arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2109.07020cs.CL
收藏

Transformers句法规则学习的频率效应

Frequency Effects on Syntactic Rule Learning in Transformers

Google Research(谷歌研究院);New York University(纽约大学);Brown University(布朗大学)

Jason Wei, Dan Garrette, Tal Linzen, Ellie Pavlick

摘要:

预先训练的语言模型在各种需要符号推理的语言任务上表现良好,这就提出了这样的模型是否隐含地表示抽象符号和规则的问题。本文通过对BERT在英语主谓一致方面的表现的个案研究来探讨这一问题。与之前的工作不同,我们从零开始训练多个BERT实例,允许我们在训练前执行一系列受控干预。我们发现,BERT通常可以很好地概括训练中从未出现过的主谓对,这表明一定程度的规则支配的行为。然而,我们也发现,表现受到词频的严重影响,实验表明,无论是动词形式的绝对频率,以及频率相对于交替的变化,有因果关系牵连的预测BERT在推理时间。对这些频率效应的进一步分析表明,BERT的行为与一般正确应用SVA规则的系统是一致的,但该系统难以克服强训练先验,并难以估计不常见词汇项的一致性特征(单数与复数)。

英文摘要:

Pre-trained language models perform well on a variety of linguistic tasks that require symbolic reasoning, raising the question of whether such models implicitly represent abstract symbols and rules. We investigate this question using the case study of BERT's performance on English subject-verb agreement. Unlike prior work, we train multiple instances of BERT from scratch, allowing us to perform a series of controlled interventions at pre-training time. We show that BERT often generalizes well to subject-verb pairs that never occurred in training, suggesting a degree of rule-governed behavior. We also find, however, that performance is heavily influenced by word frequency, with experiments showing that both the absolute frequency of a verb form, as well as the frequency relative to the alternate inflection, are causally implicated in the predictions BERT makes at inference time. Closer analysis of these frequency effects reveals that BERT's behavior is consistent with a system that correctly applies the SVA rule in general but struggles to overcome strong training priors and to estimate agreement features (singular vs. plural) on infrequent lexical items.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑