Form and function, computing and people - acdh-oeaw/wugsy GitHub Wiki

Opinions here are all Daniel's!

All our work concerns interactions between computers and people. We therefore need to accumulate some ideas about how this interaction works, and what would improve the state of things.

First, let's recognise that computer-computer interactions are, in effect, perfect, and none of our business. But when humanity gets involved, and ask computers to do things, computers vary in competence. The one certainty is that they always improve, however.

So, where should this improvement be? I think the problem can be approached from a functional-grammatical linguistic viewpoint. Machines have no problem with lower-level language tasks (segmenting sentences, words, pos tagging, and frequency calculations). They can also homogenise things really well. Where they struggle is meaning and context. For example, while a computer can easily recognise an English interrogative by looking at subject/verb ordering, it may not know if the interrogative functions as a question, as a command, an accusation, or what. A simple interrogative like 'Is it cold?' varies so much in function based on (a) the relationship and identities of speakers, (b), the topic being discussed, and (c) surrounding language. Humans figure this out instantly, naturally, and it's fun.

All this is to say that our games need to divide labour correctly between the computer and the machine. Human time and brainpower is precious; computing time is not. So, game questions should be derivable automatically based on a database of language forms; meanwhile, the answers returned for these games need to target that which is not easily done by a computer. We therefore begin to fill in the current gap in computational understanding of meaning and context.