Pith. sign in

REVIEW 4 major objections 5 minor 37 references

Leveraging Information Retrieval to Enhance Spoken Language Understanding Prompts in Few-Shot Learning

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Selecting few-shot prompt examples with BM25 lexical retrieval improves slot-filling F1 by roughly 29 points on average over random selection across four spoken-language-understanding benchmarks.

desk verdict A practical, honest empirical study showing BM25 beats random and intent-based example selection for few-shot SLU prompts, but the gains are called 'significant' without variance estimates. read the letter →

arxiv 2506.03035 v1 pith:MN3AHHEK submitted 2025-06-03 cs.CL cs.AIcs.IR

Pierre Lepagnol , Sahar Ghannay , Thomas Gerald , Christophe Servan , Sophie Rosset This is my paper
classification cs.CLcs.AIcs.IR
keywords spokenlanguageunderstandingslotfillingfew-shotlearningpromptengineeringinformationretrievalBM25in-contextlargemodels
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 tries to establish that the choice of examples in a few-shot prompt is a first-order factor in spoken language understanding, and that a simple lexical retrieval step can make that choice well. Specifically, it claims that selecting the ten prompt examples by BM25 similarity to the current utterance improves slot-filling F1 by roughly 29 points on average over random selection across ATIS, SNIPS, SLURP, and MEDIA, without lengthening the prompt. The authors argue this matters because intent-based selection, the common prior choice, can miss utterances that share the same slots even when the intent differs. If the claim holds, few-shot SLU systems can replace random or intent-anchored example picking with an inexpensive, parameter-free retriever and get better annotations under the same prompt budget.

What carries the argument

The load-bearing object is BM25, a probabilistic lexical ranking function that scores each training utterance against the test utterance by word matching, with term-frequency saturation and document-length normalization. For each utterance to label, the pipeline retrieves the top-K most similar training utterances, formats them into a fixed prompt that lists the intent, slot definitions, and labeled examples, then asks an instruction-tuned 8-billion-parameter large language model to annotate the utterance. The comparison against ColBERT, a late-interaction neural retriever, isolates lexical overlap as the mechanism: BM25's advantage tracks its higher rate of retrieving examples that contain the expected slot types, not semantic relatedness.

What would settle it

Re-run the same BM25-versus-random comparison on a SLU dataset released after the model's training cutoff, or on paraphrased versions of ATIS and SNIPS that remove verbatim overlap with pretraining text; the central claim fails if the BM25 advantage disappears while random selection is unchanged.

Watch

Extended reading notes

Core claim

The paper's central claim is that lexical information retrieval, specifically BM25, is a more effective way to choose the ten examples in a few-shot slot-filling prompt than random selection, intent-based selection, or semantic retrieval with ColBERT. Across ATIS, SNIPS, SLURP, and MEDIA, BM25-selected prompts raise mean F1 from 38.43 (random) and 55.23 (intent-based) to 67.61, and combining BM25 with intent filtering does even better on the datasets with heavy slot overlap (69.31 for BM25-to-intent). The paper also reports that five BM25-chosen examples outperform one hundred randomly chosen ones, and that BM25 retrieves the expected slot types far more often (about 90% of prompts) than intent-based or random selection. The authors interpret this as evidence that lexical alignment between the utterance and the prompt examples is the most valuable signal to pass to the LLM.

Load-bearing premise

The load-bearing premise is that the LLM has not already been trained on the evaluation corpora; the authors themselves flag in Sections 4.2 and 6 that ATIS and SNIPS may have leaked into pretraining data, and if that contamination is substantial the BM25-versus-random comparison would not measure genuine few-shot learning.

Editorial extensions

If this is right

  • Few-shot slot-filling prompts should retrieve examples by lexical similarity rather than picking randomly or by intent alone, since BM25 improves F1 on all four benchmarks tested.
  • The best configuration is dataset-dependent: plain BM25 wins on ATIS, while intent-filtered BM25 (in either order) wins on MEDIA and SLURP, where slots overlap across intents.
  • Prompt length is not the main lever: five BM25-chosen examples outperform 100 random examples, so shorter prompts can be both cheaper and more accurate.
  • Lexical alignment between the utterance and the examples matters more than semantic embedding similarity for this task, since ColBERT underperforms BM25 on every benchmark.
  • Retrieval quality is the bottleneck: when the retriever picks poor examples, as ColBERT does on SLURP, performance falls below intent-based selection.

Reading between the lines

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

  • The dataset-dependent ordering (plain BM25 on ATIS, intent-plus-BM25 on MEDIA and SLURP) suggests a cheap adaptive rule, such as measuring slot-overlap entropy per dataset and choosing whether to filter by intent before retrieval; the paper does not propose such a rule.
  • If the ATIS and SNIPS contamination the authors flag is real, the cleanest test of the mechanism is on fresh SLU data; until then the magnitude of the BM25 gain on those two datasets should be treated with caution.
  • The same retrieval-based example selection could transfer to other structured extraction tasks and other instruction-tuned models, but the paper's evidence is limited to one model and one slot-filling template.
  • A hybrid retriever that combines BM25's lexical precision with semantic embeddings might recover cases where the test utterance paraphrases slot vocabulary, an extension the paper's ColBERT results do not rule out.
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. This paper addresses few-shot slot-filling in spoken language understanding by selecting prompt examples with information retrieval. The authors compare BM25, ColBERT, intent-based selection, and random selection, plus intent-filtered BM25 variants, on four benchmarks (ATIS, SNIPS, SLURP, MEDIA) using Hermes-3-Llama-3.1-8B. They report F1 scores, slot/intent presence rates, and scaling curves. The main claim is that lexical BM25 retrieval consistently improves over random and intent-based selection without lengthening the prompt. The authors also report a comparison to fine-tuned SOTA and acknowledge possible training-data contamination for ATIS and SNIPS.

Significance. If the claim holds, the contribution is practical and useful: a cheap, model-agnostic lexical retrieval step can improve few-shot SLU prompts, with no increase in prompt length. The paper includes a multi-benchmark comparison, a semantic retrieval baseline (ColBERT), an analysis of retrieved slot coverage, and a public code repository. The main weakness is that the headline results are presented without variance or significance testing, and the per-dataset choice of the best BM25 variant is made on the test set, so the magnitude and generalizability of the claimed improvement are not yet fully established.

major comments (4)
  1. [§4.1, Table 2, Abstract] The abstract and §4.1 describe the BM25 gains as 'significant' and 'consistent', but Table 2 reports a single F1 value per condition with no confidence intervals, standard deviations, or significance tests. Random selection is stochastic, and several comparisons are close (e.g., ATIS: BM25 86.63 vs. Intent→BM25 86.38; SNIPS: BM25 84.38 vs. BM25→Intent 85.21). Without repeated seeds or a paired test, the reader cannot distinguish a robust improvement from draw-to-draw variation. Please report means and standard deviations over at least 3–5 seeds, or bootstrap confidence intervals, and apply a paired significance test or an equivalence test for the key comparisons. This support is essential for the central claim.
  2. [§4.2, §6] The authors correctly acknowledge that ATIS and SNIPS may overlap with the pretraining data of Hermes-3-Llama-3.1-8B. This is more than a caveat: if the model has memorized those datasets, the comparison on ATIS and SNIPS may reflect memorization rather than prompt example selection. Since the paper's cross-dataset claim relies partly on those two corpora, please provide additional evidence that the effect is not contamination-driven, for example by evaluating on a held-out subset not plausibly in pretraining, or by reporting results separately for datasets with and without contamination risk. The trend on SLURP and MEDIA is encouraging, but the current text does not quantify the extent of the problem.
  3. [§4.2, Table 2] The statement that the optimal configuration is dataset-dependent and the selection of the best BM25 variant per dataset are based on test-set F1. In a genuine few-shot deployment, a practitioner would not know which variant (BM25, BM25→Intent, Intent→BM25) is best. Please either provide a principled selection rule using a validation split or restrict the contribution to the family-level claim that all BM25 variants outperform random and intent-based selection. As written, the per-dataset 'best' configuration risks overfitting to the test set.
  4. [§4.3, Figure 3] The scaling curves in Figure 3 have no error bars, and the claim that five BM25 examples always outperform one hundred random examples is not statistically anchored. Since the random baseline varies with the draw, the crossover point may be an artifact of one run. Please provide error bars, repeated runs, or the underlying per-shot data points for the scaling experiments.
minor comments (5)
  1. [§3.3] The text refers to 'see Appendix in supplementary materials' for ColBERT details, but no appendix is present in the manuscript; please add the details or a pointer to the supplementary material.
  2. [§3.3] BM25 hyperparameters (k1, b) and the retrieval implementation are not specified; these are needed for reproducibility, especially since BM25 is the core method.
  3. [§3.2, §4.1] The random seed and the number of random draws are not stated; without this, the stochastic baselines cannot be reproduced.
  4. [§5] The conclusion says the method improves performance 'without additional computational cost', but §5 also states that 'the retrieval process can introduce additional computational overhead'; please reconcile these statements.
  5. [Table 2] The row label 'Fully Fine-tunned' contains a typo; it should read 'Fully Fine-tuned'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the retrieval pipeline uses fixed training-set examples and evaluates on held-out test labels, so the F1 gains are empirical rather than definitional.

full rationale

The paper's central claim is that BM25-based retrieval of prompt examples improves few-shot slot-filling F1. The derivation chain is not circular: retrieval scores are computed between the test utterance and fixed training-set utterances using a standard lexical ranking function (BM25) or a fixed ColBERT model, and the resulting F1 is evaluated on held-out test labels that never enter the retrieval or prompt-construction process. No parameter is fitted to the test labels and no predicted quantity is redefined as its own input. The reported 'best configuration' comparisons (BM25 vs BM25→Intent vs Intent→BM25) are run on the same test data, and any selection of the best configuration per dataset is an evaluation-choice concern rather than a circularity. The few self-citations (MEDIA benchmark version [25] and intent-label enrichment [24]) are data-resource citations used to describe the benchmark, not load-bearing justifications of the method; the paper does not invoke an author-specific uniqueness theorem or smuggle in an ansatz via citation. The acknowledged limitation that Hermes-3/Llama-3 may have been pretrained on ATIS and SNIPS (Section 6) and the absence of multi-seed error bars or significance tests are external-validity and statistical concerns, not circularity. The method is self-contained with respect to the claimed comparison, so the appropriate finding is no significant circularity.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the LLM's in-context learning capability, the availability of a labeled retrieval pool, and the reuse of an unmodified prompt template. There are no invented entities, and the only hand-set numerical parameter is the prompt size K=10, plus unreported BM25 hyperparameters.

free parameters (2)
  • Number of prompt examples K = 10
    The main experiments fix 10 examples following Mirza et al. [16]; Section 4.3 varies this parameter to study its impact.
  • BM25 hyperparameters (k1, b)
    Not reported in the paper; presumably library defaults, which affects exact reproducibility.
assumptions (3)
  • domain assumption The Hermes-3-Llama-3.1-8B instruction-tuned model can perform slot filling from a few prompt examples
    The entire method relies on the model's in-context learning ability, which is not formally guaranteed.
  • domain assumption A training or development set with gold annotations is available for retrieval at inference time
    The method requires a labeled pool of utterances, which may not exist in truly low-resource settings; the authors note this in Section 6.
  • domain assumption The prompt template from [16] is appropriate and is used without optimization
    The paper reuses the template and does not test alternatives, as acknowledged in Section 6.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Leveraging Information Retrieval to Enhance Spoken Language Understanding Prompts in Few-Shot Learning." pith.science (2026). https://pith.science/paper/MN3AHHEK

@misc{pith2026250603035,
  author       = {Pith},
  title        = {Pith review of: Leveraging Information Retrieval to Enhance Spoken Language Understanding Prompts in Few-Shot Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MN3AHHEK}},
  note         = {Machine review of arXiv:2506.03035}
}
read the original abstract

Understanding user queries is fundamental in many applications, such as home assistants, booking systems, or recommendations. Accordingly, it is crucial to develop accurate Spoken Language Understanding (SLU) approaches to ensure the reliability of the considered system. Current State-of-the-Art SLU techniques rely on large amounts of training data; however, only limited annotated examples are available for specific tasks or languages. In the meantime, instruction-tuned large language models (LLMs) have shown exceptional performance on unseen tasks in a few-shot setting when provided with adequate prompts. In this work, we propose to explore example selection by leveraging Information retrieval (IR) approaches to build an enhanced prompt that is applied to an SLU task. We evaluate the effectiveness of the proposed method on several SLU benchmarks. Experimental results show that lexical IR methods significantly enhance performance without increasing prompt length.

Figures

Figures reproduced from arXiv: 2506.03035 by the authors.

Figure 1
Figure 1. Upset plot illustrating the intersections of slots across different intents. In the SNIPS dataset, only a few slots are shared between intents, indicating minimal overlap. In contrast, the MEDIA dataset shows significant overlap, with many slots being shared among multiple intents. proaches select examples randomly, using only the intention as an anchor [15, 18]. However, an analysis of the SF discrepancy linked to … view at source ↗
Figure 2
Figure 2. Prompt Template: In red, the context. In green are the instructions. In blue, examples with labeled slots. In gray is the current utterance we want to label. Generating & Parsing The formatted prompt is fed into the LLM after applying the proper prompt template to gener￾ate 256 tokens using greedy decoding. The response is fi￾nally parsed using regular expressions that catch each slot type/value pair. 3. Experimenta… view at source ↗
Figure 3
Figure 3. Performance of the different methods for each number of examples in the prompt for ATIS, SLURP, SNIPS, and MEDIA SNIPS and SLURP). In the same way, ColBERT miss more of￾ten the appropriate slots compared to Intent-Based Selection. In summary, the results demonstrate that retrieval meth￾ods integrating BM25, especially when combined with intent information (either before or after retrieval), substantially out￾perform… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

37 extracted references · 34 canonical work pages

  1. [1]

    What is the weather like in Abu Dhabi tomorrow?

    Introduction Spoken Language Understanding (SLU) is a critical component of task-oriented dialogue systems [1]. It consists of extracting information from user utterances and providing faithful infor- mation to a more extensive system. The extracted information is crucial to the dialogue systems for different sub-tasks, from querying a database (e.g., ret...

    work page

  2. [2]

    Leveraging Information Retrieval to Enhance Spoken Language Understanding Prompts in Few-Shot Learning

    Proposed Approach While selecting the right examples is a key challenge in order to enhance the quality of the prompt [19], to the best of our knowledge, the current state of the art concerning SLU tasks using prompt engineering for slot-filling (SF) does not select examples based on utterance similarity. In contrast, related ap- arXiv:2506.03035v1 [cs.CL...

    work page Pith review arXiv 2025

  3. [3]

    We present the datasets used, describe the chosen language model, detail the retrieval mechanisms em- ployed, and specify the baseline and comparison methods

    Experimental Setup This section outlines the experimental methodology for eval- uating our approach. We present the datasets used, describe the chosen language model, detail the retrieval mechanisms em- ployed, and specify the baseline and comparison methods. 3.1. Datasets & Metrics Our study focuses on four well-known SLU benchmarks (see details in table...

    work page 2022

  4. [4]

    Results & analysis In this section, we report and discuss the results of the different experiments. It includes a comparison to state-of-the-art SLU methods, the evaluation of the different example selection meth- ods, the impact of retrieved utterances on performances, and the impact of the number of selected examples. 4.1. Comparative analysis with Stat...

    work page

  5. [5]

    By integrating IR methods, specifically BM25, into the ex- ample selection process, we have addressed the challenges of overlapping intents and slots in complex datasets

    Conclusion This study investigates the use of information retrieval (IR) methods to enhance prompt construction in spoken language understanding (SLU) task. By integrating IR methods, specifically BM25, into the ex- ample selection process, we have addressed the challenges of overlapping intents and slots in complex datasets. Our com- prehensive evaluatio...

    work page

  6. [6]

    Limitations Despite promising advances, this study on language models for SLU tasks may have several limitations. Firstly, the reliance on access to a training dataset, for example retrieval, poses chal- lenges, particularly in low-resource settings where such data may be scarce or unavailable, thus limiting the applicability of the method. In addition, t...

    work page

  7. [7]

    This work was performed using HPC resources from GENCI-IDRIS (Grant 2023-AD011014242)

    Acknowledgments This work is supported by the ANRT (Association nationale de la recherche et de la technologie) with a CIFRE fellowship granted to SCIAM2 (CIFRE N°2022/1608). This work was performed using HPC resources from GENCI-IDRIS (Grant 2023-AD011014242)

    work page 2022

  8. [8]

    End-to-end sequence labeling via bi- directional LSTM-CNNs-CRF,

    X. Ma and E. Hovy, “End-to-end sequence labeling via bi- directional LSTM-CNNs-CRF,” inACL 2016, K. Erk and N. A. Smith, Eds. Berlin, Germany: Association for Computational Linguistics, Aug. 2016

    work page 2016

Show all 37 references
  1. [9]

    Tur and R

    G. Tur and R. De Mori,Spoken language understanding: Systems for extracting semantic information from speech, 2011

    2011

  2. [10]

    Results of the French Evalda-Media evaluation campaign for literal understanding,

    H. Bonneau-Maynard, C. Ayache, F. B ´echet, A. Denis, A. Kuhn, F. Lefevre, D. Mostefa, M. Quignard, S. Rosset, C. Servan, and J. Villaneau, “Results of the French Evalda-Media evaluation campaign for literal understanding,” inLREC 2006, Genes, Italy, Mai 2006

    2006

  3. [11]

    The ATIS spoken language systems pilot corpus,

    C. T. Hemphill, J. J. Godfrey, and G. R. Doddington, “The ATIS spoken language systems pilot corpus,” inSpeech and Natural Language: Proc. of a Workshop Held at Hidden Valley, Pennsylvania, June 24-27,1990, 1990. [Online]. Available: https://www.aclweb.org/anthology/H90-1021

    1990

  4. [12]

    Semantic annotation of the French media dialog cor- pus,

    H. Bonneau-Maynard, S. Rosset, C. Ayache, A. Kuhn, and D. Mostefa, “Semantic annotation of the French media dialog cor- pus,” inProc. Interspeech 2005, 2005, pp. 3457–3460

    2005

  5. [13]

    SLURP: A spoken language understanding resource package,

    E. Bastianelli, A. Vanzo, P. Swietojanski, and V . Rieser, “SLURP: A spoken language understanding resource package,” inEMNLP, B. Webber, T. Cohn, Y . He, and Y . Liu, Eds. Association for Computational Linguistics, Nov. 2020, pp. 7252–7262

    2020

  6. [14]

    Conceptual decoding from word lattices: application to the spoken dialogue corpus MEDIA,

    C. Servan, C. Raymond, F. B ´echet, and P. Nocera, “Conceptual decoding from word lattices: application to the spoken dialogue corpus MEDIA,” inInterspeech 2006 - ICSLP, Pittsburgh, United States, Sep. 2006

    2006

  7. [15]

    Compar- ing stochastic approaches to spoken language understanding in multiple languages,

    S. Hahn, M. Dinarelli, C. Raymond, F. Lefevre, P. Lehnen, R. De Mori, A. Moschitti, H. Ney, and G. Riccardi, “Compar- ing stochastic approaches to spoken language understanding in multiple languages,”IEEE Transactions on Audio, Speech, and Language Processing, vol. 19, no. 6, ...

    2011

  8. [16]

    BM25→Intent

    using theHermes-3-Llama-3.1-8Bmodel, which reaches 83.98 of F1-scores. This approach uses intent selection from a manually selected pool of examples from the SNIPS cor- pus. Since it is not possible to reproduce this manual selection for every corpus, we propose to select the ...

  9. [17]

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

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of deep bidirectional transformers for language understanding,” inProc. of NAACL, 2019. [Online]. Available: https://www.aclweb.org/anthology/N19-1423

    2019

  10. [18]

    Neural Networks ap- proaches focused on French Spoken Language Understanding: application to the MEDIA Evaluation Task,

    S. Ghannay, C. Servan, and S. Rosset, “Neural Networks ap- proaches focused on French Spoken Language Understanding: application to the MEDIA Evaluation Task,” inCOLING’2020, Barcelona (online), Spain, Dec. 2020

    2020

  11. [19]

    Multi-lingual intent detection and slot filling in a joint bert-based model,

    G. Castellucci, V . Bellomaria, A. Favalli, and R. Romagnoli, “Multi-lingual intent detection and slot filling in a joint bert-based model,”arXiv preprint arXiv:1907.02884, 2019

    1907 arXiv

  12. [20]

    Bert for joint intent classifica- tion and slot filling,

    Q. Chen, Z. Zhuo, and W. Wang, “Bert for joint intent classifica- tion and slot filling,”arXiv preprint arXiv:1902.10909, 2019

    1902 arXiv

  13. [21]

    A joint learning framework with bert for spoken language understanding,

    Z. Zhang, Z. Zhang, H. Chen, and Z. Zhang, “A joint learning framework with bert for spoken language understanding,”IEEE Access, 2019

    2019

  14. [22]

    A Survey of Joint Intent Detection and Slot Filling Models in Natural Lan- guage Understanding,

    H. Weld, X. Huang, S. Long, J. Poon, and S. C. Han, “A Survey of Joint Intent Detection and Slot Filling Models in Natural Lan- guage Understanding,”ACM Computing Surveys, vol. 55, no. 8, pp. 1–38, Aug. 2023

    2023

  15. [23]

    Can ChatGPT Detect Intent? Evaluating Large Language Models for Spoken Language Understanding,

    M. He and P. N. Garner, “Can ChatGPT Detect Intent? Evaluating Large Language Models for Spoken Language Understanding,” Tech. Rep., Aug. 2023, arXiv:2305.13512 [cs, eess] type: article. 2https://www.sciam.fr/

    2023 arXiv

  16. [24]

    Illuminer: Instruction-tuned large language models as few-shot intent clas- sifier and slot filler,

    P. Mirza, V . Sudhi, S. Sahoo, and S. R. Bhat, “Illuminer: Instruction-tuned large language models as few-shot intent clas- sifier and slot filler,” inLREC-COLING, Torino, Italy, May 2024

    2024

  17. [25]

    Zero-shot spoken language understanding via large language models: A preliminary study,

    Z. Zhu, X. Cheng, H. An, Z. Wang, D. Chen, and Z. Huang, “Zero-shot spoken language understanding via large language models: A preliminary study,” inLREC-COLING 2024, Torino, Italia, May 2024, pp. 17 877–17 883

    2024

  18. [26]

    Croprompt: Cross-task interactive prompting for zero- shot spoken language understanding,

    L. Qin, F. Wei, Q. Chen, J. Zhou, S. Huang, J. Si, W. Lu, and W. Che, “Croprompt: Cross-task interactive prompting for zero- shot spoken language understanding,” 2024

    2024

  19. [27]

    Retrieval-based prompt selection for code-related few-shot learning,

    N. Nashid, M. Sintaha, and A. Mesbah, “Retrieval-based prompt selection for code-related few-shot learning,” in2023 IEEE/ACM 45th International Conference on Software Engineering (ICSE), 2023, pp. 2450–2462

    2023

  20. [28]

    The probabilistic relevance framework: Bm25 and beyond,

    S. Robertson and H. Zaragoza, “The probabilistic relevance framework: Bm25 and beyond,” vol. 3, no. 4, pp. 333–389, apr 2009

    2009

  21. [29]

    Colbert: Efficient and effective pas- sage search via contextualized late interaction over BERT,

    O. Khattab and M. Zaharia, “Colbert: Efficient and effective pas- sage search via contextualized late interaction over BERT,” in ACM SIGIR 2020, Virtual. ACM, 2020, pp. 39–48

    2020

  22. [30]

    Colbertv2: Effective and efficient retrieval via lightweight late interaction,

    K. Santhanam, O. Khattab, J. Saad-Falcon, C. Potts, and M. Za- haria, “Colbertv2: Effective and efficient retrieval via lightweight late interaction,” 2022

    2022

  23. [31]

    Snips voice platform: an embedded spoken language under- standing system for private-by-design voice interfaces,

    A. Coucke, A. Saade, A. Ball, T. Bluche, A. Caulier, D. Leroy, C. Doumouro, T. Gisselbrecht, F. Caltagirone, T. Lavrilet al., “Snips voice platform: an embedded spoken language under- standing system for private-by-design voice interfaces,”arXiv preprint arXiv:1805.10190, 2018

    2018 arXiv

  24. [32]

    New Semantic Task for the French Spoken Language Under- standing MEDIA Benchmark,

    N. Alavoine, G. Laperriere, C. Servan, S. Ghannay, and S. Ros- set, “New Semantic Task for the French Spoken Language Under- standing MEDIA Benchmark,” inLREC-COLING 2024, Torino, Italy, May 2024

    2024

  25. [33]

    The spoken language un- derstanding MEDIA benchmark dataset in the era of deep learn- ing: data updates, training and evaluation tools,

    G. Laperri `ere, V . Pelloin, A. Caubri`ere, S. Mdhaffar, N. Camelin, S. Ghannay, B. Jabaian, and Y . Est`eve, “The spoken language un- derstanding MEDIA benchmark dataset in the era of deep learn- ing: data updates, training and evaluation tools,” inLREC, Mar- seille, France,...

    2022

  26. [34]

    The llama 3 herd of models,

    A. Dubey et al., “The llama 3 herd of models,” 2024

    2024

  27. [35]

    Hermes 3 technical report,

    R. Teknium, J. Quesnelle, and C. Guang, “Hermes 3 technical report,” 2024

    2024

  28. [36]

    Towards joint intent detection and slot filling via higher-order attention,

    D. Chen, Z. Huang, X. Wu, S. Ge, and Y . Zou, “Towards joint intent detection and slot filling via higher-order attention,” inPro- ceedings of the Thirty-First International Joint Conference on Ar- tificial Intelligence, IJCAI-22, 7 2022, pp. 4072–4078

    2022

  29. [37]

    Hierarchical multi-task natural language understanding for cross-domain conversational AI: HERMIT NLU,

    A. Vanzo, E. Bastianelli, and O. Lemon, “Hierarchical multi-task natural language understanding for cross-domain conversational AI: HERMIT NLU,” inProceedings of the 20th Annual SIGdial Meeting on Discourse and Dialogue, S. Nakamura, M. Gasic, I. Zukerman, G. Skantze, M. Nakan...

    2019

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.