Zipf's law in aphasic speech: An investigation of word frequency distributions

Abstract

Word frequencies in a text follow a curious pattern. A few of them appear extremely frequently, while by far most of them appear only once or twice. This pattern is considered a law of word frequencies, and better known as Zipf’s law. Its universality renders Zipf’s law one of the very few true language universals. The existence of this law has been known for more than a century. It has extensively been studied in large corpora of written texts and in many different languages, but hardly so in more singular cases such as spoken language and speech from people with a language disorder such as aphasia. This dissertation aims to fill this gap, to gain more insight into the workings of word frequencies on the one hand, and in the organization of the mental lexicon in people with and without aphasia on the other. First, the process of speech production is discussed, both in people with and without aphasia. Next, a thorough discussion of the hypotheses for the existence of Zipf’s law is presented. These explanations fall into a number of categories: The Principle of Least Effort, intermittent silence, preferential attachment, new optimization models (present day versions of the Principle of Least Effort) and a variety of semantic explanations. It is argued that models based on preferential attachment are most plausible. Next, it is shown how the values of the parameters of Zipf’s law vary depending on medium (written or spoken) and text length. In addition, a text is created for which Zipf’s law does not hold, to test if this is possible and how readers respond to such a text. These insights are then used to study Zipf’s law in different types of aphasic speech: in long samples from Dutch non-fluent aphasic speakers at 2, 5 and 8 months post onset and in short samples from English, Greek and Hungarian fluent and non-fluent aphasic speakers in the chronic stage of aphasia. It is shown that aphasia influences the values of the parameters, as does the language under consideration. But in all cases, Zipf’s law continues to apply. This finding strengthens the hypothesis that the system for word retrieval in aphasia is still intact.