mlsanityData to 7 Oct 2026

When a sentence does not introduce a discourse entity, Transformer-based models still sometimes refer to it

arXiv 2205.03472 ↗First posted 6 May 2022cs.CLcited by 22

Not on the map. Why?

Abstract (arXiv)

Understanding longer narratives or participating in conversations requires tracking of discourse entities that have been mentioned. Indefinite noun phrases (NPs), such as 'a dog', frequently introduce discourse entities but this behavior is modulated by sentential operators such as negation. For example, 'a dog' in 'Arthur doesn't own a dog' does not introduce a discourse entity due to the presence of negation. In this work, we adapt the psycholinguistic assessment of language models paradigm to higher-level linguistic phenomena and introduce an English evaluation suite that targets the knowledge of the interactions between sentential operators and indefinite NPs. We use this evaluation suite for a fine-grained investigation of the entity tracking abilities of the Transformer-based models GPT-2 and GPT-3. We find that while the models are to a certain extent sensitive to the interactions we investigate, they are all challenged by the presence of multiple NPs and their behavior is not systematic, which suggests that even models at the scale of GPT-3 do not fully acquire basic entity tracking abilities.

Key facts. Cited by 22 papers in the map (data through 7 Oct 2026). Builds on 26 of its 42 references that are in the map. First posted on arXiv 6 May 2022.

Builds on

26 of 42 references.

References · 26

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  2. 2020Language Models are Few-Shot Learnerscited by 32,199
  3. 2019Transformer-XL: Attentive Language Models Beyond a Fixed-Length Contextcited by 1,830
  4. 2019BERT Rediscovers the Classical NLP Pipelinecited by 1,046
  5. 2019What Does BERT Look At? An Analysis of BERT's Attentioncited by 994
  6. 2016The LAMBADA dataset: Word prediction requiring a broad discourse contextcited by 960
  7. 2019BLiMP: The Benchmark of Linguistic Minimal Pairs for Englishcited by 553
  8. 2014Memory Networkscited by 354
  9. 2016Assessing the Ability of LSTMs to Learn Syntax-Sensitive Dependenciescited by 350
  10. 2019What BERT is not: Lessons from a new suite of psycholinguistic diagnostics for language modelscited by 317
  11. 2021All That's 'Human' Is Not Gold: Evaluating Human Evaluation of Generated Textcited by 287
  12. 2018Targeted Syntactic Evaluation of Language Modelscited by 239
14 more
  1. 2021The GEM Benchmark: Natural Language Generation, its Evaluation and Metricscited by 239
  2. 2018Colorless green recurrent networks dream hierarchicallycited by 210
  3. 2021Implicit Representations of Meaning in Neural Language Modelscited by 179
  4. 2020A Systematic Assessment of Syntactic Generalization in Neural Language Modelscited by 163
  5. 2019Neural Language Models as Psycholinguistic Subjects: Representations of Syntactic Statecited by 137
  6. 2020PlotMachines: Outline-Conditioned Generation with Dynamic Plot State Trackingcited by 104
  7. 2018What do RNN Language Models Learn about Filler-Gap Dependencies?cited by 86
  8. 2021Provable Limitations of Acquiring Meaning from Ungrounded Form: What Will Future Language Models Understand?cited by 55
  9. 2019Probing What Different NLP Tasks Teach Machines about Function Word Comprehensioncited by 44
  10. 2021Refining Targeted Syntactic Evaluation of Language Modelscited by 41
  11. 2021Sorting through the noise: Testing robustness of information processing in pre-trained language modelscited by 27
  12. 2021Pragmatic competence of pre-trained language models through the lens of discourse connectivescited by 16
  13. 2021The Language Model Understood the Prompt was Ambiguous: Probing Syntactic Uncertainty Through Generationcited by 15
  14. 2019Attending to Entities for Better Text Understandingcited by 6

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 4; 2023: 6; 2024: 6; 2025: 5; 2026: 1.

22 papers.

Sort:

By area:

  1. 2022Neural Theory-of-Mind? On the Limits of Social Intelligence in Large LMscited by 220
  2. 2022Language Models as Agent Modelscited by 131
  3. 2023Language Model Behavior: A Comprehensive Surveycited by 92
  4. 2023Eight Things to Know about Large Language Modelscited by 73
  5. 2023Entity Tracking in Language Modelscited by 44
  6. 2022Can language models handle recursively nested grammatical structures? A case study on comparing models and humanscited by 41
  7. 2023Dissociating language and thought in large language modelscited by 27
  8. 2022Model Criticism for Long-Form Text Generationcited by 21
  9. 2024Scope Ambiguities in Large Language Modelscited by 11
  10. 2024RST-LoRA: A Discourse-Aware Low-Rank Adaptation for Long Document Abstractive Summarizationcited by 10
  11. 2024Do Large Language Models Mirror Cognitive Language Processing?cited by 9
  12. 2023Understanding Natural Language Understanding Systems. A Critical Analysiscited by 4
10 more
  1. 2024Experimental Contexts Can Facilitate Robust Semantic Property Inference in Language Models, but Inconsistentlycited by 3
  2. 2025Language Models Largely Exhibit Human-like Constituent Ordering Preferencescited by 1
  3. 2025Meaning Beyond Truth Conditions: Evaluating Discourse Level Understanding via Anaphora Accessibilitycited by 1
  4. 2025Hey, wait a minute: on at-issue sensitivity in Language Modelscited by 1
  5. 2025Mary, the Cheeseburger-Eating Vegetarian: Do LLMs Recognize Incoherence in Narratives?cited by 1
  6. 2023On General Language Understandingcited by 0
  7. 2024LIEDER: Linguistically-Informed Evaluation for Discourse Entity Recognitioncited by 0
  8. 2024To Drop or Not to Drop? Predicting Argument Ellipsis Judgments: A Case Study in Japanesecited by 0
  9. 2025Do Large Language Models know who did what to whom?cited by 0
  10. 2026A Circuit for Plural Reference: How LLMs Represent and Retrieve Singular and Plural Entitiescited by 0