On 25 January 2023, 6pm CET, the next talk of the series will take place: Anna Stein (Heinrich Heine University Düsseldorf) with “Probabilistic reduction and the mental lexicon”. Please join us for this talk if you are interested in the topic or our talk series in general and can find the time! We wish to make our speakers feel welcome by as many listeners as possible.
Many papers have repeatedly proven that statistical properties or probabilistic measures of language affect the way we speak. Early mentions of frequency effects in language can be traced back to Arabic grammarians from the 8th century. This is a phenomenon called probabilistic reduction. It refers to redundancy-based effects that cause acoustic reduction. For example, high frequency words tend to be acoustically reduced compared to words with a low frequency, because low frequency words are less familiar and their acoustic information therefore less redundant. This behaviour has also been observed at the syllable and segment level. Aylett & Turk 2004 and Ennever et al. 2017 respectively showed that the probabilistic measures of syllables and segments influence their duration, and consequently also the duration of the word they are a part of. What is unclear, however, is what happens when different levels make conflicting predictions for the duration of a word. When, for example, high frequency of a word predicts a reduced duration, but its syllables have a low frequency. This results in a competition between the levels, where it is unclear whether a certain level has more predictive power than another for determining word duration. By using a computational model to represent lexical processes, I model this competition between the levels to see whether each ‘level’ of a word contributes equally to its duration (reduction), or if there are differences in how the word, its syllables and segments contribute to it.