Keyword Trainer

Sanitising History?

2010-03-01T17:11:00.005Z

I thought I'd use this brief break from my fully booked timetable to post an interesting story my friend just told me - I think it illustrates really well the effect keywords can have on the perception of images:

While I was studying for an MA History & Theory of Photography our class was given a tour of a picture library’s historical archive. In one particular room several people were scanning and others were captioning, old images. We were free to talk to the employees. I looked over the shoulder of one of the keyworders and asked her what she was doing. She said that she was PC'ing the keywords. I asked her what that meant and she said that some of these old images have captions and words that we can't use, such as the N word, so I'm re-keywording them. I told her that you are re-writing the past and she just looked straight through me, with the look of someone who has been keywording continuously for hours on end. Later I brought this up with the manager who had given the tour. I said that you are sanitising history, you are changing it to make it more palatable and he just shrugged his shoulders and said that unless they ‘corrected’ these images they wouldn’t sell.

Historical accuracy and commercial sales aren't easy bedfellows but I cannot but wonder that if by making captions and keywords more PC we are promoting a kind of deliberate social amnesia? Shouldn't we learn from the past rather than cover it up? Photographs have been retouched and manipulated from its inception but I think that my friend's story is a good example of how metadata can also have a subtle yet crucial part to play in the way in which we understand images. Good and bad comments are welcomed.

Excerpt of my recent comment on keywording standards

2009-12-15T06:36:00.005Z

I recently posted some thoughts on controlled vocabularies and whether there is such a thing as ‘industry standard’ at the Controlled Vocabulary forum. Here’s a modified excerpt of that text (please excuse if the paragraphs seem disjointed):

As someone who prepares data (including keywording and captioning) for Getty, Corbis and Alamy as well as small specialist libraries on a daily basis, I recognise that there isn't a single industry keywording standard. I can only wish for one as this would make my job a bit easier. Yet, Getty, Corbis and some other generic stock libraries are in many ways similar and, on the whole, the apparent differences are, as has been said in the Keywording Central blog, “stylistic rather than substantive”. Both agencies typically use terms such as Teenager, Mid Adult and Young Adult for age; Teenage Girl, Mid Adult Man and Young Woman for gender; and Caucasian and Asian ethnicity and for ethnicity. This would be a good starting point for standardisation.

As I said in my previous post, I believe it is possible to define good keywording and I would go even further by adding that it is necessary to define good keywording if image searching in the stock industry is to be meaningful (note: I regard social tagging a different kettle of fish which is why I limit the discussion to stock). Keywords are crucial for accessing images (at the moment content based retrieval cannot achieve the same level of accuracy and consistency as text based retrieval) and overarching rules on how to make visual content accessible by text apply. To summarise my previous post, good keywording is a) consistent (terms are used consistently and in the same way across the collection), b) relevant (meet the needs and search methods of the searchers) and finally c) diverse (cater as many needs as possible). How exactly these principles are applied varies from agency to agency but the overall criterion is the same.

Some argue that regardless of its advantages a controlled vocabulary runs into problems because language is a moving target, with new words entering the English vocabulary and old terms becoming obsolete. To an extent this is true but I don’t think this is where the limitations of a controlled vocabulary lie. In practice, keeping up with changing language isn’t a problem - I tweak my vocabulary maybe once a month and it’s never more than a couple of hours’ work. I don’t think the sheer size of the English language is a huge problem either as most of us only use a fraction of the entire English vocabulary (even Shakespeare who is regarded as having an unusually large vocabulary covered only a small part of the entire language) and our everyday language, which is one criteria for choosing search terms, is fairly limited. Further, the key aim of a controlled vocabulary is precisely to deal with this vastness of language - a well-constructed keyword tree will bring consistency to the way in which words are used whilst allowing people to find images regardless of the term they are searching for (and contrary to some claims, choosing a preferred term over another isn’t arbitrary as this depends on the target audience and how terms are used and understood in a given field). For me, the real limitations of a controlled vocabulary lie in the differences between text and visuals, which is another topic all together.

There has been a trend to imitate the major industry players’ vocabulary with the result of peculiar terms such as Human Hand and Full Suit being used. You may think that this is counter-intuitive and unhelpful as it is not how people normally talk and therefore it is not how people would search for images. However, the use of such terms arise from the need to display words as unambiguously as possible. This doesn’t mean that the searching experience is jeopardized. People can still use commonly used terms such as Hand and Suit and find the images they are after. Even in a search engine that doesn’t employ a thesaurus I think the use of neologisms together with ‘common’ words can be justified as it adds more search value, which is surely desirable.

Going back to my earlier point and to echo Sarah Saunders, there is no one solution that fits all but there are common sense principles that can dramatically improve image findability.

another one...

2009-12-08T21:10:00.004Z

How about this - indoors or outdoors?

How would you keyword this?

2009-12-03T09:46:00.008Z

Welcome to my Mind Bogglers series which deals with typical problems with keywording and controlled vocabularies. There is no correct answer to these as I simply want to encourage discussion. So, here's the first one:

Would you keyword the general setting for this image as 'indoors' or 'outdoors' ?

What is good keywording?

2009-09-01T09:44:00.002+01:00

What is good keywording and how do you achieve it? A summary of my recent talk at the Imperial War Museum in London.
My sincere apologies for taking so long to post this article. Life’s been very busy with new and exciting projects in all the main areas - controlled vocabularies, keywording and training. Admittedly, a big chunk of my spare time has also been taken up by my passion for triathlons - it’s my first season and I have just completed the first couple of races.

One of the many things I have been occupied with was a talk I gave at an ACE (Association for Cultural Enterprises) seminar for the Cultural Heritage sector at the Imperial War Museum in London. It was a very interesting day and the other excellent talkers gave a lot of food for thought. Here’s a summary of mine:

Images are used for different purposes by different people and the original context, although important, is only one way of approaching visual content. To make images accessible, keywords are imperative. When I was preparing for this presentation there was a particular painting I wanted to use but, alas, I didn’t remember who painted it or when it was painted. The only thing I could remember was that it depicted young boys at a beach playing with home-made toy boats, sunlight flickering on the surface of the water with some large stones in the foreground. I searched for ‘boys’, ‘Finnish painting’ ‘sea’, ‘traditional’, ‘playing’ and so on but I simply could not find this image. After an hour, I stumbled across it by pure chance. Had I been a potential customer wanting to use the image, say, for a book cover this would have been a lost sale. The point I want to make here is that images, particularly those in a specialist area such as fine art, archeology, architecture and music should be approached from a non-academic point of view if a wider audience is to be reached. Recording only the title, artist’s name and the year the object was created may serve well the experts but simply isn’t enough for someone like me who can only remember a few random aspects of the object. Therefore diversity is a critical aspect of good keywording.

The second component of good keywording is consistency. This often means having a controlled vocabulary which comes in many shapes and forms. A controlled vocabulary can dramatically improve search results as it stabilises and disambiguates terms. Only one term is chosen for keywording when many variations are available (candy or sweets, car or automobile?) and the meaning of ambiguous terms are clearly defined (Paris Hilton or Paris the capital, pool as in a billiards game or a swimming pool?). When adding keywords to images you will also need to ensure that the terms are always applied in the same way. For example, should ‘Transportation’ be applied to all images that have to do with transport related things or should it be more narrowly defined only to cover images which show things or people actually being transported from place to place. These decisions are part and parcel of image findability and it is also what makes the keywording challenging and fascinating!

The third and last characteristic of good keywording is relevancy. Understand your audience and how they search images. What do they expect to see when searching for a particular term? One of the most common mistakes people new to keywording make is that they apply too many keywords to images in the hope of increasing sales. Nothing frustrates a picture searcher more than irrelevant search results and having to scan through hundreds of pages to find that one relevant image. Be accurate yet brave enough to leave small, peripheral things out (unless, of course, this is what your users want).

That’s it for now. I hope you have enjoyed this article and that you found it worth the wait! Please keep checking my website every so often, I’ve got some new features in the pipeline and I hope to release them later this year.

Reflections on keywording after my first course

2009-02-01T09:45:00.000Z

A big thank you to the participants of my – and I’m told, the industry’s – first ever keywording course which was recently held in London. The full class (thanks a million for persevering in that tiny hot room!) reaffirmed the rising popularity of keywording. Contrary to the old saying, images without words are worthless in today’s digital environment.

Yet there is some truth in the old phrase. Images are complex and interpreting them logically is a difficult job (ask anyone studying semiotics). It never ceases to amaze me how people new to image cataloguing think that anyone can add keywords to images – it is just a matter of adding words that correspond with what is shown in the image. But what is seen in the image varies from viewer to viewer and when we add to this the flexibility of natural language (pool can refer to a body of water, billiards game, swimming pool, etc., and Wellington boots can be called Wellies, Rubber boots, Galoshes, Gum boots, etc.) we end up with a mixture of ingredients which requires a skilful chef to dish up a complete meal.

 I certainly had my preconceived ideas about keywording when I started as a keyworder at Photonica some years ago. Surely I was well-equipped for image cataloguing having studied photography, art history and aesthetics? My patient manager thought the contrary – I was considered good enough for the job only after three months of training and even after this we had ongoing training one day a month.
 
Having just returned from cold Finland to equally cold England I’m working on adding new features to my site. In this blog you will find my thoughts on image classification and there will be a separate page for tips on keywording and controlled vocabularies. I’m also planning to add a fun page where you can participate in keywording – by doing so I’m hoping to make this solitary business a little less isolated.