Comparing Plausibility Estimates in Base and Instruction-Tuned Large Language Models

Anonymous

Abstract: Instruction-tuned LLMs can respond to explicit queries formulated as prompts, which greatly facilitates interaction with human users. However, prompt-based approaches might not always be able to tap into the wealth of implicit knowledge acquired by LLMs during pre-training. This paper presents a comprehensive study of ways to evaluate semantic plausibility in LLMs. We compare base and instruction-tuned LLM performance on an English sentence plausibility task via (a) explicit prompting and (b) implicit estimation via direct readout of the probabilities models assign to strings. Experiment 1 shows that, across model architectures and plausibility datasets, (i) log likelihood (LL) scores are the most reliable indicator of sentence plausibility, with zero-shot prompting yielding inconsistent and typically poor results; (ii) LL-based performance is still inferior to human performance; (iii) instruction-tuned models have worse LL-based performance than base models. In Experiment 2, we show that LL scores across models are modulated by context in the expected way, showing high performance on three metrics of context-sensitive plausibility and providing a direct match to explicit human plausibility judgments. Overall, LL estimates remain a more reliable measure of plausibility in LLMs than direct prompting.

Paper Type: long

Research Area: Interpretability and Analysis of Models for NLP

Contribution Types: Model analysis & interpretability

Languages Studied: English