I am fairly new to support vector machines and I was thinking about whether or not I could use them to tackle a specific research question in linguistics:

My goal would be to find the probabilities for the position of the main stress in a given set of words. I would use a training set which has stress and frequencs information of several thousand words. This would be the learning set for the SVM. In a next step I would like to check what probabilities for the stress position the SVM would return for a given set of test words. Class probabilities are preferred since there are more than two options for the location of the main stress in a given word, i.e. the binary case would not work and a multi-class SVM is needed. However, SVMs usually only use numerical values as input, my training set would consist of strings (the words in the dictionary).

Does anybody have an idea if there is a way to use a SVM for my research question? I tried to use the string kernel from the ksvm() function in R but this kernel seems to be no longer supported. If anyone has ever worked with SVMs and strings as input data, I would be very happy to hear about your experiences and recommendations. Moreover, I am also interested in other machine learning approaches that could be used in order to make stress prediction for a set of test words based on a training set.

  • There are really two options here : either you have identified classes of words according to their stress e.g. 'final-stressed', 'penultimate-stress'… and you want to classify your words in them, or you want to classify the individual phonemes as either stressed or not stressed. both cases sounds equally impractical to me (then again, no phonetician here), why wouldn't you try CRF or NN instead ?
    – Evpok
    Oct 8 '16 at 16:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.