Beschrijving
Free download at https://doi.org/10.54195/9789465152059
When we listen to speech, we usually understand words quickly and effortlessly. This is especially remarkable because languages consist of thousands of thousands of words, some of which can only be differentiated based on a single sound. In many languages, some words share their stems. For instance, feather, feathered, and featherbed, all contain the stem feather. These words can be seen as consisting of several meaningful parts (e.g. feather + bed). This thesis investigates whether and how the different parts of these words are used in the recognition processes especially in listening.
To investigate the role of the different parts of a word, the thesis analyses the results from large-scale listening experiments with computational models, simulating how word recognition might work in the human brain. The research shows that listeners rely both on the whole words (e.g. feathered) and on its parts (e.g. feather and -ed). Whether the whole word or its parts are most important depends on the type of word and whether it is heard or read. The thesis also shows that spoken words are recognised more quickly when they belong to a larger group of words sharing their stem (such as feather, feathered, and featherbed). This family size effect is strongest when related words are similar in both meaning and sound. By comparing different types of computational models, the research demonstrates that family size effects can arise naturally from how listeners narrow down possible word candidates as spoken words unfold, even without explicitly breaking words into parts.
In conclusion, this work contributes to building more accurate theories of human spoken word recognition and clarifies the role of a words’ internal structure in the process of recognising spoken words.
