Response to Mrunal Chavda's "Sentiment Analysis"

Response to Mrunal Chavda's "Sentiment Analysis"

Posted by denise on December 3, 2014

I met Mrunal in September 2014, during the GW4 workshop on Collaboration in Bristol. We were assigned to the same team when it came to discussing possible collaborations between researchers of the GW4. We engaged in a really inspiring conversation about theatre, computing and potential overlaps that could be used to spark new ideas and discussions and might eventually lead to new research projects.

One outcome of this is the following text. I wrote it in response to Mrunal's most recent blog post which you can find here. In his post, Mrunal expresses his ideas which formed the basis of our conversation in Bristol. This is the continuation of our discussion. And hopefully the beginning of an interesting research journey.

"Sentiment Analysis: An Interdisciplinary Approach"

Now, the title already makes me think hard. In my opinion, we'd have to specify what we mean by sentiment. It might be, that we do or don't understand the same thing summed up by this term. My approach would be influenced by psychology research which differentiates between emotions. mood, sentiments and personality. Duration and focus would be two of the dimensions used to distinguish one from the other. Emotions for example are shorter in duration than mood or personality. And they are also said to be more narrow in focus, e.g. related to a specific event. There is also an approach, suggested by Dave Moffat in 1997, that emotions can be viewed as short term personality and personality as a long term emotion. But before I follow this tangent any further, it would be better to say what theatre researchers understand by sentiment analysis. This is important to clarify because we are lead by the following general question:

"In this collaborative work between a computer science researcher (Denise Lengyel)and theatre researcher (Mrunal Chavda), the attempt is to answer the following question:

Can we trace actors’ physical manifestation and provide a link their internal states using a computer generated model?"

This questions seems to raise several research problems in one. First of all, what are physical manifestations of actors? Are we talking about verbal behaviour? Non-verbal behaviour (and if so, which of its many subcategories)?

Secondly, how can we then trace the manifestations?--which would mean tracking them and then analysing certain aspects of interest (parts of the manifestations) over time to create traces and maybe search for patterns within those traces. But what are those aspects? Are they covered comprehensiveley by Bharata's determinants (see quote from Mrunal below)? Or would they need further refinement?

"Bharata listed the determinants, such as unseemly dress, near obscene utterances and pointing out the faults of others, displaying deformed limbs, quarrels, covetousness, misplaced ornaments, and impudence. Bharata listed among its consequences biting lips, throbbing of the nose and cheek, widening of the eyes, contracting eyes, perspiration, and colouring of the face and holding one’s sides. The transitory states are lethargy, dissimulation, drowsiness, sleeplessness, waking up and envy."

Internal states seems to relate to the notion of sentiments. One challenge would be to measure them. We could rely on self-reported data. Or maybe we should consider psychophysiological methods, i.e. EMG or ECG? I personally would be interested in empathy as an internal state of the actor. Which would mean having to rely on self-reported data for lack of better instruments.

The question how the link or mapping between physical manifestations and internal states could be investigated it related to that. And at the same time is a research quest of its own. I'm sure there is research in social science and psychology out there that investigates that. We'd have to dig deeper there. Get talking to researchers from appropriate fields.

Last but, it seems, definitely not least, how should and could that be modeled on a computer? And can one model be responsible for tracing physical manifestations AND linking of internal states?


--to be continued--