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_9640a72d73253957
2022.emnlp-main.262
2,022
Spread estimates intrinsic few-shot hardness using feature-space invariances between training and test samples.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_96017a8e2ac86c85
2022.emnlp-main.262
2,022
Spread accounts for intrinsic few-shot hardness better than existing metrics.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_d5fe44533ee9703c
2022.emnlp-main.262
2,022
Spread is faster to compute than existing metrics for intrinsic few-shot hardness.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_8328b69150eb9005
2023.emnlp-main.679
2,023
FFLM enables zero-shot faithfulness evaluation for text summarization using a moderately-sized foundation language model.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_94b8495c70fdce43
2023.emnlp-main.679
2,023
FFLM measures faithfulness by combining probability changes that occur when prefixing text consistent with the output.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_486c44bbb767a3a6
2023.emnlp-main.679
2,023
FFLM achieves competitive or superior performance to ChatGPT on inconsistency detection and faithfulness rating with a substantially smaller model size.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_f368c911cbde2a6f
2020.emnlp-main.402
2,020
Sequence ranking is the best training objective for in-domain pronoun resolution.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_f98f437464252273
2020.emnlp-main.402
2,020
Semantic similarity between candidates and pronoun is the best training objective for out-of-domain pronoun resolution.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_c1b9f12903f994ca
2020.emnlp-main.402
2,020
Sequence ranking training objective exhibits seed-wise instability in pronoun resolution, but other training objectives do not.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_63a952970e94698d
2021.emnlp-main.802
2,021
XTREME-R is a multilingual natural language understanding benchmark with an improved set of ten tasks.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_c145d666108a1207
2021.emnlp-main.802
2,021
XTREME-R includes challenging language-agnostic retrieval tasks.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_2bd0915a6a3a5a4c
2021.emnlp-main.802
2,021
XTREME-R covers 50 typologically diverse languages.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_f3eca0de4446b107
2021.emnlp-main.802
2,021
The massively multilingual diagnostic suite diagnoses multilingual model behavior.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_b82670f46a3bea44
2021.emnlp-main.802
2,021
Fine-grained multi-dataset evaluation capabilities assess multilingual models.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_2bb2fb461b167191
2025.emnlp-main.451
2,025
Language-Guided Temporal Token Pruning (LGTTP) leverages temporal cues from queries to adaptively prune video tokens while preserving contextual continuity and reducing computational overhead.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_05bf9920eebb57a5
2025.emnlp-main.451
2,025
Language-Guided Temporal Token Pruning (LGTTP) retains higher token density in temporally relevant segments compared to uniform pruning and keyframe selection.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_c0e2d15e6e0fc417
2025.emnlp-main.451
2,025
Language-Guided Temporal Token Pruning (LGTTP) is model-agnostic and integrates with TimeChat and LLaVA-Video.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_c1be86981ffdd3ff
2025.emnlp-main.451
2,025
Language-Guided Temporal Token Pruning (LGTTP) reduces computation while preserving task performance.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_cad427ccdbc32788
2025.emnlp-main.451
2,025
Language-Guided Temporal Token Pruning (LGTTP) excels on queries with explicit temporal markers.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_ed035893bc95a27e
2025.emnlp-main.451
2,025
Language-Guided Temporal Token Pruning (LGTTP) remains effective across general video understanding tasks.
5
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_2edace547078260f
2024.emnlp-main.497
2,024
Jellyfish models are instruction-tuned local large language models that serve as universal data preprocessing task solvers on a local, single, low-priced GPU, ensuring data security and enabling customization.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_b927e038c69f2b03
2024.emnlp-main.497
2,024
Instruction data for data preprocessing tasks is constructed using data configuration, knowledge injection, and reasoning data distillation techniques tailored to data preprocessing.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_9bde7ec7979f4b3c
2024.emnlp-main.497
2,024
Jellyfish models are competitive with GPT-3.5 and GPT-4 in data preprocessing tasks.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_392e4ed8aa65cc4d
2024.emnlp-main.497
2,024
Jellyfish models exhibit strong generalizability to unseen data preprocessing tasks.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_6471385a4aa779da
2024.emnlp-main.497
2,024
Jellyfish models have enhanced reasoning capabilities compared to GPT-3.5.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_d047fdc8b3fe2170
2025.emnlp-main.840
2,025
Database-Augmented Query representation (DAQu) augments the original query with query-related metadata across multiple tables.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_1d4e507011076bea
2025.emnlp-main.840
2,025
Database-Augmented Query representation (DAQu) uses a graph-based set-encoding strategy to encode query-related metadata by considering hierarchies of features in the database without order.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_3ec4e8c18fa51c21
2021.emnlp-main.834
2,021
Three novel features in advanced semantics are used in text readability assessment.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_f114b8b6adc8fb70
2021.emnlp-main.834
2,021
Handcrafted features improve model performance on smaller datasets in text readability assessment.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_d378bc821738238f
2021.emnlp-main.834
2,021
Hybrid models achieve state-of-the-art accuracy in text readability assessment.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_48c79cbccceb266b
2023.emnlp-main.969
2,023
KANGAROO captures the implicit graph structure among entities in closed domains.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_5d70bd04ee6a309c
2023.emnlp-main.969
2,023
KANGAROO uses hyperbolic embeddings of deep hierarchical entity-class structures for knowledge fusion.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_46aa746efff8651f
2023.emnlp-main.969
2,023
KANGAROO constructs hard negative samples via contrastive learning over subgraphs.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_aa6e5af32306415e
2023.emnlp-main.969
2,023
KANGAROO outperforms existing KEPLM training paradigms on knowledge-aware and general NLP tasks in closed domains in both full and few-shot learning settings.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_f9511a32882ec219
2025.emnlp-main.30
2,025
Target-driven multi-modal alignment captures target-specific relationships in multimodal stance detection.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_40b63136b76a64ac
2025.emnlp-main.30
2,025
Dynamic weighting balances modality contributions in multimodal stance detection.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_48d6238506ec389b
2025.emnlp-main.30
2,025
T-MAD achieves robust multimodal stance detection in both in-target and zero-shot settings.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_f24d5b16d1c7eb75
2025.emnlp-main.30
2,025
Target-driven multi-modal alignment enhances multimodal stance detection effectiveness.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_f1850bb883418e42
2025.emnlp-main.30
2,025
Dynamic weighting enhances multimodal stance detection effectiveness.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_d43961080f73ef22
2024.emnlp-main.86
2,024
HiCore learns multi-level user interests by boosting multi-interest self-supervised learning with item-, entity-, and word-oriented multiple-channel hypergraphs.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_f574293a7bd2b4d9
2023.emnlp-main.632
2,023
An exhaustive overview defines the task and challenges of model editing for large language models.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_d41f380a94d625d2
2023.emnlp-main.632
2,023
An in-depth empirical analysis evaluates the most progressive methods for editing large language models.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_895262e3e5546e41
2023.emnlp-main.632
2,023
A new benchmark dataset facilitates robust evaluation and identifies enduring issues in existing techniques for editing large language models.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_223fe981107911e6
2023.emnlp-main.688
2,023
LEAP incorporates explicit planning into the inference procedure for multi-step logical reasoning.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_66d807b00789accd
2023.emnlp-main.688
2,023
Explicit planning makes logical reasoning decisions more informed by looking ahead into future effects.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_f00459164387ba6a
2023.emnlp-main.688
2,023
LEAP uses a training strategy that safeguards explicit planning from spurious features.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_83283e28494bde46
2023.emnlp-main.688
2,023
LEAP with small T5 models achieves competitive performance with GPT-3 on logical reasoning tasks while using significantly fewer parameters.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_0924a54ff176133b
2023.emnlp-main.688
2,023
LEAP with GPT-3.5 outperforms chain-of-thought prompting on the PrOntoQA dataset.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_f3ab8d36ebaf79c9
2023.emnlp-main.688
2,023
Explicit planning is crucial for logical reasoning performance.
5
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_ed8e5c9f873c0188
2021.emnlp-main.843
2,021
Structurally-diverse sampling selects structurally-diverse synthetic utterance-program pairs.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_3e22dbd6e2921881
2021.emnlp-main.843
2,021
A new split of the schema2QA dataset evaluates compositional generalization in semantic parsing.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_0f0ac313f01c436d
2021.emnlp-main.843
2,021
Structurally-diverse sampling improves performance in the traditional i.i.d. setup for semantic parsing.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_966273f04ce30e7e
2021.emnlp-main.843
2,021
Structurally-diverse sampling achieves compositional generalization improvements in semantic parsing with fewer synthetic examples than uniform random sampling.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_17faa441ea9eba75
2025.emnlp-main.1189
2,025
For English and Spanish, monolingual speech in monolingual discourse contexts is prosodically different from monolingual speech in multilingual discourse contexts.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_150a71efd03b725f
2025.emnlp-main.1189
2,025
For English and Spanish, the prosodic differences between monolingual speech in monolingual and multilingual discourse contexts are more marked with increased proximity to multilingual discourse.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_3f538b8df69020f2
2025.emnlp-main.255
2,025
MIO understands and generates speech, text, images, and videos in an end-to-end, autoregressive manner.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_b3a223740d92532e
2025.emnlp-main.255
2,025
MIO is trained on a mixture of discrete tokens across four modalities using causal multimodal modeling.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_a871c7b04471607c
2025.emnlp-main.255
2,025
MIO uses a four-stage training process including alignment pre-training, interleaved pre-training, speech-enhanced pre-training, and comprehensive supervised fine-tuning.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_8cc2f1937b7240ce
2025.emnlp-main.255
2,025
MIO generates interleaved video-text sequences.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_a3c0133662c29fc0
2025.emnlp-main.255
2,025
MIO performs chain-of-visual-thought reasoning.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_1b029a65081bcdbd
2025.emnlp-main.255
2,025
MIO generates visual guidelines.
5
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_44542c97df668fc2
2025.emnlp-main.255
2,025
MIO performs instructional image editing.
6
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_b3c6e4698122138c
2020.emnlp-main.630
2,020
A new formulation for multilingual entity linking resolves language-specific mentions to a language-agnostic Knowledge Base.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_c8849fb2336eb215
2020.emnlp-main.630
2,020
An auxiliary entity-pairing task enhances dual encoder training for multilingual entity linking.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_da3c4f16f3043e78
2020.emnlp-main.630
2,020
A single entity retrieval model covers 100+ languages and 20 million entities for multilingual entity linking.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_12202eb670695681
2020.emnlp-main.630
2,020
A multilingual entity linking model covering 100+ languages outperforms state-of-the-art methods on a limited cross-lingual linking task.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_285de01653806cbd
2020.emnlp-main.630
2,020
Mewsli-9 is a multilingual dataset for entity linking matched to a language-agnostic Knowledge Base setting.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_72be859aaf31f21b
2020.emnlp-main.630
2,020
Frequency-based analysis provides key insights for model and training enhancements in multilingual entity linking.
5
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_9242297bfa5261e2
2020.emnlp-main.626
2,020
MedFilter identifies and categorizes medically relevant utterances in doctor-patient conversations.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_bd3560ea89a6779e
2020.emnlp-main.626
2,020
MedFilter integrates discourse structure and ontological knowledge to identify and categorize medically relevant utterances in doctor-patient conversations.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_ff17cd309f933d04
2020.emnlp-main.626
2,020
MedFilter improves the extraction of symptoms, medications, and complaints in medical conversations.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_2309e8e0d33ac5f2
2025.emnlp-main.911
2,025
N-Gram Trie Speculative Decoding leverages the overlap between context and model output.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_2088cdd9d37eee62
2025.emnlp-main.911
2,025
N-Gram Trie Speculative Decoding constructs an n-gram trie from the context to generate draft tokens.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_1e0b550c9a049e27
2025.emnlp-main.911
2,025
N-Gram Trie Speculative Decoding accelerates token generation for LLMs without compromising accuracy.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_715a47613cc51aad
2024.emnlp-main.113
2,024
A novel Chinese document-level evidence retriever is developed for real-world fact-checking.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_5af13dc14bec85fe
2024.emnlp-main.113
2,024
A Chinese document-level evidence retriever achieves state-of-the-art performance in evidence retrieval.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_5d48627fda649f29
2024.emnlp-main.113
2,024
An adversarial dataset is constructed for analyzing token-level biases in fact-checking systems, where each instance has high word overlap with the original but opposite veracity labels.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_8063ff844f9247e9
2024.emnlp-main.113
2,024
A language-specific fact-checking system for Chinese outperforms translation-based methods and multilingual language models in fact-checking.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_da3a8bd7834888f1
2024.emnlp-main.113
2,024
A language-specific fact-checking system for Chinese is more robust toward token-level biases.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_eb69da9d630533a3
2024.emnlp-main.573
2,024
Translating images to make them culturally relevant is a task in machine translation.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_5b576bae9e8a0c04
2024.emnlp-main.573
2,024
Three pipelines comprising state-of-the-art generative models perform image translation for cultural relevance.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_3877054b403d8a99
2024.emnlp-main.573
2,024
An evaluation dataset for image translation for cultural relevance has two parts: a concept part with 600 cross-culturally coherent images focusing on a single concept per image and an application part with 100 images curated from real-world applications.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_77bf424612991633
2024.emnlp-main.573
2,024
The failure of image-editing models to translate images for cultural relevance is improved by leveraging LLMs and retrievers in the loop.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_c8bf3d0f51d8eb03
2024.emnlp-main.573
2,024
Current pipelines for image translation for cultural relevance fail for some countries in real-world applications.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_ad475d56585133df
2022.emnlp-main.543
2,022
An intent-slot co-occurrence graph captures global intent-slot correlation from the entire training corpus.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_c5a934c82a514d1c
2022.emnlp-main.543
2,022
A graph neural network models the interaction between intent detection and slot filling using global intent-slot co-occurrence.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_6c3bd4f94e17308b
2022.emnlp-main.543
2,022
Using global intent-slot co-occurrence improves joint multiple intent detection and slot filling over state-of-the-art models.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_4bf05a6f9f82b041
2023.emnlp-main.105
2,023
SCALE is a task-agnostic model that detects factual inconsistencies in long documents using a Natural Language Inference approach with large text chunks.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_1effdbdd7939a8c8
2023.emnlp-main.105
2,023
SCALE achieves state-of-the-art performance in factual inconsistency detection for diverse tasks and long inputs.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_89009021c2316f9d
2023.emnlp-main.105
2,023
SCALE explains decisions through relevant source sentence retrieval using a novel algorithm.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_319f17f91ce72766
2023.emnlp-main.105
2,023
ScreenEval is a long-form dialogue dataset for factual inconsistency detection.
3
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_42db8f77bded7a56
2023.emnlp-main.105
2,023
SCALE is more efficient than existing methods for factual inconsistency detection.
4
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_bfa80f1eec7a3309
2021.emnlp-main.753
2,021
Value-aware objective incorporates value vectors in attention approximation.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_c429e4bf12e11b5d
2021.emnlp-main.753
2,021
In language modeling, optimal attention approximation incorporating value vectors outperforms optimal attention approximation ignoring value vectors.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_76bc47dbcd604dee
2021.emnlp-main.753
2,021
For less skewed kernel functions, value vectors have a greater impact on sparse attention approximation quality.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_b8033a19f2fb3610
2022.emnlp-main.136
2,022
ToKen decomposes the few-shot hate speech detection task into constituent parts.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_b670a97c6a7a13a5
2022.emnlp-main.136
2,022
ToKen infuses knowledge from reasoning datasets such as ATOMIC2020 for few-shot hate speech detection.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_3369bb6aaa5ecd0a
2022.emnlp-main.136
2,022
ToKen generalizes to out-of-distribution datasets in few-shot hate speech detection.
2
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_8a27db7bd1e41633
2023.emnlp-main.91
2,023
UL2R improves language models and their scaling curves with minimal extra compute and without new data sources.
0
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00
clm_b98f08225bd60764
2023.emnlp-main.91
2,023
UL2R continues training a state-of-the-art large language model with UL2’s mixture-of-denoiser objective for a few additional steps.
1
qwen/qwen3-235b-a22b-thinking-2507
f9a9b434c72c
extract_20260524T085601Z
2026-05-24T08:56:02.723585+00:00