Pith. sign in

REVIEW 5 major objections 6 minor 1 cited by

Adaptive Two-Phase Finetuning LLMs for Japanese Legal Text Retrieval

T0 review · 5 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 This paper claims that a two-phase fine-tuning pipeline, in which a LLaMA-based bi-encoder is first trained on broad positives and in-batch negatives and then on query-specific hard negatives, outperforms existing sparse, dense, and…

desk verdict A useful but statistically fragile extension of RepLLaMA: the two-phase recipe and Gemini-generated Japanese legal dataset are worth a look, but the reported gains need error bars and a clean label audit. read the letter →

arxiv 2412.13205 v1 pith:44VTQY4R submitted 2024-12-03 cs.IR cs.CLcs.LG

Quang Hoang Trung , Nguyen Van Hoang Phuc , Le Trung Hoang , Quang Huu Hieu , Vo Nguyen Le Duy This is my paper · ORCID
classification cs.IRcs.CLcs.LG
keywords Japaneselegaltextretrievaltwo-phasefine-tuningLLMbi-encoderhardnegativeminingdensedatasetMSMARCOensemble
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 is trying to establish that large language models can be made to retrieve Japanese legal texts well by fine-tuning them in two deliberate stages rather than one. In the first stage the model learns broad context from positive documents, BM25+ negatives, and in-batch easy negatives; in the second it is pushed to distinguish the query's own top-ranked non-relevant documents, called hard negatives, from true positives. The authors support the claim with a new Japanese legal corpus, 743 statute articles and roughly 3.5 thousand contract chunks labeled with relevant articles, and report that the pipeline beats sparse, dense, and generative baselines, including RepLLaMA, on recall and ranking metrics. They further report that the same pipeline transfers to a split of MS MARCO, suggesting the recipe is not Japanese-specific. A sympathetic reader would care because it offers a parameter-efficient path to building retrievers for low-resource legal domains without expert annotation at scale.

What carries the argument

The load-bearing mechanism is contrastive training of a LLaMA-2 bi-encoder whose query and document vectors are the hidden states of an appended <EOS> token, optimized by InfoNCE loss. Phase 1 assembles a globalDataset of true positives, top-BM25+ negatives, and in-batch negatives from other queries, teaching broad generalization; Phase 2 builds a domainSpecificDataset containing only positives and hard negatives mined by the Phase 1 model itself, with batch size 1 so each query competes only against its own near-miss documents. LoRA and 16-bit quantization make the 7-billion-parameter model trainable on two 24GB GPUs, and an ensemble variant adds BM25+ lexical scores and multiple checkpoints with learned weights.

What would settle it

Have several licensed Japanese labor-law attorneys independently re-annotate the 130 test queries against the 743-article corpus, blind to the model outputs; if inter-annotator agreement is low or the corrected labels shift, recompute Table 4 and check whether the margin over RepLLaMA survives. A cheaper supplementary test is to evaluate only on the 302 human-authored contract chunks, excluding the 3,172 Gemini-generated sentences, to see whether the phase-2 gains depend on generated training data.

Watch

Extended reading notes

Core claim

The central claim, stated on the paper's own terms, is that a two-phase fine-tuning schedule built on RepLLaMA's LLaMA-based bi-encoder yields better retrieval than any of the compared single-phase, sparse, dense, or generative methods. On the Japanese legal test set of 130 queries, the single two-phase model reaches Recall@10 of 76.87 and MRR@10 of 82.54, against 73.73 and 81.58 for RepLLaMA; adding BM25+ and a second checkpoint to the ensemble raises Recall@10 to 84.12. On a split of MS MARCO, the same two-phase approach reaches Recall@10 of 80.71 versus 77.55 for RepLLaMA. The paper also claims its dataset, built in Japanese with Gemini 1.5 Pro and manual verification rather than translated, is itself a contribution for future research in non-English legal retrieval.

Load-bearing premise

The dataset's relevance labels are accurate enough to support the reported margins: the labels were produced by Gemini 1.5 Pro and then manually verified, but the paper reports no legal experts, no annotator counts, and no agreement statistics, and the test set has only 130 queries.

Editorial extensions

If this is right

  • If the two-phase claim holds, the same pipeline should transfer to other non-English legal systems, since the only Japanese-specific components are the corpus and the BM25+ tokenizer settings.
  • Phase 2's batch-size-1 hard-negative deepening is credited with the gains over single-phase RepLLaMA; removing it should lower precision at small k on the Japanese test set.
  • Combining BM25+ with fine-tuned LLM embeddings in an ensemble gives the best published numbers on this corpus (Recall@10 84.12), implying lexical and semantic signals are complementary rather than redundant.
  • The dataset itself, 743 statute articles and 3,259 labeled training rows, gives the community a reusable non-English legal retrieval benchmark.

Reading between the lines

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

  • Extension beyond the paper: because the test set has only 130 queries and all labels trace back to one LLM plus unnamed manual verification, the reported margins are likely to shrink under independent re-labeling by several legal experts; that check is the most direct way to test the headline gains.
  • Extension beyond the paper: the same pipelined generation (LLM writes contract-like sentences, LLM evaluates their relevance) may create a label source that correlates with the retriever's own inductive biases, so a retriever trained and evaluated on it could look better than it would on fully human-authored contracts.
  • Extension beyond the paper: the phase-1/phase-2 schedule with self-mined hard negatives should be testable as a drop-in addition to other LLM bi-encoders, such as those based on LLaMA-3, without changing the loss or the batch-size-1 deepening step.
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 / 6 minor

Summary. The paper proposes a two-phase fine-tuning pipeline for LLM-based dense retrieval, tailored to Japanese legal text, and introduces a new corpus: 743 Japanese statutes, 302 human-identified contract-chunk-law pairs (with 3,172 Gemini-generated training sentences), split into 99/73/130 train/validation/test examples. Phase 1 trains a LLaMA-2-7B retriever with BM25+ negatives and in-batch negatives; Phase 2 fine-tunes on hard negatives mined from Phase 1. An ensemble variant combines checkpoints and/or BM25+ scores. Experiments report improved Recall@k, MRR@10, MAP@10, and nDCG@10 over sparse, dense, and generative baselines on the Japanese legal dataset, and improved results on a small custom split of MS MARCO. The authors release code and checkpoints.

Significance. If the reported results are reliable, the two-phase training strategy is a plausible and useful contribution for non-English legal retrieval, and the Japanese legal dataset (code, checkpoints released) could be a resource for the community. The paper's inclusion of well-known baselines and its attempt to show English generalization are positive. However, the Japanese label construction relies on a single LLM with light manual verification, the test set is small (130 queries), and no statistical uncertainty is reported; the ensemble weights are tuned on the test set. These issues currently prevent the paper from establishing its central empirical claims. The MS MARCO split is non-standard, further weakening the English-generalization claim. The release of code and model checkpoints is a strength that enables independent verification.

major comments (5)
  1. [§3.1 (iv)–(v), §5.1] The test-set relevance labels are produced by Gemini 1.5 Pro and then only manually verified by removing 'clearly unrelated' chunks; no legal experts, annotator counts, or inter-annotator agreement are reported. The training positives in stage (v) are also Gemini-generated. The reported gains over RepLLaMA on the Japanese dataset (R@10 76.87 vs 73.73, MRR@10 82.54 vs 81.58) may therefore reflect agreement with Gemini's notion of relevance rather than legal relevance. The paper should either provide an expert audit of the test labels (with agreement statistics) or explicitly condition the conclusions on the validity of the Gemini-generated labels.
  2. [§4, Table 3] The ensemble weights w1=0.6 and w2=0.4 for 'Ours(ckpt*)' are selected based on test-set performance ('based on our experiments ... yields the best performance'). This is test-set tuning, so the reported ensemble numbers (e.g., R@10 83.42, MRR@10 83.00 in the text) are optimistically biased. Model selection should be performed on the validation set or via nested cross-validation, and the paper should report the variance of the chosen weights.
  3. [§5.1, Table 4] With 130 test queries, a per-query binary R@10 has a standard error of approximately 5.4 percentage points. The reported difference over RepLLaMA (76.87 vs 73.73, a 3.14-point margin) is well within this noise. No confidence intervals, significance tests, or per-query variance are reported anywhere in the paper. The in-domain superiority claim is therefore not statistically supported in its current form.
  4. [§5.2, Table 5] The MS MARCO experiment uses a non-standard 'split of 17,132 rows' (15,270 train / 1,000 validation / 862 test) and a corpus of ~134,000 documents, rather than the full MS MARCO passage collection or the standard development set. It is not clear whether the baselines in Table 5 were retrained and evaluated on this same split with matched hyperparameters. Without a common, well-defined evaluation protocol, the claim that the pipeline 'proves effective in English contexts' and 'surpasses comparable baselines on the MS MARCO dataset' is not established.
  5. [§5.1, Table 4 (DSI-QG row)] The DSI-QG baseline on the Japanese dataset uses the English docT5query generator (as the paper acknowledges), which is known to be a poor fit for Japanese. Listing this as a standard baseline and then using its very low scores (e.g., R@10 28.77) to demonstrate superiority is unfair. Either run DSI-QG with a Japanese-appropriate query generator, or remove the row and clearly state that DSI-QG was evaluated only under a degenerate configuration.
minor comments (6)
  1. [Table 3] The 'Single' row appears to contain merged digits (e.g., '98.2183'), and the text states MRR@10 = 83.00 while the table's MRR@10 column parses inconsistently. Please correct the table formatting and verify the numbers against the text.
  2. [§4 and algorithms] The notation 'LoRa' is used inconsistently (it should be 'LoRA', Low-Rank Adaptation), and the terms 'globalDataset', 'domainSpecif icDataset', and 'humanLabeledDos' in Algorithms 1–2 are informal; please use standard camelCase or mathematical notation.
  3. [§5.1] The hyperparameters a1 and a2 from Algorithm 1 are not explicitly defined in the text; Section 5.1 describes 'sampled 50 relevant documents from BM25+ results' and 'top 50 relevant documents from the Phase 1 model,' but the mapping to a1 and a2 should be stated in the algorithm or implementation details.
  4. [§3.2, Eq. (1)] The embedding notation 'VT = Decoder("t1 t2 . . . tk <EOS>")[−1]' is ambiguous; clarify that [−1] denotes the hidden state at the <EOS> position, and define the vector for queries and documents separately.
  5. [§1 and §3.1] There is a stray period in 'Japanese e-Gov website. 1,' and the sentence 'Laws were retrieved from the Japanese e-Gov website. 1, which provides access...' should be reworded.
  6. [§5.1 vs §5.2] The two experiments use different max sequence lengths (512 vs 128), batch sizes, and numbers of negatives; the paper should discuss whether these differences are justified by dataset characteristics and whether the MS MARCO experiment is intended as a matched-domain test or a domain-shift test.

Circularity Check

2 steps flagged · score 4.0 of 10

Partially circular: ensemble weights are selected on the test set and the Japanese relevance labels share one generator with the training data; the single-model and MS MARCO results remain independent.

  1. fitted input called prediction [Section 4, 'Single' ensemble paragraph and Algorithm 3; results in Table 3]
    "By assigning weighted scores to each checkpoint, based on our experiments, we observe that the combination of checkpoint 1 and checkpoint 2 (with weights w1 = 0.6 and w2 = 0.4) yields the best performance across multiple evaluation metrics."

    The weights w1 and w2 (and implicitly the BM25+ weight alpha in Algorithm 3) are chosen because that combination yields the best scores on the very metrics reported in Table 3; no separate validation-based selection is described. The 'Single Ours(ckpt*)' row therefore reports a post-hoc fit, not a prediction: the ensemble gain (R@10 83.42 vs 76.87, MRR@10 83.00 vs 82.54) is forced by optimizing the evaluation criterion itself.

  2. self definitional [Section 3.1, stages (iv) and (v)]
    "Relevant rules for these chunks were identified using the Gemini 1.5 Pro to match rules with chunks. Subsequently, manual verification was conducted to remove any chunks that were not directly related to the identified rules. ... for each law in the set, we generated five sentences that mirrored the typical content of a contract related to that law using Gemini 1.5 Pro."

    The Japanese benchmark's relevance labels (stage iv) are produced by Gemini 1.5 Pro and only filtered, not re-labeled, by humans; the synthetic training positives (stage v) are also produced by Gemini 1.5 Pro from the same law articles. The test metric therefore measures how well the retriever reproduces Gemini's own chunk-to-article mapping, and a model trained on Gemini-generated positives is rewarded for matching Gemini's notion of relevance. The reported Japanese-dataset lead over RepLLaMA (R@10 76.87 vs 73.73) can thus reflect shared generator provenance rather than independent legal relevance. This is a benchmark-level circularity; the MS MARCO split is unaffected.

full rationale

The core pipeline derivation is not circular in the equation sense: Phase 2 hard-negative mining is a standard self-mining step, the RepLLaMA-style EOS embedding architecture is adopted from an external paper, and the single-model 'Ours' rows in Tables 4 and 5 are genuine measurements against fixed labels. No load-bearing self-citation or imported uniqueness argument appears. The two concrete circularity pressures are: (1) the Section 4 ensemble weights are selected by optimizing the reported evaluation metrics, so the ensemble improvements are post-hoc fits rather than predictions; and (2) the Japanese dataset's test/validation labels and the bulk of its training positives are both generated by Gemini 1.5 Pro, making the Japanese-dataset comparison a self-consistency check against Gemini. The latter is partly a label-validity concern: if the Gemini labels are accepted as ground truth, the ranking comparison remains internally consistent, but the shared generator weakens the claim that the gains reflect legal relevance. Independent support comes from the MS MARCO split (R@10 80.71 vs RepLLaMA's 77.55) and from the single-model RepLLaMA comparison, which is not affected by the ensemble-weight selection. The 130-query test set and absent significance tests are additional statistical risks, but they are not circularity. Overall, the central single-model claim has independent content, so the score is 4 rather than higher.

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

The central performance claims rest on an LLM-generated Japanese corpus, two test-set-selected ensemble weights, and standard contrastive-retrieval assumptions. No new physical entities are introduced.

free parameters (3)
  • Ensemble weight w1 for checkpoint 1 = 0.6
    Selected on the test set in Section 4 because it 'yields the best performance'; no held-out tuning.
  • Ensemble weight w2 for checkpoint 2 = 0.4
    Selected together with w1 on the test set in Section 4; no held-out tuning.
  • Phase-1/Phase-2 sampling hyperparameters (a1, in-batch count, a2, batch sizes) = 50/30/50/B=2 then B=1 (JA); 15/8/7/B=8 then B=1 (MS MARCO)
    Manual choices not swept; the reported metrics depend on them.
assumptions (4)
  • domain assumption The last-token (EOS) embedding of an autoregressive LLM is a reliable dense representation for retrieval.
    Adopted from RepLLaMA (Ma et al., 2024) in Section 3.2; no independent verification in this paper.
  • domain assumption InfoNCE contrastive loss with dot-product similarity is the right objective for ranking.
    Used throughout Section 3.2.4; standard but assumed.
  • ad hoc to paper Manual review by the authors is sufficient to validate Gemini-generated legal labels.
    Section 3.1 states manual verification 'provided a reasonable degree of accuracy' with no expert auditors or agreement statistics.
  • domain assumption A 130-query test set is large enough to compare retrieval models.
    Section 5.1 reports all Japanese conclusions from this test set without error bars.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adaptive Two-Phase Finetuning LLMs for Japanese Legal Text Retrieval." pith.science (2026). https://pith.science/paper/44VTQY4R

@misc{pith2026241213205,
  author       = {Pith},
  title        = {Pith review of: Adaptive Two-Phase Finetuning LLMs for Japanese Legal Text Retrieval},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/44VTQY4R}},
  note         = {Machine review of arXiv:2412.13205}
}
read the original abstract

Text Retrieval (TR) involves finding and retrieving text-based content relevant to a user's query from a large repository, with applications in real-world scenarios such as legal document retrieval. While most existing studies focus on English, limited work addresses Japanese contexts. In this paper, we introduce a new dataset specifically designed for Japanese legal contexts and propose a novel two-phase pipeline tailored to this domain. In the first phase, the model learns a broad understanding of global contexts, enhancing its generalization and adaptability to diverse queries. In the second phase, the model is fine-tuned to address complex queries specific to legal scenarios. Extensive experiments are conducted to demonstrate the superior performance of our method, which outperforms existing baselines. Furthermore, our pipeline proves effective in English contexts, surpassing comparable baselines on the MS MARCO dataset. We have made our code publicly available on GitHub, and the model checkpoints are accessible via HuggingFace.

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. Optimizing Multi-Stage Language Models for Effective Text Retrieval

    cs.IR 2024-12 reject novelty 3.0 of 10

    A language-model-only, two-phase retrieval pipeline with hard-negative training and a grid-searched ensemble is reported to outperform sparse, dense, and generative baselines on a Japanese legal retrieval test set and...

Reference graph

Works this paper leans on

65 extracted references · 20 canonical work pages · cited by 1 Pith paper

  1. [1]

    write newline

    " write newline " cite write " FUNCTION editor.postfix editor num.names #1 > "( )" "( )" if FUNCTION editor.trans.postfix editor num.names #1 > "( )" "( )" if FUNCTION trans.postfix translator num.names #1 > "( )" "( )" if FUNCTION authors.editors.reflist.apa5 'field := 'dot := field num.names 'numnames := numnames 'format.num.names := format.num.names na...

  2. [2]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter doi edition editor eid howpublished institution journal key keywords month note number organization pages publisher school series title type url volume year eprint archive archivePrefix primaryClass adsurl adsnote version label INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.sta...

  3. [3]

    write newline

    " write newline "" before.all 'output.state := FUNCTION if.digit duplicate "0" = swap duplicate "1" = swap duplicate "2" = swap duplicate "3" = swap duplicate "4" = swap duplicate "5" = swap duplicate "6" = swap duplicate "7" = swap duplicate "8" = swap "9" = or or or or or or or or or FUNCTION n.separate 't := "" #0 'numnames := t empty not t #-1 #1 subs...

  4. [4]

    write newline

    " write newline "" before.all 'output.state := FUNCTION output.doi doi empty skip "doi:" doi * "" * output if FUNCTION format.archive archivePrefix empty "" archivePrefix ":" * if FUNCTION format.primaryClass primaryClass empty "" " [" primaryClass * "] " * if FUNCTION format.eprint eprint empty "" archive empty " https://arxiv.org/abs/" eprint * " " * " ...

  5. [5]

    , " * write output.state after.block = add.period write newline

    ENTRY address archive author booktitle chapter edition editor eprint howpublished institution journal key keywords month note number organization pages publisher school series title type url doi volume year archivePrefix primaryClass eid adsurl adsnote version label INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.sta...

  6. [6]

    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...

  7. [7]

    Available from:

    ENTRY address assignee author booktitle chapter cartographer day edition editor howpublished institution inventor journal key keywords month note number organization pages part publisher school series title type volume word year eprint doi url lastchecked updated archive archivePrefix primaryClass eid adsurl adsnote version label INTEGERS output.state bef...

  8. [8]

    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 ":" * " " *...

Show all 65 references
  1. [9]

    , Adler , S

    botherref Achiam , J. , Adler , S. , Agarwal , S. , Ahmad , L. , Akkaya , I. , Aleman , F.L. , Almeida , D. , Altenschmidt , J. , Altman , S. , Anadkat , S. , et al.: Gpt-4 technical report. arXiv preprint arXiv:2303.08774 (2023) botherref

    2023 arXiv

  2. [11]

    , Hudson , D.A

    botherref Bommasani , R. , Hudson , D.A. , Adeli , E. , Altman , R. , Arora , S. , Arx , S. , Bernstein , M.S. , Bohg , J. , Bosselut , A. , Brunskill , E. , et al.: On the opportunities and risks of foundation models. arXiv preprint arXiv:2108.07258 (2021) botherref

    2021 arXiv

  3. [12]

    , Fergadiotis , M

    botherref Chalkidis , I. , Fergadiotis , M. , Malakasiotis , P. , Androutsopoulos , I. : Large-scale multi-label text classification on eu legislation. arXiv preprint arXiv:1906.02192 (2019) botherref

    2019 arXiv

  4. [13]

    , Fisch , A

    bchapter Chen , D. , Fisch , A. , Weston , J. , Bordes , A. : Reading W ikipedia to answer open-domain questions . In: Barzilay , R. , Kan , M.-Y. (eds.) Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pp. 1870 ...

    2017 doi

  5. [14]

    : Unsupervised cross-lingual representation learning at scale

    botherref Conneau , A. : Unsupervised cross-lingual representation learning at scale. arXiv preprint arXiv:1911.02116 (2019) botherref

    2019 arXiv

  6. [15]

    : Flashattention-2: Faster attention with better parallelism and work partitioning

    botherref Dao , T. : Flashattention-2: Faster attention with better parallelism and work partitioning. arXiv preprint arXiv:2307.08691 (2023) botherref

    2023 arXiv

  7. [16]

    , Chang , M.-W

    bchapter Devlin , J. , Chang , M.-W. , Lee , K. , Toutanova , K. : BERT : Pre-training of deep bidirectional transformers for language understanding . In: Burstein , J. , Doran , C. , Solorio , T. (eds.) Proceedings of the 2019 Conference of the North A merican Chapter of the ...

    2019 doi

  8. [17]

    , Dumais , S.T

    barticle Deerwester , S. , Dumais , S.T. , Furnas , G.W. , Landauer , T.K. , Harshman , R. : Indexing by latent semantic analysis . Journal of the American society for information science 41 ( 6 ), 391 -- 407 ( 1990 ) barticle

    1990

  9. [18]

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

    botherref Devlin , J. : Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805 (2018) botherref

    2018 arXiv

  10. [19]

    , Du , J

    barticle Fu , X. , Du , J. , Zheng , H.-T. , Li , J. , Hou , C. , Zhou , Q. , Kim , H.-G. : Ss-bert: A semantic information selecting approach for open-domain question answering . Electronics 12 ( 7 ), 1692 ( 2023 ) barticle

    2023

  11. [20]

    , Callan , J

    botherref Gao , L. , Callan , J. : Unsupervised corpus aware language model pre-training for dense passage retrieval. arXiv preprint arXiv:2108.05540 (2021) botherref

    2021 arXiv

  12. [21]

    , Callan , J

    bchapter Gao , L. , Callan , J. : Unsupervised corpus aware language model pre-training for dense passage retrieval . In: Muresan , S. , Nakov , P. , Villavicencio , A. (eds.) Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Lo...

    2022 doi

  13. [22]

    , Marasovi \'c , A

    botherref Gururangan , S. , Marasovi \'c , A. , Swayamdipta , S. , Lo , K. , Beltagy , I. , Downey , D. , Smith , N.A. : Don't stop pretraining: Adapt language models to domains and tasks. arXiv preprint arXiv:2004.10964 (2020) botherref

    2020 arXiv

  14. [23]

    : An overview of the japanese legal system

    barticle Hahn , E.J. : An overview of the japanese legal system . Nw. J. Int'l L. & Bus. 5 , 517 ( 1983 ) barticle

    1983

  15. [24]

    , Shen , Y

    botherref Hu , E.J. , Shen , Y. , Wallis , P. , Allen-Zhu , Z. , Li , Y. , Wang , S. , Wang , L. , Chen , W. : Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685 (2021) botherref

    2021 arXiv

  16. [25]

    , Shin , A

    bchapter Jin , Q. , Shin , A. , Lu , Z. : Lader: Log-augmented dense retrieval for biomedical literature search . In: Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval , pp. 2092 -- 2097 ( 2023 ) bchapter

    2023

  17. [26]

    , Oguz , B

    bchapter Karpukhin , V. , Oguz , B. , Min , S. , Lewis , P. , Wu , L. , Edunov , S. , Chen , D. , Yih , W.-t. : Dense passage retrieval for open-domain question answering . In: Webber , B. , Cohn , T. , He , Y. , Liu , Y. (eds.) Proceedings of the 2020 Conference on Empirical ...

    2020 doi

  18. [27]

    , Abrego , G.H

    bchapter Lu , J. , Abrego , G.H. , Ma , J. , Ni , J. , Yang , Y. : Multi-stage training with improved negative contrast for neural passage retrieval . In: Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing , pp. 6091 -- 6103 ( 2021 ) bchapter

    2021

  19. [28]

    : Roberta: A robustly optimized bert pretraining approach

    botherref Liu , Y. : Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692 (2019) botherref

    2019 arXiv

  20. [29]

    , Liu , Z

    botherref Li , C. , Liu , Z. , Xiao , S. , Shao , Y. : Making large language models a better foundation for dense retrieval. arXiv preprint arXiv:2312.15503 (2023) botherref

    2023

  21. [30]

    u ttler , H. , Lewis , M. , Yih , W.-t. , Rockt \

    barticle Lewis , P. , Perez , E. , Piktus , A. , Petroni , F. , Karpukhin , V. , Goyal , N. , K \"u ttler , H. , Lewis , M. , Yih , W.-t. , Rockt \"a schel , T. , : Retrieval-augmented generation for knowledge-intensive nlp tasks . Advances in Neural Information Processing Sys...

    2020

  22. [31]

    , Roy , R

    botherref Lee , C. , Roy , R. , Xu , M. , Raiman , J. , Shoeybi , M. , Catanzaro , B. , Ping , W. : Nv-embed: Improved techniques for training llms as generalist embedding models. arXiv preprint arXiv:2405.17428 (2024) botherref

    2024 arXiv

  23. [32]

    , Finin , T

    bchapter Martineau , J. , Finin , T. : Delta tfidf: An improved feature space for sentiment analysis . In: Proceedings of the International AAAI Conference on Web and Social Media , vol. 3 , pp. 258 -- 261 ( 2009 ) bchapter

    2009

  24. [33]

    , Wang , L

    bchapter Ma , X. , Wang , L. , Yang , N. , Wei , F. , Lin , J. : Fine-tuning llama for multi-stage text retrieval . In: Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval , pp. 2421 -- 2425 ( 2024 ) bchapter

    2024

  25. [34]

    , Zaheer , M

    bchapter Monath , N. , Zaheer , M. , Allen , K. , McCallum , A. : Improving dual-encoder training through dynamic indexes for negative mining . In: International Conference on Artificial Intelligence and Statistics , pp. 9308 -- 9330 ( 2023 ). PMLR bchapter

    2023

  26. [35]

    , Lin , J

    botherref Nogueira , R. , Lin , J. , Epistemic , A. : From doc2query to doctttttquery. Online preprint 6(2) (2019) botherref

    2019

  27. [36]

    , Rosenberg , M

    botherref Nguyen , T. , Rosenberg , M. , Song , X. , Gao , J. , Tiwary , S. , Majumder , R. , Deng , L. : MS MARCO: A human generated machine reading comprehension dataset. CoRR abs/1611.09268 (2016) https://arxiv.org/abs/1611.09268 arXiv:1611.09268 botherref

    2016 arXiv

  28. [37]

    , Nguyen , H.-T

    botherref Pham , N.-M. , Nguyen , H.-T. , Do , T.-H. : Multi-stage information retrieval for vietnamese legal texts. arXiv preprint arXiv:2209.14494 (2022) botherref

    2022 arXiv

  29. [38]

    , Ding , Y

    botherref Qu , Y. , Ding , Y. , Liu , J. , Liu , K. , Ren , R. , Zhao , W.X. , Dong , D. , Wu , H. , Wang , H. : Rocketqa: An optimized training approach to dense passage retrieval for open-domain question answering. arXiv preprint arXiv:2010.08191 (2020) botherref

    2020 arXiv

  30. [39]

    : Sentence-bert: Sentence embeddings using siamese bert-networks

    botherref Reimers , N. : Sentence-bert: Sentence embeddings using siamese bert-networks. arXiv preprint arXiv:1908.10084 (2019) botherref

    2019 arXiv

  31. [40]

    , Gurevych , I

    bchapter Reimers , N. , Gurevych , I. : Sentence- BERT : Sentence embeddings using S iamese BERT -networks . In: Inui , K. , Jiang , J. , Ng , V. , Wan , X. (eds.) Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International ...

    2019 doi

  32. [41]

    , Lv , S

    botherref Ren , R. , Lv , S. , Qu , Y. , Liu , J. , Zhao , W.X. , She , Q. , Wu , H. , Wang , H. , Wen , J.-R. : Pair: Leveraging passage-centric similarity relation for improving dense passage retrieval. arXiv preprint arXiv:2108.06027 (2021) botherref

    2021 arXiv

  33. [42]

    , Shazeer , N

    barticle Raffel , C. , Shazeer , N. , Roberts , A. , Lee , K. , Narang , S. , Matena , M. , Zhou , Y. , Li , W. , Liu , P.J. : Exploring the limits of transfer learning with a unified text-to-text transformer . Journal of machine learning research 21 ( 140 ), 1 -- 67 ( 2020 ) barticle

    2020

  34. [43]

    , Walker , S

    bchapter Robertson , S.E. , Walker , S. : Some simple effective approximations to the 2-poisson model for probabilistic weighted retrieval . In: SIGIR’94: Proceedings of the Seventeenth Annual International ACM-SIGIR Conference on Research and Development in Information Retrie...

    1994

  35. [44]

    , Walker , S

    barticle Robertson , S.E. , Walker , S. , Jones , S. , Hancock-Beaulieu , M.M. , Gatford , M. , : Okapi at trec-3 . Nist Special Publication Sp 109 , 109 ( 1995 ) barticle

    1995

  36. [45]

    , Zaragoza , H

    bchapter Robertson , S. , Zaragoza , H. , Taylor , M. : Simple bm25 extension to multiple weighted fields . In: Proceedings of the Thirteenth ACM International Conference on Information and Knowledge Management , pp. 42 -- 49 ( 2004 ) bchapter

    2004

  37. [46]

    : Modern information retrieval

    botherref Salton , G. : Modern information retrieval. (No Title) (1983) botherref

    1983

  38. [47]

    : Choice of law: an empirical analysis

    barticle Sanga , S. : Choice of law: an empirical analysis . Journal of Empirical Legal Studies 11 ( 4 ), 894 -- 928 ( 2014 ) barticle

    2014

  39. [48]

    , Buckley , C

    barticle Salton , G. , Buckley , C. : Term-weighting approaches in automatic text retrieval . Information Processing & Management 24 ( 5 ), 513 -- 523 ( 1988 ) 10.1016/0306-4573(88)90021-0 barticle

    1988 doi

  40. [49]

    , Croft , W.B

    bchapter Song , F. , Croft , W.B. : A general language model for information retrieval . In: Proceedings of the Eighth International Conference on Information and Knowledge Management , pp. 316 -- 321 ( 1999 ) bchapter

    1999

  41. [50]

    , Min , S

    botherref Shi , W. , Min , S. , Yasunaga , M. , Seo , M. , James , R. , Lewis , M. , Zettlemoyer , L. , Yih , W.-t. : Replug: Retrieval-augmented black-box language models. arXiv preprint arXiv:2301.12652 (2023) botherref

    2023 arXiv

  42. [51]

    , Sen , J

    botherref Sil , A. , Sen , J. , Iyer , B. , Franz , M. , Fadnis , K. , Bornea , M. , Rosenthal , S. , McCarley , S. , Zhang , R. , Kumar , V. , et al.: Primeqa: the prime repository for state-of-the-art multilingual question answering research and development. arXiv preprint a...

    2023 arXiv

  43. [52]

    , Wong , A

    barticle Salton , G. , Wong , A. , Yang , C.-S. : A vector space model for automatic indexing . Communications of the ACM 18 ( 11 ), 613 -- 620 ( 1975 ) barticle

    1975

  44. [53]

    , Yan , L

    botherref Sun , W. , Yan , L. , Chen , Z. , Wang , S. , Zhu , H. , Ren , P. , Chen , Z. , Yin , D. , Rijke , M. , Ren , Z. : Learning to tokenize for generative retrieval. Advances in Neural Information Processing Systems 36 (2024) botherref

    2024

  45. [54]

    , Yokote , K

    botherref Sasazawa , Y. , Yokote , K. , Imaichi , O. , Sogawa , Y. : Text retrieval with multi-stage re-ranking models. arXiv preprint arXiv:2311.07994 (2023) botherref

    2023 arXiv

  46. [55]

    , Lavril , T

    botherref Touvron , H. , Lavril , T. , Izacard , G. , Martinet , X. , Lachaux , M.-A. , Lacroix , T. , Rozi \`e re , B. , Goyal , N. , Hambro , E. , Azhar , F. , et al.: Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971 (2023) botherref

    2023 arXiv

  47. [56]

    , Martin , L

    botherref Touvron , H. , Martin , L. , Stone , K. , Albert , P. , Almahairi , A. , Babaei , Y. , Bashlykov , N. , Batra , S. , Bhargava , P. , Bhosale , S. , et al.: Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288 (2023) botherref

    2023 arXiv

  48. [57]

    , Reimers , N

    botherref Thakur , N. , Reimers , N. , R \"u ckl \'e , A. , Srivastava , A. , Gurevych , I. : Beir: A heterogenous benchmark for zero-shot evaluation of information retrieval models. arXiv preprint arXiv:2104.08663 (2021) botherref

    2021 arXiv

  49. [58]

    , Tran , V

    barticle Tay , Y. , Tran , V. , Dehghani , M. , Ni , J. , Bahri , D. , Mehta , H. , Qin , Z. , Hui , K. , Zhao , Z. , Gupta , J. , : Transformer memory as a differentiable search index . Advances in Neural Information Processing Systems 35 , 21831 -- 21843 ( 2022 ) barticle

    2022

  50. [59]

    , Vlachos , A

    bchapter Thorne , J. , Vlachos , A. , Christodoulopoulos , C. , Mittal , A. : FEVER : a large-scale dataset for fact extraction and VER ification . In: Walker , M. , Ji , H. , Stent , A. (eds.) Proceedings of the 2018 Conference of the North A merican Chapter of the Associatio...

    2018 doi

  51. [60]

    , Cer , D

    bchapter Yang , Y. , Cer , D. , Ahmad , A. , Guo , M. , Law , J. , Constant , N. , Hernandez Abrego , G. , Yuan , S. , Tar , C. , Sung , Y.-h. , Strope , B. , Kurzweil , R. : Multilingual universal sentence encoder for semantic retrieval . In: Celikyilmaz , A. , Wen , T.-H. (e...

    2020 doi

  52. [61]

    , Ren , H

    botherref Zhuang , S. , Ren , H. , Shou , L. , Pei , J. , Gong , M. , Zuccon , G. , Jiang , D. : Bridging the gap between indexing and retrieval for differentiable search index with query generation. arXiv preprint arXiv:2206.10128 (2022) botherref

    2022 arXiv

  53. [62]

    , Zhu , W

    barticle Zhai , Q. , Zhu , W. , Zhang , X. , Liu , C. : Contrastive refinement for dense retrieval inference in the open-domain question answering task . Future Internet 15 ( 4 ), 137 ( 2023 ) barticle

    2023

  54. [63]

    , " * write output.state after.block = add.period write newline

    ENTRY address archive author booktitle chapter doi edition editor eid eprint howpublished institution journal key keywords month note number organization pages publisher school series title type url volume year archivePrefix primaryClass adsurl adsnote version label extra.labe...

  55. [64]

    write newline

    " write newline "" before.all 'output.state := FUNCTION add.period duplicate empty 'skip "." * add.blank if FUNCTION if.digit duplicate "0" = swap duplicate "1" = swap duplicate "2" = swap duplicate "3" = swap duplicate "4" = swap duplicate "5" = swap duplicate "6" = swap dupl...

  56. [65]

    write newline

    " write newline "" before.all 'output.state := FUNCTION string.to.integer 't := t text.length 'k := #1 'char.num := t char.num #1 substring 's := s is.num s "." = or char.num k = not and char.num #1 + 'char.num := while char.num #1 - 'char.num := t #1 char.num substring FUNCTI...

  57. [66]

    write newline

    " write newline "" before.all 'output.state := FUNCTION string.to.integer 't := t text.length 'k := #1 'char.num := t char.num #1 substring 's := s is.num s "." = or char.num k = not and char.num #1 + 'char.num := while char.num #1 - 'char.num := t #1 char.num substring FUNCTI...

Pith tools

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