Consider this sentence, in Portuguese:
The red syllables are always stressed in normal speech.
Speakers of Brazilian Portuguese (BP) and European Portuguese (EP) usually assign secondary stress to other syllables, but do it differently, as noted in brown.
The alternation of stressed and unstressed syllables reflects the rythmic patterns of a language. It is quite perceptible that EP and BP have distinct rhythms. We try to grasp this distinction via a predictive theory of secondary stress placement, and applying it to both varieties of Portuguese.
This is being pursued by a model based on Optimality Theory. Divide a sentence into segments of consecutive syllables, and place a stress in one syllable in each segment. There are many ways this can be done for a given sentence, and several have the correct secondary stresses, say, for BP. The model finds these stresses by solving an optimization problem over the segmentations. The crux of the model is that the function being optimized embodies several linguistic theoretical hypotheses. They are expressed by means of several constraints and an ordering of those. Different orderings yield different functions to be optimized, and should be enough to model stresses in BP and EP.
To test that model, we developed a computer program called sotaq. It allows an easy implementation of constraints of linguistic significance and also permits experimenting with different orderings of these. So, lots of data can be checked against the program, so that relevant contraints can emerge from experimentation and the program and model can be refined.
Previous: The Algorithm Vocale
Next: Working Sessions at ZiF