How does Google choose which items to display at the top of their lists? It depends on many factors (80 or so, I've heard) amongst them factors which don't depend on the contents of the page itself
- whether the item is on a popular site
- whether there are many links to the item
- whether the item appears as a link on a popular page
Ranking items is useful, but some programs (trueknowledge for example) try to do more, attempting to construct new knowledge. The process is not dissimilar to what humans do when they perform literature searches - pulling at loose threads, following leads, making connections, and building an "internal map". It's similar to what people do when familiarising themselves with an area of knowledge, giving structure (or even meaning) to amorphous data. Chase and Simon in an article mainly about chess (1973) surmised that to become an expert required about 10 years of experience - the result of learning roughly 50,000 chunks of information. What this process of acquiring expertise entails, and the implications for poetry readers, is what I'm going to write about here.
I'll start with some vocabulary and ideas from Artificial Intelligence. Some of these terms have already leaked into common parlance.
- Crowdsourcing - getting answers using a large (mixed) group of people rather than an expert
- Swarm intelligence - the collective behaviour of decentralized, self-organized systems, natural or artificial. A swarm of bees is cleverer than a bee.
- Unsupervised learning - a class of problems in which one seeks to determine how the data are organized, looking for categories (using Clustering), pattern and structure (using pattern recognition). Adaptive resonance theory (ART) allows the number of clusters to vary with problem size and lets the user control the degree of similarity between members of the same clusters.
- Data mining - extracting patterns from large data sets by combining methods from statistics and artificial intelligence with database management.
- Chunking - collecting data into bigger units (for easier recall and manipulation)
Expert Performance and Deliberate Practice (by Ericsson) contains many references which I won't repeat here. S/he says that
- there are, at least, some domains where "experts" perform no better than less trained individuals
- In those domains where performance consistently increases, aspiring experts seek out particular kinds of experience ... the accumulated amount of deliberate practice is closely related to the attained level of performance of many types of experts, such as musicians..., chessplayers... and athletes
- the difference between experts and less skilled subjects ... reflects qualitative differences in the organization of knowledge and its representation ... Less skilled subjects' knowledge, in contrast, is encoded using everyday concepts that make the retrieval of even their limited relevant knowledge difficult and unreliable. The same acquired representations appear to be essential for experts' ability to monitor and evaluate their own performance ... so they can keep improving their own performance by designing their own training and assimilating new knowledge."
The resulting knowledge acquired is perhaps akin to the results of epidemiological studies in medicine - individual cases aren't studied deeply, nor are explanations necessarily sought. Instead, clusters, patterns and common features are identified. Another name for this type of knowledge might be "experience". Acquiring it's a skill (perhaps even a type of intelligence) that deserves a name.
In poetry the mined data is already chunked and inter-linked (a feature that prospective experts can exploit or ignore). Perhaps experienced poet would be expected to have read 50,000 poems. It doesn't seem hard to reach a certain level of publishable competence. Plateauing is common, and might be defeated by subsequent supervised learning.
Suppose (as is likely) some people are better than others at this type of learning. What characteristics might these people have?
- A multi-displinary approach (the general technique can successfully be applied to several fields)
- An openness to crowdsourcing - canvassing opinion from many sources (at a workshop, for example), not necessarily just experts
- A sympathy for the notion that you can be whatever you choose to be
- Analogical rather than analytical tendencies - not depending on detailed analysis, and not always able to explain their opinions from within the text
- They may have accurate expectations of how a poem will develop based on the first few lines
- They may not have a good memory for detail
In fact even without "understanding" a text, it's sometimes possible to assess it with surprising accuracy. Automated essay assessment is scarily successful. In practise critiquers will employ other methods too (close-reading, etc), though I suspect some critiquers lean more heavily on experience than others do (older people may depend more on experience not because they have more of it than young poets, but because their detailed memory is fading) . Poetry experience (like experience in chess) can provide an initial short-list of relevant features to consider at the analytical phase. And analysis can help guide the reader towards new regimes of self-training, enriching experience.
What remains unclear are examples of the chunks/categorisations that poetry experts employ and the domain-specific memory skills that let them keep in working memory relevant information. In chess the "chunks" are common formations (e.g. a fianchetto'd phalanx) but also collections of pieces that are somehow cooperating even if they're not close. Poets may may develop idiosyncratic categories ("allotment poems", "first person pieces where the punchline reveals that the persona's not the poet", "contradictory voices that resolve", etc). They may be able to remember blocks of words as a unit because they can feel how the words interact.
How transferable is this experience? Knowing the nature of the chunks might help with transfer, but it's not easy for experts to introspect. Employing experts as mentors is the most common method.