Recent studies suggest that self-reflective prompting can significantly enhance the reasoning capabilities of Large Language Models (LLMs). However, the use of external feedback as a stop criterion raises doubts about the true extent of LLMs’ ability to emulate human-like self-reflection. In this paper, we set out to clarify these capabilities under a more stringent evaluation setting in which we disallow any kind of external feedback. Our findings under this setting show a split: while self-reflection enhances performance in TruthfulQA, it adversely affects results in HotpotQA.We conduct follow-up analyses to clarify the contributing factors in these patterns, and find that the influence of self-reflection is impacted both by reliability of accuracy in models’ initial responses, and by overall question difficulty: specifically, self-reflection shows the most benefit when models are less likely to be correct initially, and when overall question difficulty is higher. We also find that self-reflection reduces tendency toward majority voting. Based on our findings, we propose guidelines for decisions on when to implement self-reflection. We release the codebase for reproducing our experiments at https://github.com/yanhong-lbh/LLM-SelfReflection-Eval.

In this paper, we describe TTIC’s submission to WMT 2023 Sign Language Translation task on the Swiss-German Sign Language (DSGS) to German track. Our approach explores the advantages of using large-scale self-supervised pre-training in the task of sign language translation, over more traditional approaches that rely heavily on supervision, along with costly labels such as gloss annotations. The proposed model consists of a VideoSwin transformer for image encoding, and a T5 model adapted to receive VideoSwin features as input instead of text. In WMT-SLT 22’s development set, this system achieves 2.03 BLEU score, a 59% increase over the previous best reported performance. In the official test set, our primary submission achieves 1.1 BLEU score and 17.0 chrF score.