Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Surface Form Competition: Why the Highest Probability Answer Isn't Always Right

About

Large language models have shown promising results in zero-shot settings (Brown et al.,2020; Radford et al., 2019). For example, they can perform multiple choice tasks simply by conditioning on a question and selecting the answer with the highest probability. However, ranking by string probability can be problematic due to surface form competition-wherein different surface forms compete for probability mass, even if they represent the same underlying concept, e.g. "computer" and "PC." Since probability mass is finite, this lowers the probability of the correct answer, due to competition from other strings that are valid answers (but not one of the multiple choice options). We introduce Domain Conditional Pointwise Mutual Information, an alternative scoring function that directly compensates for surface form competition by simply reweighing each option according to a term that is proportional to its a priori likelihood within the context of the specific zero-shot task. It achieves consistent gains in zero-shot performance over both calibrated (Zhao et al., 2021) and uncalibrated scoring functions on all GPT-2 and GPT-3 models over a variety of multiple choice datasets.

Ari Holtzman, Peter West, Vered Shwartz, Yejin Choi, Luke Zettlemoyer• 2021

Related benchmarks

TaskDatasetResultRank
Natural Language InferenceRTE
Accuracy64.3
367
Subjectivity ClassificationSubj
Accuracy64
266
Text ClassificationAG-News
Accuracy75.4
248
Sentiment ClassificationSST-2
Accuracy71.4
174
Sentiment ClassificationMR
Accuracy76.3
148
Sentiment ClassificationCR
Accuracy70
142
Text ClassificationAGNews
Accuracy74.7
28
Topic ClassificationYahoo Answers Topics
Accuracy54.7
26
Sentiment AnalysisRotten Tomato
Accuracy75.5
25
Open-set knowledge retrievalT-REx (All)
Macro-averaged EM16.4
19
Showing 10 of 18 rows

Other info

Follow for update