CoChat

Quantity doesn’t buy quality syntax with neural language models

Open access
van Schijndel, Marten, Mueller, Aaron, Linzen, Tal · 2019 · Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)

The DOI matches its Crossref record; 32 of the 32 references it cites were found in a registry or an index.

Abstract

Marten van Schijndel, Aaron Mueller, Tal Linzen. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.18653/v1/d19-1592

Cites

32 of 32 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 LSTM agreement performance in several syntactic constructions. The solid horizontal line indicates chance performance. The dashed lines show the performance of GPT and BERT as reported by Wolf (2019), the performance of humans as reported by Marvin and Linzen (2018), and the performance of GRNN. Error bars reflect standard deviation across the five models in each category.
Figure 1 LSTM agreement performance in several syntactic constructions. The solid horizontal line indicates chance performance. The dashed lines show the performance of GPT and BERT as reported by Wolf (2019), the performance of humans as reported by Marvin and Linzen (2018), and the performance of GRNN. Error bars reflect standard deviation across the five models in each category.CC-BY · From the PDF
Figure 2 Lines depict number of training tokens needed for LSTMs to achieve human-like (left) or 99.99% accuracy (right) in each syntactic agreement condition, according to our estimates. Bars depict the amount of data on which each model was trained.
Figure 2 Lines depict number of training tokens needed for LSTMs to achieve human-like (left) or 99.99% accuracy (right) in each syntactic agreement condition, according to our estimates. Bars depict the amount of data on which each model was trained.CC-BY · From the PDF

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

  • Attention Is All You Need

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404 · cited 2×

    “Comparison with Transformers: Finally, we report results from two publicly available LMs based on non-recurrent self-attention (Transform-ers; Vaswani et al., 2017): GPT (Radford et al., 2018) and BERT (Devlin et al., 2019).”
  • Long Short-Term Memory

    Hochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318 · cited 1×

    “We train 125 RNN LMs with long short-term memory (LSTM, Hochreiter and Schmidhuber, 1997) units, systematically varying the size of the training corpus and the dimensionality of the models’ hidden layer, and track the relationship between these parameters and the performance of the models on…”
  • Exploiting Generative AI to Scale up Intelligent Tutoring Systems

    Jakubův, J., Chvalovský, K., Goertzel, Z., Kaliszyk, C., Olšák, M., Piotrowski, B., Schulz, S., Suda, M., & Urban, J. (2023) · DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · doi:10.4230/lipics.itp.2023.19 · Cited by 76,658

  • BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

    Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805 · Cited by 73,173 · cited 3×

    “Comparison with Transformers: Finally, we report results from two publicly available LMs based on non-recurrent self-attention (Transform-ers; Vaswani et al., 2017): GPT (Radford et al., 2018) and BERT (Devlin et al., 2019).”
  • BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

    Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1423 · Cited by 54,325 · cited 3×

    “Comparison with Transformers: Finally, we report results from two publicly available LMs based on non-recurrent self-attention (Transform-ers; Vaswani et al., 2017): GPT (Radford et al., 2018) and BERT (Devlin et al., 2019).”
  • Glove: Global Vectors for Word Representation

    Pennington, J., Socher, R., & Manning, C. (2014) · Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.3115/v1/d14-1162 · Cited by 34,071 · cited 2×

    “The vocabulary consisted of an intersection of the 400k word GloVe vocabulary (Pennington et al., 2014) with the 50k words used by GRNN (see below); the resulting vocabulary had 28,438 words.”
  • LSTM: A Search Space Odyssey

    Greff, K., Srivastava, R. K., Koutnik, J., Steunebrink, B. R., & Schmidhuber, J. (2017) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2016.2582924 · Cited by 6,955 · cited 2×

    “The limited effect of network size is consistent with previous findings on sequence labeling tasks (Reimers and Gurevych, 2017; Greff et al., 2017).”
  • Recurrent neural network based language model

    Mikolov, T., Karafiát, M., Burget, L., Černocký, J., & Khudanpur, S. (2010) · Interspeech 2010 · doi:10.21437/interspeech.2010-343 · Cited by 5,390 · cited 2×

    “Recurrent neural network language models (LMs) can learn to predict upcoming words with remarkably low perplexity (Mikolov et al., 2010; Jozefowicz et al., 2016; Radford et al., 2019).”
  • Bayesian t tests for accepting and rejecting the null hypothesis

    Rouder, J. N., Speckman, P. L., Sun, D., Morey, R. D., & Iverson, G. (2009) · Psychonomic Bulletin & Review · doi:10.3758/pbr.16.2.225 · Cited by 3,688 · cited 1×

    “We performed statistical tests of our hypotheses using Bayes factors, which quantify the support the data provide for a more complex model compared to a simpler one (Rouder et al., 2009).”
  • Transformer-XL: Attentive language models beyond a fixed-length context

    Dai, Z., Yang, Z., Yang, Y., Carbonell, J., Le, Q., & Salakhutdinov, R. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1285 · Cited by 3,169 · cited 12×

    “RNN LMs have been shown to favor the grammatical variant in the vast majority of cases sampled at random from a corpus (Linzen et al., 2016), but their accuracy decreases in the presence of distracting nouns intervening between the head of the subject and the verb, especially when those nouns are in relative clauses (…”
    Influential
  • The Unreasonable Effectiveness of Data

    Halevy, A., Norvig, P., & Pereira, F. (2009) · IEEE Intelligent Systems · doi:10.1109/mis.2009.36 · Cited by 1,767 · cited 2×

    “Can we hope to address these deficits by training larger and larger networks on larger and larger corpora, relying on the “unreasonable effective-ness” of massive datasets (Halevy et al., 2009) and computational power?”
  • BayesFactor: Computation of Bayes Factors for Common Designs

    Morey, R. D., & Rouder, J. N. (2012) · CRAN: Contributed Packages · doi:10.32614/cran.package.bayesfactor · Cited by 1,173

  • Exploring the Limits of Language Modeling

    Józefowicz, R., Vinyals, O., Schuster, M., Shazeer, N., & Wu, Y. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1602.02410 · Cited by 823 · cited 1×

    “Recurrent neural network language models (LMs) can learn to predict upcoming words with remarkably low perplexity (Mikolov et al., 2010; Jozefowicz et al., 2016; Radford et al., 2019).”
  • Assessing the Ability of LSTMs to Learn Syntax-Sensitive Dependencies

    Linzen, T., Dupoux, E., & Goldberg, Y. (2016) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00115 · Cited by 737 · cited 3×

    “One such evaluation strategy compares the probability assigned by the LM to a minimal pair of sentences differing only in grammaticality (Linzen et al., 2016).”
    InfluentialSelf-citation
  • Using the Output Embedding to Improve Language Models

    Press, O., & Wolf, L. (2017) · Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 2, Short Papers · doi:10.18653/v1/e17-2025 · Cited by 596

  • Colorless Green Recurrent Networks Dream Hierarchically

    Gulordava, K., Bojanowski, P., Grave, E., Linzen, T., & Baroni, M. (2018) · Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers) · doi:10.18653/v1/n18-1108 · Cited by 465

    Self-citation
  • Reporting Score Distributions Makes a Difference: Performance Study of LSTM-networks for Sequence Tagging

    Reimers, N., & Gurevych, I. (2017) · Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d17-1035 · Cited by 463 · cited 1×

    “The limited effect of network size is consistent with previous findings on sequence labeling tasks (Reimers and Gurevych, 2017; Greff et al., 2017).”
  • Assessing BERT's Syntactic Abilities

    Goldberg, Y. (2019) · arXiv · doi:10.48550/arxiv.1901.05287 · Cited by 385 · cited 2×

    “Both of these models have been argued to learn powerful syntactic representations (Goldberg, 2019; Wolf, 2019).”
  • Targeted Syntactic Evaluation of Language Models

    Marvin, R., & Linzen, T. (2018) · Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d18-1151 · Cited by 334 · cited 1×

    “ality of the models’ hidden layer, and track the relationship between these parameters and the performance of the models on agreement dependencies in a range of syntactic constructions (Marvin and Linzen, 2018).”
    Self-citation
  • Tying Word Vectors and Word Classifiers: A Loss Framework for Language Modeling

    Inan, H., Khosravi, K., & Socher, R. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1611.01462 · Cited by 288 · cited 1×

    “Input and output weights were tied during training (Press and Wolf, 2017; Inan et al., 2017); consequently, the input embedding had the same dimensionality as the hidden layers.”
  • Theory of Probability (3rd Edition)

    Litton, C. D. (1984) · Journal of the Operational Research Society · doi:10.1057/jors.1984.141 · Cited by 282

  • Cloze-driven Pretraining of Self-attention Networks

    Baevski, A., Edunov, S., Liu, Y., Zettlemoyer, L., & Auli, M. (2019) · Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) · doi:10.18653/v1/d19-1539 · Cited by 215 · cited 1×

    “Dra-matically increasing the pre-training corpus for a BERT-like model from 562M words to 18G words only leads to a modest improvement in its natu-ral language inference accuracy, from 81.7% to 82.3% (Baevski et al., 2019).”
    Influential
  • Cloze-driven Pretraining of Self-attention Networks

    Baevski, A., Edunov, S., Liu, Y., Zettlemoyer, L., & Auli, M. (2019) · arXiv · doi:10.48550/arxiv.1903.07785 · Cited by 215 · cited 1×

    “Dra-matically increasing the pre-training corpus for a BERT-like model from 562M words to 18G words only leads to a modest improvement in its natu-ral language inference accuracy, from 81.7% to 82.3% (Baevski et al., 2019).”
    Influential
  • LSTMs Can Learn Syntax-Sensitive Dependencies Well, But Modeling Structure Makes Them Better

    Kuncoro, A., Dyer, C., Hale, J., Yogatama, D., Clark, S., & Blunsom, P. (2018) · Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p18-1132 · Cited by 137

  • Learning Longer-term Dependencies in RNNs with Auxiliary Losses

    Trinh, T. H., Dai, A. M., Luong, M.-T., & Le, Q. V. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1803.00144 · Cited by 84

  • Neural Arithmetic Logic Units

    Trask, A., Hill, F., Reed, S., Rae, J. W., Dyer, C., & Blunsom, P. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1808.00508 · Cited by 74 · cited 1×

    “…known to struggle with generalizing short patterns to longer sequences, this pattern constitutes a failure to generalize to shorter sequences (cf. Trask et al., 2018); techniques for improving longer distance dependency learning in LMs (e.g., Trinh et al., 2018; Dai et al., 2019) are unlikely to…”
  • Structural Supervision Improves Learning of Non-Local Grammatical Dependencies

    Wilcox, E., Qian, P., Futrell, R., Ballesteros, M., & Levy, R. (2019) · Proceedings of the 2019 Conference of the North · doi:10.18653/v1/n19-1334 · Cited by 57 · cited 2×

    “Overall, this body of results points to the limited data efficiency of standard RNNs and Transformers, and indicates that learning syntax from realistic amounts of data—in particular the amount of data available to humans when they learn language—may require syntactically structured architectures or explicit syntactic…”
  • Theory of Probability (3rd Edition).

    Litton, C. D., & Jefferys, H. (1984) · Journal of the Operational Research Society · doi:10.2307/2582798 · Cited by 50

  • Distinct patterns of syntactic agreement errors in recurrent networks and humans

    Linzen, T., & Leonard, B. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1807.06882 · Cited by 45 · cited 1×

    “While modeling human errors is of considerable interest to cognitive scientists (Linzen and Leonard, 2018), we believe that in most applied contexts it is desirable for the model to make no errors at all.”
    Self-citation
  • Scalable Syntax-Aware Language Models Using Knowledge Distillation

    Kuncoro, A., Dyer, C., Rimell, L., Clark, S. R. L., & Blunsom, P. (2019) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/p19-1337 · Cited by 36 · cited 1×

    “…indicates that learning syntax from realistic amounts of data—in particular the amount of data available to humans when they learn language—may require syntactically structured architectures or explicit syntactic supervision (Enguehard et al., 2017; Kuncoro et al., 2018, 2019; Wilcox et al., 2019).”
  • Exploring the Syntactic Abilities of RNNs with Multi-task Learning

    Enguehard, É., Goldberg, Y., & Linzen, T. (2017) · Proceedings of the 21st Conference on Computational Natural Language Learning (CoNLL 2017) · doi:10.18653/v1/k17-1003 · Cited by 24 · cited 2×

    “Overall, this body of results points to the limited data efficiency of standard RNNs and Transformers, and indicates that learning syntax from realistic amounts of data—in particular the amount of data available to humans when they learn language—may require syntactically structured architectures or explicit syntactic…”
    Self-citation
  • Does BERT agree? Evaluating knowledge of structure dependence through agreement relationsPreprint

    Bacon, G., & Regier, T. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1908.09892 · Cited by 20 · cited 1×

    “BERT’s agreement accuracy decreases as the subject becomes more distant from its verb (Bacon and Regier, 2019).”

Research works printed but not linked: 3

  1. Kathryn Bock and Carol A. Miller. 1991. Broken agreement. Cognitive Psychology, 23(1):45-93.
  2. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Gar- nett, editors, Advances in Neural Information Pro- cessing Systems 30, pages 5998-6008.
  3. Ethan Wilcox, Peng Qian, Richard Futrell, Miguel Ballesteros, and Roger Levy. 2019. Structural su- pervision improves learning of non-local grammat- ical dependencies. In Proceedings of the 2019 An- nual Conference of the North American Chapter of the Association for Computational Linguistics. As- sociation for Computational Linguistics.

Cited by

See all 68

Metadata

Authors
van Schijndel, M., Mueller, A., & Linzen, T.
Affiliations
Johns Hopkins University (US)
Venue
Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)
Published
2019
Pages
5830-5836
OpenAlex
W2971016963
Cited by
65
References
30
Citations by year
2026: 1 · 2025: 1 · 2024: 4 · 2023: 12 · 2022: 7 · 2021: 17 · 2020: 23
FWCI
4.59
Type
conference paper
Language
en
Open access
Licence
cc-by
Locations
Conference on Empirical Methods in Natural Language Processing (EMNLP) (publishedVersion, cc-by)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
  • Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
  • Social Sciences › Arts and Humanities › Language and Linguistics › Syntax, Semantics, Linguistic Variation
Keywords
neural language models, syntax, sentence length, model size

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.