HOME PROJECTS


Nadja Althaus




CURRENT AND RECENT PROJECTS

TSNE visualisation of Swedish accent data

Speech corpus analysis with deep networks (LSTMs)

A current projects investigates the use of LSTM deep neural networks to classify prosodic patterns, and in the first instance this is applied to identifying pitch accent patterns in a Swedish speech corpus. This sheds light on the production of each type of pitch accent in every-day language.

Time course model plot

Orthography and the sound of words in our minds

To what extent does the spelling of a word affect our mental representation of its sounds? Here we used a regular case of vowel raising in Bengali, which sometimes is and sometimes isn't reflected in writing, to investigate how spelling and sound changes each affect the speed of word processing. We used growth curves to examine participants' eye movements in a fragment completion task.

Time course plot with stimuli

The nature of babies' first word representations

Do children store every detail of the sound of a new word, or just as much detail as is necessary to discriminate it from other words they know? Our eye tracking project looks at how 18- and 24-month-olds process words with coronal sounds like [n], [d], [t], as evidence from adults suggests we have to be particularly flexible with this category of consonants.