The role of learning mechanisms in understanding spoken language

近年来，有几个值得注意的例子表明，学习单词及其含义的能力在我们成年后的整个生活中仍然很重要。学习机制不仅使我们能够获得全新的信息(例如，单词“Twitter”(2)的社交网络含义，见相关学生身份)，而且它们还可以修改我们现有的关于我们已经知道的单词的知识，以增强它们的后续处理。这种学习机制有一个明显的好处，那就是促进了像“树皮”这样具有多种含义的单词的处理。这种形式的歧义在语言中普遍存在，超过80%的单词有一个以上的词典定义(3)。学习机制可以通过根据经验调整词义的可获得性，对我们理解这些棘手的单词的能力做出重要贡献。例如，“黑桃”这个词既可以指“挖掘工具”，也可以指“纸牌套装”，而我们对这个词的偏好解释可能会受到我们最近是否把园艺或打牌作为一种爱好的影响。通过学习我们最近使用这些单词的经验，我们可以更好地预测未来更有可能遇到哪些含义。有几个理由相信，这种学习机制确实对理解歧义单词做出了重要贡献。首先，许多研究表明，在整个语言中出现频率最高的词义对加工有很大的好处(4)，这样，在像“蓝笔”这样的短语中，“笔”这个词更容易理解，因为它使用的是主导意义，而像“猪笔”这样的短语使用的是频率较低的意义。这些发现间接地表明，我们对一生中遇到不同词义的总体频率很敏感。此外，我们最近提供的第一个直接证据表明，听者对词义的偏好受到最近一次与词义之一的接触的强烈影响，因此，在听到像“农场主把羊放进羊圈”这样的句子后，“笔”的低频意义会变得更容易获得(5)。尽管这一证据表明学习机制在我们理解歧义单词的能力中发挥了关键作用，但目前的歧义解决模型在解释这些机制时明显不足。这项研究的主要目的是获得关于听者如何以及何时可以利用他们的经验来调整他们对个别词义的偏好的全面的实证证据。首先，我们将确定听者是否从所有与歧义单词的接触中学习，或者如果特定条件是学习所必需的(实验1，6)。其次，我们将测量听者近期和长期使用歧义单词的经验的相对贡献，以确定我们如何整合我们一生中获得的信息(实验2-5，9)。最后，我们将探索意义偏好是否会受到说话人的身份或口音等因素的影响(实验7，8)。这些数据将指导开发一种新的歧义解决模式，该模式明确了这些学习机制的性质。这一新的研究路线提供了在这一重要主题上取得快速进展的机会，并可能导致对语言理解这一关键方面的根本不同的理解。通过了解这些过程在健康成年人中是如何运作的，我们还将提供经验证据，使未来的研究能够评估那些已知理解含有歧义单词的句子有困难的人的加工和学习缺陷的贡献(见影响总结)。

In recent years there have been several notable demonstrations that the ability to learn about words and their meanings continues to be important throughout our adult life (1). Not only do learning mechanisms allow us to acquire entirely new information (e.g., the social-networking meaning of the word "twitter" (2), see associated studentship), but they can also modify our existing knowledge about words that we already know in order to enhance their subsequent processing. One case where such a learning mechanism would be of clear benefit is to facilitate the processing of words like "bark" that have multiple meanings. This form of ambiguity is ubiquitous in language, with over 80% of words having more than one dictionary definition (3). Learning mechanisms could make an important contribution to our ability to understand these tricky words by adjusting the availability of their meanings on the basis of experience. For example the word "spade" can refer to either a "digging tool" or "card suit", and our preferred interpretation of this word could be influenced by whether we had recently taken up gardening or card-playing as a hobby. By learning from our recent experience with such words, we can make better predictions about which meaning is more likely to be encountered in the future. There are several reasons to believe that such learning mechanisms do indeed make an important contribution to understanding ambiguous words. First, numerous studies have shown a strong processing benefit for the meaning that is most frequent in the language as a whole (4), such that the word "pen" is easier to understand in a phrase like "the blue pen", which uses its dominant meaning compared a with the phrase like "the pig pen" where its low frequency meaning is used. These findings suggest, indirectly, that that we are sensitive to the overall frequency with which we encounter alternative word meanings across our life-time. In addition, we have recently provided the first direct evidence that listeners' preferences for word meanings are strongly influenced by a single recent encounter with one of its meanings, such that the low-frequency meaning of "pen" would become more readily available after hearing a sentence like "the farmer put the sheep in the pen"(5).Despite this evidence that learning mechanisms play a key role in our ability to understand ambiguous words, current models of ambiguity resolution are remarkably inadequate when it comes to explaining these mechanisms. The primary aim of the proposed research is to obtain comprehensive empirical evidence about how and when listeners can use their experience to adjust their preferences for individual word meanings. First we will establish whether listeners learn from all encounters with an ambiguous word or if particular conditions necessary for learning to occur (Experiment 1, 6). Second, we will measure the relative contributions of listeners' recent and longer-term experience with ambiguous words to determine how we integrate information acquired across our lifespan (Experiments 2-5, 9). Finally we will explore whether meaning preferences can be modulated by factors such as the identity or accent of the speaker (Experiments 7, 8). These data will guide the development of a new model of ambiguity resolution which specifies the nature of these learning mechanisms. This novel line of research offers an opportunity to make rapid progress on this important topic, and may lead to a radically different understanding of this critical aspect of language comprehension.By understanding how these processes operate within healthy adults we will also provide the empirical evidence that will allow future research to assess the contributions of both processing and learning deficits in those individuals who are known to have difficulties understanding sentences that contain ambiguous words (see Impact Summary).