I saw a really cool talk the other day by doctoral student Daniel Avrahami from CMU. Daniel is looking at correlations between properties of IM conversations (e.g. length of message, duration of conversation, time between messages sent, etc.) and the type of relationship shared by the two interlocutors (e.g. work, social, etc.). He started with knowledge of the relationship type (as categorized by the participants in his study, i.e. the conversationalists themselves) and looked at the conversation properties described above. Then he tried to build a model that could start with the conversation properties and reverse engineer the type of relationship shared by the conversationalists. Perhaps unsurprisingly to anyone who IMs, he has found that closer friends tend to exchange individual messages that are shorter, more widely spaced, and spead over conversations whose duration is quite long, while co-workers' individual messages tend to be longer, in conversations that take much less time.

I found his talk very interesting for a few reasons beyond just the fact that the area of human-computer interaction tickles the cognitive scientist in me. First, the narcissistic: I IM, all the time, for everything. If friends at a distance aren't instant messaging, then I'm probably not in very close contact with them. I IM people at work who are just across the hall for all kinds of reasons: quickly resolving work issues, making lunch plans, telling people to close their doors so I can't hear their speaker phones, and so on. It's interesting to see whether my patterns of usage (and the patterns of usage I observe in my IM contacts) match his larger findings.

I was also interested to see that he wasn't using very many metalinguistic properties to enrich his model for accurately predicting social relationship type. It seems like he could collect data regarding canonical vs. non-canonical syntax, punctuation, and even more purely linguistic indicators such as vocabulary choice, code switching among bilinguals, and so on. He's even gotten some participants to agree to let him use the content of their discussions unmasked, mitigating some of the privacy concerns. I'm sure if someone started digging into that stuff, they would find a whole sociolinguistic world to explore, in a new medium that is relatvely unstudied for this kind of interaction.

In any case, I'm looking forward to checking out the complete paper that the talk was based on. I came away with all kinds of ideas for things to look at, which is pretty much my definition of a good talk.