claim_id
large_stringlengths
20
20
paper_id
large_stringlengths
17
20
year
int32
2.02k
2.03k
atomic_claim
large_stringlengths
18
387
claim_index
int32
0
8
extractor_model
large_stringclasses
1 value
prompt_hash
large_stringclasses
1 value
run_id
large_stringclasses
2 values
created_at
large_stringclasses
2 values
clm_0dedcbb2666f727e
2025.emnlp-main.969
2,025
The benchmark for solving linguistic puzzles is extended to include Linguistic Olympiad puzzles.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_b9a0f63cf2525b1d
2025.emnlp-main.969
2,025
Large Language Models outperform humans on most types of linguistic puzzles, except for puzzles centered on writing systems and puzzles involving understudied languages.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_fb457533d5e584f2
2020.emnlp-main.600
2,020
Data augmentation for active learning uses labeled guided outputs from a language generation model to train text classifiers.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_82ce7b5c330aee3d
2020.emnlp-main.600
2,020
Text generation for data augmentation is optimized using Monte Carlo Tree Search to maximize output usefulness with entropy as an optimization criterion.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_829e92d79d0662e5
2020.emnlp-main.600
2,020
Guided text generation with Monte Carlo Tree Search improves text classification performance on the TREC-6 and SST-2 datasets compared to non-guided data generation.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_931c9b1b4d05a2e2
2021.emnlp-main.539
2,021
Levenshtein Training uses the Levenshtein Transformer for word-level quality estimation.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_783e2e6ca5e381a3
2021.emnlp-main.539
2,021
A two-stage transfer learning procedure that uses augmented data and human post-editing data reduces the mismatch between machine translation and word-level quality estimation tasks.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_81c8e4d7e2a1092f
2021.emnlp-main.539
2,021
Heuristics for word-level quality estimation construct reference labels compatible with subword-level finetuning and inference.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_f3acb7e23cc4a09b
2021.emnlp-main.539
2,021
Levenshtein Training has superior data efficiency for word-level quality estimation in data-constrained settings.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_4545aa61f2031441
2021.emnlp-main.539
2,021
Levenshtein Training achieves competitive performance for word-level quality estimation in unconstrained settings.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_68b67acc19207263
2023.emnlp-main.803
2,023
Imbalanced label distribution in question answering causes model collapse during test-time adaptation.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_039f0b619b75e7df
2023.emnlp-main.803
2,023
Anti-CF utilizes the source model's output to regularize the update of the adapted model during test time.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_5f4457444bd0a71c
2023.emnlp-main.803
2,023
Anti-CF includes an efficient side block that reduces inference time.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_6fc72ea382dcc4ce
2023.emnlp-main.803
2,023
Anti-CF achieves comparable or better results than previous test-time adaptation methods.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_380f3c3c92cd2f6a
2023.emnlp-main.803
2,023
Anti-CF maintains inference speed close to that of vanilla forward propagation.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_662e788aa6cca767
2022.emnlp-main.85
2,022
Fine-grained Category Discovery under Coarse-grained supervision (FCDC) discovers fine-grained categories using only coarse-grained labeled data.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_5a7e0562b951193c
2022.emnlp-main.85
2,022
Hierarchical weighted self-contrastive learning combines a weighted self-contrastive module with supervised learning in a hierarchical manner to discover fine-grained categories from coarse-grained labeled data.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_4ae764b646e211fd
2025.emnlp-main.1041
2,025
Modern large language models can be effectively adapted for information reranking using minimal, high-quality supervision.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_8214864722e662f8
2025.emnlp-main.1041
2,025
LIMRANK-SYNTHESIZER generates diverse, challenging, and realistic reranking examples.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_6eb6204a648e4e24
2025.emnlp-main.1041
2,025
LIMRANK is a reranker model for reasoning-intensive information reranking.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_54427430bb939265
2025.emnlp-main.1041
2,025
LIMRANK achieves competitive performance for information reranking with minimal training data.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_6f1573f04f349d27
2025.emnlp-main.1041
2,025
LIMRANK generalizes to downstream tasks including scientific literature search and retrieval-augmented generation for knowledge-intensive problem solving.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_52fa922087765119
2023.emnlp-main.631
2,023
The information gap between dialogue contexts and desired inferences renders the inductive inference process more challenging.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_69b39ac367212260
2023.emnlp-main.631
2,023
Negative samples enable dialogue models to identify incorrect inferences.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_374f87805e5adda7
2021.emnlp-main.695
2,021
A new Logical Form grammar models a wide range of knowledge graph queries and is sufficiently simple for efficient supervision data generation.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_05f2129942ddac21
2021.emnlp-main.695
2,021
A Transformer-based model incorporates knowledge graph and conversational contexts via a JSON-like structured input, which is transformed to embeddings and processed by standard attention layers.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_eeb014e2fb61fb98
2021.emnlp-main.695
2,021
A conversational question answering system over knowledge graphs with high-coverage grammar and structured context achieves higher grammar coverage and logical form execution accuracy on the CSQA dataset than prior work.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_0322ef7023fa3ed3
2022.emnlp-main.483
2,022
RLET performs iterative single step reasoning for entailment tree generation using sentence selection and deduction generation modules.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_a690cf4da900a08e
2022.emnlp-main.483
2,022
RLET accumulates training signals across the entailment tree using an aligned reward function consistent with the evaluation.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_bfd91a275985fef4
2022.emnlp-main.483
2,022
Reinforcement learning is introduced for entailment tree generation for the first time.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_458cf63c08b05381
2023.emnlp-main.491
2,023
A systematic evaluation method assesses multilingual capabilities of large language models using qualitative and quantitative approaches.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_789a621bc3689275
2023.emnlp-main.491
2,023
A novel prompt back-translation method evaluates cross-language generalization in large language models.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_62828758586eb499
2023.emnlp-main.491
2,023
Large language models effectively transfer learned knowledge across different languages in tasks where the correct output is independent of the input language.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_007abdafee71c841
2023.emnlp-main.491
2,023
Large language models struggle to provide accurate results in tasks where the correct output depends on the input language.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_42e38c535fafc291
2025.emnlp-main.590
2,025
CEMTM infers coherent and interpretable topic structures from short and long documents containing text and images.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_791cbfb4a4dd33a7
2025.emnlp-main.590
2,025
CEMTM uses fine-tuned large vision language models to obtain contextualized embeddings.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_d0735c405d72c58b
2025.emnlp-main.590
2,025
CEMTM employs a distributional attention mechanism to weight token-level contributions to topic inference.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_7ef321ff6bb00d54
2025.emnlp-main.590
2,025
CEMTM processes multiple images per document without repeated encoding.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_0bbe5dab51790b1b
2025.emnlp-main.590
2,025
CEMTM captures visually grounded semantics in complex domains such as scientific articles.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_603d005331f18e9f
2021.emnlp-main.101
2,021
MS-Mentions provides entity mention annotations for Material Science synthesis procedural texts.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_f1893129e0ad3067
2021.emnlp-main.101
2,021
A new label inventory is designed to provide consistent entity mention annotations for Material Science synthesis procedural texts.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_d4c186375c07a46b
2021.emnlp-main.101
2,021
A new annotation approach is designed to maximize consistency and annotation speed for domain experts annotating Material Science synthesis procedural texts.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_efd6fc90606e9b2c
2021.emnlp-main.101
2,021
MS-Mentions entity mention annotations achieve high inter-annotator agreement.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_39465c10b97a7732
2021.emnlp-main.233
2,021
Lifelong Explanation (LLE) continuously trains a student explainer under the supervision of a teacher explanation algorithm on different tasks in lifelong learning.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_9737005cfa7f7438
2021.emnlp-main.233
2,021
Lifelong Explanation (LLE) uses Experience Replay to prevent catastrophic forgetting in the student explainer.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_7adeced41cc31ef4
2021.emnlp-main.233
2,021
Lifelong Explanation (LLE) enhances the stability of explanations for all seen tasks.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_2d1194454904950d
2021.emnlp-main.233
2,021
Lifelong Explanation (LLE) maintains the same level of faithfulness to the black-box model as the teacher explanation algorithm.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_d158810991f6e3bf
2021.emnlp-main.233
2,021
Lifelong Explanation (LLE) is faster at test time than existing explanation methods.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_8003d20814746e79
2021.emnlp-main.233
2,021
The Experience Replay mechanism in Lifelong Explanation (LLE) enhances the learning capabilities of the student explainer.
5
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_02c8ab5f0243dd3f
2021.emnlp-main.264
2,021
Scheduled sampling based on decoding steps increases the selection chance of predicted tokens as decoding steps increase.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_92af2eb971b45bdc
2021.emnlp-main.264
2,021
Scheduled sampling based on decoding steps more realistically simulates the inference scene during training.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_09f7451675758c94
2021.emnlp-main.264
2,021
Scheduled sampling based on both training steps and decoding steps improves neural machine translation performance.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_e847e05855e6dd78
2021.emnlp-main.264
2,021
Scheduled sampling based on decoding steps outperforms the Transformer baseline and vanilla scheduled sampling on neural machine translation tasks.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_ffca3f16ff9c04b7
2021.emnlp-main.264
2,021
Scheduled sampling based on decoding steps generalizes to text summarization tasks.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_e948d30e25f61afa
2024.emnlp-main.72
2,024
The predictive preferences of large language models are highly brittle and skewed even with semantically equivalent instructions.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_7c2110dd151f38cd
2024.emnlp-main.72
2,024
Fairer predictive preferences from large language models lead to judgments that are better aligned with humans.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_21d02f23498289e7
2024.emnlp-main.72
2,024
ZEPO produces fairer preference decisions and improves the alignment of large language model evaluators with human judgments without requiring labeled data.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_c8a08d6072c40922
2024.emnlp-main.72
2,024
ZEPO uses a zero-shot learning objective based on preference decision fairness.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_9394ecabeba37e0b
2021.emnlp-main.57
2,021
Training with gold AMR and machine-translated sentences improves multilingual AMR-to-text generation over training with generated silver AMR.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_254e44ab1641bf16
2021.emnlp-main.57
2,021
Training with a combination of gold AMR with machine-translated sentences and generated silver AMR improves multilingual AMR-to-text generation.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_9a81ab0137b0b9b6
2021.emnlp-main.57
2,021
Training with a combination of gold AMR with machine-translated sentences and generated silver AMR sets a new state of the art for multilingual AMR-to-text generation in German, Italian, Spain, and Chinese.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_c532a92ace58cbb6
2021.emnlp-main.199
2,021
ConRPG uses a context regularizer for meaning in paraphrase generation.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_9c2ffbd7513906c1
2021.emnlp-main.199
2,021
ConRPG generates massive amounts of high-quality paraphrase pairs.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_3bf7ad3a5eada2af
2021.emnlp-main.199
2,021
ConRPG produces paraphrases with more lexical and syntactic diversity.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_a85a5ee91cd2db51
2021.emnlp-main.199
2,021
ConRPG allows developer intervention in the paraphrase generation process through human-interpretable scoring functions, resulting in a more controllable model.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_1a2c56cc00c816d1
2021.emnlp-main.199
2,021
ConRPG outperforms existing paraphrase generation methods in both supervised and unsupervised setups.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_7e7e82fe449397f0
2020.emnlp-main.132
2,020
A pre-trained encoder for entity relation extraction uses a span encoder and a span pair encoder to incorporate intra-span and inter-span information.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_483c129cab257028
2020.emnlp-main.132
2,020
A span encoder recovers the original token order within a span from a randomly shuffled sequence.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_880c8b962f08d0bb
2020.emnlp-main.132
2,020
A span pair encoder distinguishes span pairs from the same sentence as positive and from different sentences as negative using contrastive loss.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_472c036d5a42b63e
2020.emnlp-main.132
2,020
In entity relation extraction, pre-training with intra-span and inter-span information outperforms distantly supervised pre-training.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_3526133d46d8c23d
2021.emnlp-main.729
2,021
A collaborative learning framework for unsupervised text style transfer uses a pair of decoders with opposite decoding directions.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_6dc668ebf9a0cfe1
2021.emnlp-main.729
2,021
A collaborative learning framework for unsupervised text style transfer regularizes each decoder by knowledge from the peer decoder.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_0df1a7f0793efd0e
2021.emnlp-main.729
2,021
A collaborative learning framework for unsupervised text style transfer guarantees the difference in knowledge acquisition between decoders by opposite decoding directions and a distinguishability constraint.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_8d5f3747a9e51b59
2021.emnlp-main.729
2,021
A collaborative learning framework for unsupervised text style transfer employs mutual knowledge distillation to drive both decoders to a better optimum and alleviate the over-transfer and under-transfer problems.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_f255826145635a2d
2020.emnlp-main.403
2,020
NSP is detrimental to BERT pre-training due to context splitting and shallow semantic signal.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_ff2b30c11fbc660d
2020.emnlp-main.403
2,020
Sentence ordering, adjacent sentence prediction, TF prediction, TF-IDF prediction, a FastSent variant, and a Quick Thoughts variant yield better performance than pure masked language modeling when used as auxiliary pre-training tasks.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_c067ed292d178415
2020.emnlp-main.403
2,020
Multi-task pre-training with multiple auxiliary tasks yields better model performance than pre-training with any single auxiliary task.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_bc3080acdd1a5433
2020.emnlp-main.403
2,020
A language model pre-trained with multiple auxiliary tasks outperforms BERTBase on the GLUE benchmark with substantially reduced training data.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_522c41178195c2dc
2020.emnlp-main.736
2,020
UNION evaluates the quality of generated stories in open-ended story generation without references.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_8594c78774e7cd7b
2020.emnlp-main.736
2,020
UNION distinguishes human-written stories from negative samples.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_1804f4f6f058eec2
2020.emnlp-main.736
2,020
UNION recovers the perturbation in negative stories.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_c29a7e2e1e7d238d
2020.emnlp-main.736
2,020
Negative samples for open-ended story generation evaluation mimic repeated plots, conflicting logic, and long-range incoherence errors from existing NLG models.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_2098941135495550
2020.emnlp-main.736
2,020
UNION correlates better with human judgments for open-ended story generation evaluation than existing state-of-the-art metrics.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_2c302c4fb3a2e249
2020.emnlp-main.736
2,020
UNION is more generalizable for open-ended story generation evaluation than existing state-of-the-art metrics.
5
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_8eb16e538660576f
2024.emnlp-main.962
2,024
MedCoT performs Medical Visual Question Answering using a hierarchical chain of specialists where an Initial Specialist proposes diagnostic rationales, a Follow-up Specialist validates these rationales, and a sparse Mixture of Experts within a Diagnostic Specialist reaches consensus through a vote for definitive diagno...
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_79caeb009f7f84a3
2024.emnlp-main.962
2,024
MedCoT improves performance and interpretability in Medical Visual Question Answering.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_4a736f6adb30e922
2022.emnlp-main.481
2,022
FLUTE is a dataset of figurative language natural language inference instances with textual explanations for Sarcasm, Simile, Metaphor, and Idioms.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_247e277c2d720ac6
2022.emnlp-main.481
2,022
A Human-AI collaboration framework utilizing GPT-3 and human annotators (novices and experts) aids in scaling up the creation of datasets for complex linguistic phenomena such as figurative language.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_ee4dd43110927ca8
2023.emnlp-main.965
2,023
Measuring variance of word vector distributions detects semantic differences in words appearing in two corpora without requiring corpus or word alignments.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_2bc0af6932ef790a
2023.emnlp-main.965
2,023
The norm of the mean word vector measures the coverage of meanings of a word in a corpus.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_dad8ca5220026455
2023.emnlp-main.965
2,023
Measuring variance of word vector distributions is robust to corpus size skew.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_c288cd0d30533b06
2023.emnlp-main.965
2,023
Measuring variance of word vector distributions detects semantic differences in infrequent words appearing in two corpora.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_c28e3ed02984611c
2023.emnlp-main.965
2,023
Measuring variance of word vector distributions pinpoints word instances with a meaning missing in one of two corpora.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_f4cc23f133bcafea
2024.emnlp-main.727
2,024
Plan-guided retrieval uses the planning capabilities of instruction-tuned LLMs to guide retrieval for grounded text generation.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_a8c69e8912bf20b7
2024.emnlp-main.727
2,024
Plan-guided retrieval improves the coverage of relevant facts in long-form text generation.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_4aba05fdcbc45860
2024.emnlp-main.727
2,024
Plan-guided retrieval produces more informative responses in long-form text generation.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_3f497a291e78a48c
2024.emnlp-main.727
2,024
Plan-guided retrieval provides a higher rate of attribution to source documents in long-form text generation.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_946d84cb8d8362df
2022.emnlp-main.70
2,022
Authorship Representation Regularization (ARR) creates authorship representation with reduced reliance on topic-specific information.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_70ffc191b7923eec
2022.emnlp-main.70
2,022
A cross-topic-open-set evaluation method evaluates authorship representation performance for unseen authors and unseen topics.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_6e0b37a5496ca4ff
2022.emnlp-main.262
2,022
Few-shot hardness is intrinsic to datasets for a given pre-trained model.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00