Carter Georgia-Ann, Keller Frank, Hoffman Paul
Institute for Language, Cognition and Computation, School of Informatics, The University of Edinburgh.
School of Philosophy, Psychology and Language Sciences, The University of Edinburgh.
Cogn Sci. 2025 Jun;49(6):e70072. doi: 10.1111/cogs.70072.
Word embeddings derived from large language corpora have been successfully used in cognitive science and artificial intelligence to represent linguistic meaning. However, there is continued debate as to how well they encode useful information about the perceptual qualities of concepts. This debate is critical to identifying the scope of embodiment in human semantics. If perceptual object properties can be inferred from word embeddings derived from language alone, this suggests that language provides a useful adjunct to direct perceptual experience for acquiring this kind of conceptual knowledge. Previous research has shown mixed performance when embeddings are used to predict perceptual qualities. Here, we tested if we could improve performance by leveraging the ability of Transformer-based language models to represent word meaning in context. To this end, we conducted two experiments. Our first experiment investigated noun representations. We generated decontextualized ("charcoal") and contextualized ("the brightness of charcoal") Word2Vec and BERT embeddings for a large set of concepts and compared their ability to predict human ratings of the concepts' brightness. We repeated this procedure to also probe for the shape of those concepts. In general, we found very good prediction performance for shape, and a more modest performance for brightness. The addition of context did not improve perceptual prediction performance. In Experiment 2, we investigated representations of adjective-noun phrases. Perceptual prediction performance was generally found to be good, with the nonadditive nature of adjective brightness reflected in the word embeddings. We also found that the addition of context had a limited impact on how well perceptual features could be predicted. We frame these results against current work on the interpretability of language models and debates surrounding embodiment in human conceptual processing.
从大型语言语料库中衍生出的词嵌入已成功应用于认知科学和人工智能领域,用于表示语言意义。然而,关于它们对概念的感知特性编码有用信息的程度,仍存在持续的争论。这场争论对于确定人类语义中具身性的范围至关重要。如果仅从语言衍生的词嵌入中就能推断出感知对象的属性,这表明语言为获取这类概念知识提供了一种有用的辅助手段,可辅助直接的感知体验。先前的研究表明,在使用嵌入来预测感知特性时,表现参差不齐。在此,我们测试了能否通过利用基于Transformer的语言模型在上下文中表示词义的能力来提高性能。为此,我们进行了两项实验。我们的第一个实验研究了名词表征。我们为大量概念生成了去语境化(“木炭”)和语境化(“木炭的亮度”)的Word2Vec和BERT嵌入,并比较了它们预测人类对这些概念亮度评分的能力。我们重复这个过程,也探究了这些概念的形状。总体而言,我们发现形状的预测性能非常好,而亮度的预测性能则较为一般。添加上下文并没有提高感知预测性能。在实验2中,我们研究了形容词 - 名词短语的表征。感知预测性能总体上被发现良好,形容词亮度的非相加性质在词嵌入中得到了体现。我们还发现,添加上下文对感知特征的预测能力影响有限。我们将这些结果与当前关于语言模型可解释性的研究以及围绕人类概念处理中具身性的争论联系起来。