While the fact that phonetic information is evaluated in a non-discrete, probabilistic fashion is well established, there is less consensus regarding how long such encoding is maintained. Here, we examined whether people maintain in memory the amount of vowel nasality present in a word when processing a subsequent word that holds a semantic dependency with the first one. Vowel nasality in English is an acoustic correlate of the oral vs. nasal status of an adjacent consonant, and sometimes it is the only distinguishing phonetic feature (e.g., bet vs. bent). In Experiment 1, we show that people can perceive differences in nasality between two vowels above and beyond differences in the categorization of those vowels. In Experiment 2, we tracked listeners’ eye-movements as they heard a sentence that mentioned one of four displayed images (e.g., ‘money’) following a prime word (e.g., ‘bet’) that held a semantic relationship with the target word. Recognition of the target was found to be modulated by the degree of nasality in the first word’s vowel: Slightly greater uncertainty regarding the oral status of the post-vocalic consonant in the first word translated into a weaker semantic cue for the identification of the second word. Thus, listeners appear to maintain in memory the degree of vowel nasality they perceived on the first word and bring this information to bear onto the interpretation of a subsequent, semantically-dependent word. Probabilistic cue integration across words that hold semantic coherence, we argue, contributes to achieving robust language comprehension despite the inherent ambiguity of the speech signal.

虽然以非离散，概率的方式评估语音信息的事实已广为人知，但关于将这种编码保持多长时间的共识却很少。在这里，我们检查了人们在处理后一个与第一个单词具有语义依赖性的单词时，是否在内存中保留了单词中存在的元音鼻音数量。英语中的元音鼻音是相邻辅音的口腔状态与鼻音状态之间的声学​​关联，有时它是唯一的区别性语音特征（例如，下注与弯曲））。在实验1中，我们表明人们可以感知两个元音之间的鼻腔差异，而不仅仅是这些元音的分类差异。在实验2中，我们跟踪了听众的眼动，因为他们听到一个句子，该句子提到了与目标词保持语义关系的素词（例如“ bet”）后面的四个显示图像（例如“ money”）之一。发现目标的识别受到第一个单词元音中鼻音程度的调节：第一个单词中语音后辅音的口头状态的不确定性稍大，这转化为较弱的用于识别第二个单词的语义提示。从而，听众似乎在记忆中保持了他们在第一个单词上感知到的元音鼻音的程度，并将此信息带入对后续语义相关单词的解释中。我们认为，尽管语音信号具有内在的歧义性，但具有语义连贯性的单词之间的概率提示整合仍有助于实现强大的语言理解能力。