The LAMBADA dataset: Word prediction requiring a broad discourse context¶
Authors: D. Paperno; G. Kruszewski; A. Lazaridou; N. Q. Pham; R. Bernardi; S. Pezzelle; M. Baroni; G. Boleda; R. Fern´andez
Year: 2016
type: dataset collection: sok-references status: auto
Links: link
Summary¶
Auto-extracted summary (unreviewed)
Introduces LAMBADA, a word-prediction benchmark of narrative passages whose final word humans can guess from the full passage but not from the last sentence alone, so that models must track broad discourse context. In preliminary experiments, n-gram, RNN, LSTM and Memory Network language models all score below 1% accuracy on LAMBADA while reaching roughly 15-22% on a control set, and none beat the heuristic of picking a random capitalized word from the passage.
Abstract¶
We introduce LAMBADA, a dataset to evaluate the capabilities of computational models for text understanding by means of a word prediction task. LAMBADA is a collection of narrative passages sharing the characteristic that human subjects are able to guess their last word if they are exposed to the whole passage, but not if they only see the last sentence preceding the target word. To succeed on LAM- BADA, computational models cannot simply rely on local context, but must be able to keep track of information in the broader discourse. We show that LAMBADA exemplifies a wide range of linguistic phenomena, and that none of several state-ofthe-art language models reaches accuracy above 1% on this novel benchmark. We thus propose LAMBADA as a challenging test set, meant to encourage the development of new models capable of genuine understanding of broad context in natural language text.
Source: pdf
Extracted details¶
| Field | Value | Source | Confidence |
|---|---|---|---|
| contribution | LAMBADA dataset built with human filtering so the target word requires broad context | llm | 0.90 |
| contribution | Analysis of the linguistic phenomena exemplified by the passages | llm | 0.90 |
| contribution | Baseline evaluation of n-gram, RNN, LSTM, Memory Network and CBOW models | llm | 0.90 |
| contribution | Control set to show models perform normally on standard language modeling | llm | 0.90 |
| contribution | Planned public competition and release of development set and training corpus | llm | 0.90 |
| models | N-gram LM | llm | 0.90 |
| models | N-gram LM with cache | llm | 0.90 |
| models | RNN | llm | 0.90 |
| models | LSTM | llm | 0.90 |
| models | Memory Network | llm | 0.90 |
| models | CBOW (supervised and unsupervised) | llm | 0.90 |
| datasets | LAMBADA | llm | 0.90 |
| tasks | language modeling | llm | 0.90 |
| tasks | word prediction | llm | 0.90 |
| tasks | text understanding | llm | 0.90 |
| limitations | Evaluation is preliminary and only a proof-of-concept; models were not extensively tuned | llm | 0.90 |
| limitations | By construction standard language models are bound to fail, since passages were filtered by simple LMs failing on them | llm | 0.90 |
| key result | accuracy: <= 0.1% for all tested models (LAMBADA test passages) | llm | 0.90 |
| key result | accuracy: 7.3% (random capitalized word from passage heuristic on LAMBADA) | llm | 0.90 |
| key result | perplexity: 768 (N-gram with cache on LAMBADA (best model)) | llm | 0.90 |
| key result | accuracy: 21.9% (LSTM on control set) | llm | 0.90 |
| key result | accuracy: 19.1% (N-gram models on control set) | llm | 0.90 |
SoK reference [47]¶
D. Paperno, G. Kruszewski, A. Lazaridou, N. Q. Pham, R. Bernardi, S. Pezzelle, M. Baroni, G. Boleda, and R. Fern´andez, “The LAMBADA dataset: Word prediction requiring a broad discourse context,” in Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), K. Erk and N. A. Smith, Eds. Berlin, Germany: Association for Computational Linguistics, Aug. 2016, pp. 1525– 1534. [Online]. Available: https://aclanthology.org/P16-1144/
Listed on the reference page of the site.
Keyword mentions¶
Counts of vocabulary terms in the full text (first page of occurrence in brackets).
- scheme: Partially homomorphic encryption 2 [p.1]
- model: RNN / LSTM 16 [p.4], CNN 1 [p.2]
- dataset: LAMBADA 69 [p.1]
- hardware: GPU 1 [p.9]
- task: Language modeling / generation 4 [p.1], Named entity recognition 2 [p.2]
- topic: Benchmark 6 [p.1]
Files¶
| Path (under papers/) | Collection | Pages | Size | SHA-256 | Note |
|---|---|---|---|---|---|
references/sok-references/047_the-lambada-dataset-word-prediction-requiring-a-broad-discou.pdf |
sok-references | 10 | 151.7 KB | 781dd5a29bcf |
Database key: the-lambada-dataset-word-prediction-requiring-a-broad-discourse-context · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set the-lambada-dataset-word-prediction-requiring-a-broad-discourse-context COLUMN VALUE