Hello,

I suggest you to see this new study:

https://www.sciencedirect.com/science/article/abs/pii/S1050641120301279

Good luck.

Dear Kumar,

as first I suggest you to reduce the dimensionality of your data set by principal component analysis. Given you have two subsets of variables (subset A = verbal utterances, 30 binary 1/0 variables for each utterance + 8 yes/no binary codes I suppose) and subset B = 17 facial muscle movements) you must perform two separate PCAs for the two subsets. This implies you get two data matrices (for A and B respectively) both having as statistical units (rows) the subsequent observations. Given the facial movements are each other correlated, subset B will give rise to a number of relevant components much lower than 17, components are each other linearly indpendent by construction and the observation of the loadings (correlation coefficients between original variables and components, will allow you to interpret the meaning od every component). The same will happen (even if to a lesser extent) for subset A. Now you have in your hands nA principal componetnscores for subset A (PCscores(A)) and nB (PCscores(B)) components for subset B. Given the two subsets have the same statistical units you can correlate the two subsets of components by canonical analysis and in this way single out the relational structure linking utterances and facial muscle movements. The component scores are time dpendent signals (the statistical units are taken one after the other) you can in turn amalyze (if you want you can do the same analysis for each single variable) by Recurrence Quantification Analysis (RQA) that is particularly suited for this kind of data, given it does not imply any stationarity assumption and works very well for short time series. Still more important it is particularly suited for detecting transitions and/or short temporal patterns in the series.

Here below some literature:

Article Climate induced temperature effects on growth performance, f...

https://www.nature.com/articles/s41397-019-0080-6

Article The Application of Principal Component Analysis to Drug Disc...

https://www.sciencedirect.com/science/article/abs/pii/S0375960196007414

Article Combinatorics and synchronization in natural semiotics

May be it is important to know what is end objective with the data. I see a time frequency localization study with some self similarity behavior study could bring some hidden structures. Once coudl also do energy dissipation phenomena there.