Wednesday, 7 July 2010

Context and meaning in search

My 18 month old son has just discovered sentences - or, more specifically one sentence - "Where's it gone?" (pronounced as one word - "Wezzigonn?" ). He throws his ball into the bushes. He looks at me dolefully and says "Wezzigonn?". I retrieve the ball. He throws it into the bushes. He looks at me dolefully and says ... well, you get the picture. Essentially, he thinks "Where's it gone?" is a single word which means "Fetch the ball, Dad". And the interesting thing is that in the context of the "game" I understand exactly what he means.

In Wittgenstein's Philosophical Investigations he posits a "language game" involving a builder and his assistant. Every time the builder needs another slab he shouts "Slab!" and his assistant duly brings him a slab. So the single word "Slab!" functions as the sentence "Bring me a slab!" in this context.

(He also discusses a variation where every time the builder needs a slab he calls out "Bring me a slab!". An observer who doesn't speak the same language assumes that "Bring me a slab!" is the word for "slab" and when building a wall himself calls to his assistant "Pass me a 'bring-me-a-slab'!" These kind of misunderstandings are commonly found in placenames - such as Bredon Hill, meaning "Hill Hill Hill".)

I suppose the point of all this is that context gives language meaning. Which is all very well if you're building a house, buying some cabbages, throwing a party etc. But what about when you "speak" into a search box. Where do you get your context from then? Do you have to play around with clever search modifiers so the interface understands that when you search Google Images for "bondage" you're looking for pictures of serfs?

Not entirely - at least not in Google and not in Amazon, which both use contextual information to give you relevant results. [In]famously, these interfaces give very different results depending on whether you are logged in. People get pretty uncomfortable with the whole idea of this 'contextual information' - how it's gathered, where it's stored. But the use of contextual information to give language meaning is an essential part of communication.

Libraries have thousands of users and millions of resources crying out to be introduced to each other. But our search mechanisms tend to be context-less. Lorcan Dempsey has said Discovery happens Elsewhere. For users of university libraries, discovery happens in context-heavy environments such as reading lists, citations, seminars, lectures. By the time they get to our interfaces they know exactly what they want. Then they find it (or don't).

Can we start to build some context into our own systems? And what kind of context would be useful? We can say straight off that for students the most useful context is what course they're doing (something we will soon have access to, and which I've blogged about elsewhere). If we also have access to course materials (i.e. reading lists) we can really start to provide useful context for searches. How about if we have access to the content of books and articles - in particular the citations they contain? Could we start to put our searches in the context of a scholarly network based on citation?

Or do we run the risk of second-guessing what users are searching for, and getting it wrong? There are endless anecdotes about people changing their relationship status on Facebook and immediately being bombarded with ads for wedding planners/speed dating. If people change course do we start serving up different results? And if discovery happens in context, how far should the library go in providing context, and how much should it leave to others?

PS Emma has pointed out that little Henry is playing the Fort-Da game http://www.cas.buffalo.edu/classes/eng/willbern/BestSellers/Catcher/FortDa.htm and that along with Lorcan and Wittgenstein, Freud could also be added to the list of tags!

8 comments:

Tony Hirst said...

For some time, I've been wondering whether or note it would be possible to use a contextual or behavioural adserver to server not ads, but links to relevant teaching content or learning resources...

Fancy having a brainstrom around this one day, and maybe trying to hack a few demos/proofs of concept. Eg I've wondered whether OpenX might be appropriate?

Huw said...

Yes, indeed! Are you in Cam anytime soon?

Lorcan Dempsey said...

Probably the only time I'll appear in a list with Wittgenstein ;-)

Huw said...

Not only that - you also come before him in the list!

thingblogging said...

I think this is such an interesting area. Especially as it's true, libraries (potentially) already hold so much useful information to allow them to provide context for searches.

I do worry about the issue of what you call "second-guessing" though, not so much getting it wrong but more limiting what people find. Sometimes context can be too restrictive or omit things that actually might be of use/interest. So there also needs to be the option to turn context off in some way. Perhaps less an issue for undergraduates, but definitely once we're talking about research. Otherwise there's a risk of creating closed off categories that only point within themselves and do not create links or connections to things outside. Often the most interesting ideas come from combining things that live in different defined "subject" areas (faculties, classification numbers, undergraduate courses, however you define it).

Celine

Huw said...

I think you'd either have one interface that did both - i.e. provided unfiltered results accompanied by "suggestions" based on you and what you're searching for a la Amazon
OR
Have totally separate interfaces for context based searching

The problems probably arise when people don't know that context is being taken into account ...

Though of course the underlying mechanisms of search already make a range of assumptions which put us firmly in the context of a "library searcher"!

librariangoddess said...

Sounds like your son is playing the Fort/Da game - so you can add Freud into the list along with Lorcan and Wittgenstein :) (http://www.cas.buffalo.edu/classes/eng/willbern/BestSellers/Catcher/FortDa.htm)

This is a great post, and ties in with stuff I think about every time I go through a search with someone on a big, cross-disciplinary db like Scopus. Does 'cycle' relate to an interative process or a form of transport? Or the other way around - and more subtly - does 'recidivism' give you the same context as 'reoffending'?

Most recently, I saw a sign saying 'COSMOS' and immediately wanted a drink ...

Huw said...

Thanks Emma! The more I think about it the more I wonder whether "search" is the right mechanism for context-driven delivery - in fact maybe "delivery" might be a better place to start than "discovery" ...