proceedings of the 40th annual meeting of the Association for Compu- tational Linguistics. 2002:311–8

    2002

  45. [53]

    Bertscore: Evaluating text generation with bert

    Zhang T, Kishore V, Wu F, Weinberger KQ, and Artzi Y. Bertscore: Evaluating text generation with bert. arXiv preprint arXiv:1904.09675 2019. 17

    1904 arXiv

Pith tools

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