Challenge: Psycholinguists have identified one such cue in the implicit causality bias of interpersonal verbs.
Approach: They propose to use pre-trained language models to encode IC bias at inference time . they hypothesize that when a cause is explicitly stated, an incongruent IC biased leads to a delay in human processing.
Outcome: The results suggest that pre-trained language models tend to prioritize lexical patterns over higher-order signals.

Similar Papers

A Causal View of Entity Bias in (Large) Language Models (2023.findings-emnlp)

Copied to clipboard

Challenge: Entity bias affects pretrained (large) language models, causing them to rely on (biased) parametric knowledge to make unfaithful predictions.
Approach: They propose a structured causal model whose parameters are easier to estimate . they propose to perturb the original entity with neighboring entities .
Outcome: The proposed model reduces biasing information pertaining to the original entity while still preserving sufficient semantic information from similar entities.
LLMs Are Prone to Fallacies in Causal Inference (2024.emnlp-main)

Copied to clipboard

Challenge: Recent work shows that causal facts can be extracted from LLMs through prompting . but it is unclear if this success is limited to explicitly-mentioned causal facts in pretraining data .
Approach: They fine tune LLMs on synthetic data and test whether they can infer causal relations . they find that LLM can correctly deduce absence of causal relations from temporal and spatial relations if order is randomized .
Outcome: The proposed model outperforms existing methods on causal inference tasks.
When Do Pre-Training Biases Propagate to Downstream Tasks? A Case Study in Text Summarization (2023.eacl-main)

Copied to clipboard

Challenge: Existing studies have shown that large language models contain linguistic and societal biases, but it is unclear how these biase amplify to downstream tasks.
Approach: They investigate how name-nationality bias propagates from pre-training to downstream tasks . they show that these biases manifest themselves as hallucinations in summarization .
Outcome: The proposed model can reduce the rate of hallucinations, but does not change the types of biases that do appear.
“You are grounded!”: Latent Name Artifacts in Pre-trained Language Models (2020.emnlp-main)

Copied to clipboard

Challenge: Pre-trained language models perpetuate biases originating in their training corpus to downstream models.
Approach: They focus on the representations of given names in pre-trained language models and show that name perturbation can have an effect on downstream tasks.
Outcome: The proposed model can be used to model the representation of given names in pre-trained language models on reading comprehension probes where name perturbation changes the model answers.
Between Circuits and Chomsky: Pre-pretraining on Formal Languages Imparts Linguistic Biases (2025.acl-long)

Copied to clipboard

Challenge: Pretraining language models on formal language can improve their acquisition of natural language, but it is computationally limited.
Approach: They hypothesize that effective transfer occurs when the formal language captures hierarchical dependencies present in natural language and remains within the computational limitations of the model architecture.
Outcome: The proposed model can achieve lower loss on natural language and better generalization with a 33% smaller token budget than training on a matched amount of natural language.
Large Language Models Are Partially Primed in Pronoun Interpretation (2023.findings-acl)

Copied to clipboard

Challenge: Existing studies suggest large language models acquire rich linguistic representations, but little is known about whether they adapt to linguistic biases in a human-like way.
Approach: They examine whether large language models display human-like referential biases using stimuli and procedures from real psycholinguistic experiments.
Outcome: The proposed models display human-like referential biases when exposed to referential patterns in the local context.
On the Reliability of Large Language Models for Causal Discovery (2025.acl-long)

Copied to clipboard

Challenge: Existing statistical methods to identify causal relationships from observational data remain elusive.
Approach: They examine the impact of memorization for accurate causal relation prediction, the influence of incorrect causal relations in pre-training data and the contextual nuances that influence LLMs’ understanding of causal relations.
Outcome: The proposed models are effective in recognizing causal relations that occur frequently in pre-training data, but their ability to generalize to new or rare causal relations is limited.
Explicit vs. Implicit: Investigating Social Bias in Large Language Models through Self-Reflection (2025.findings-acl)

Copied to clipboard

Challenge: Existing methods to quantify and quantify social biases in Large Language Models (LLMs) focus on explicit bias, with little attention to implicit bias.
Approach: They propose a self-reflection-based evaluation framework that measures implicit bias and evaluates explicit bias by prompting LLMs to analyze their own generated content.
Outcome: The proposed framework compares explicit and implicit biases in large language models . it demonstrates that explicit bias manifests as mild stereotypes, while implicit bias exhibits strong stereotypes.
Can Large Language Models Learn Independent Causal Mechanisms? (2024.emnlp-main)

Copied to clipboard

Challenge: Large Language Models (LLMs) perform poorly on complex reasoning tasks, such as abstract, causal, or logical reasoning.
Approach: They propose to use two concepts from causality to learn ICMs within LLMs to improve out-of-distribution performance on abstract and causal reasoning tasks.
Outcome: The proposed model outperforms existing models on abstract and causal reasoning tasks and is more robust to fine-tuning.
Challenging the Explanation Based on Preceding Tokens: Discovering Transferable Non-Literal Biasing (2026.acl-short)

Copied to clipboard

Challenge: et al. (2017) show that the generated preceding tokens may push the large language model towards the target answer.
Approach: They find that generated preceding tokens may push large language models towards the target answer . they suggest that the LLM may intentionally use the semantically unrelated tokens to help generation of the target .
Outcome: The generated preceding tokens may push the large language model towards the target answer . the biased connotations of the target response can also transfer to other prompts .

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations