John praised Mary because he? Implicit Causality Bias and Its\n Interaction with Explicit Cues in LMs

2021/06/02 by Yova Kementchedjhieva, Mark Anderson, Kementchedjhieva, Yova +3 · 1 citation
Computer Science · #Topic Modeling #Natural Language Processing Techniques #Text Readability and Simplification

paper · pdf · doi:10.48550/arxiv.2106.01060

Abstract

Some interpersonal verbs can implicitly attribute causality to either their\nsubject or their object and are therefore said to carry an implicit causality\n(IC) bias. Through this bias, causal links can be inferred from a narrative,\naiding language comprehension. We investigate whether pre-trained language\nmodels (PLMs) encode IC bias and use it at inference time. We find that to be\nthe case, albeit to different degrees, for three distinct PLM architectures.\nHowever, causes do not always need to be implicit -- when a cause is explicitly\nstated in a subordinate clause, an incongruent IC bias associated with the verb\nin the main clause leads to a delay in human processing. We hypothesize that\nthe temporary challenge humans face in integrating the two contradicting\nsignals, one from the lexical semantics of the verb, one from the\nsentence-level semantics, would be reflected in higher error rates for models\non tasks dependent on causal links. The results of our study lend support to\nthis hypothesis, suggesting that PLMs tend to prioritize lexical patterns over\nhigher-order signals.\n

Cited by

• Uncertainty Quantification for Evaluating Machine Translation Bias

Related

• Implicit Causality-biases in humans and LLMs as a tool for benchmarking LLM discourse capabilities
• Implicit causality in GPT-2: a case study
• Individual differences modulate sensitivity to implicit causality bias in both native and nonnative processing
• Do Large Language Models Show Biases in Causal Learning?
• Do Large Language Models Show Biases in Causal Learning? Insights from Contingency Judgment