How Do You See Machine Learning Impacting The Study, Production, And/Or Enjoyment Of Literature? (1/12/20)

[Marc]

I don't want to limit any of the responses or how your imagination might play with this question, so no qualifiers. If you want some examples of the intersection of machine learning and literature, here's a brief article from Wired Magazine.

[Lily]

My initial reaction to the link was that sometimes a title can be more tantalizing than the article that supposedly supports it. But, on second thought, earlier tonight I was browsing a National Geographic article on the analyses of ancient (sacred) texts and the thousands of (academic) man years that are willing to be put into those efforts. Then, too, I lay contrast to my current reading of Sterne's Tristram Shandy -- blessings on automated programs mimicking/analyzing it.

Okay -- links: The Life and Opinions of Tristram Shandy, Gentleman
Galatea 2.2

[Hugh]

I hope it will be a long time before a machine could predict what I will find interesting to read - if Amazon/GoodReads recommendation algorithms are anything to go by, we are not anywhere near that yet. I can see the point of doing a certain amount of statistical analysis of text, particularly when trying to establish authenticity and perhaps for etymology, but actually understanding a complex work of literature would be pretty advanced AI.

[Neil]

Richard Powers’ Galatea 2.2 is an interesting book to read.

[Marc]

It should be relatively easy for a machine to make decent reading recommendations to a human given data like Amazon and GR possess. It would be difficult to pick a new author or that random book that seems to resonate with an individual despite not being their standard fair. Given that most of us have no trouble locating books we're excited to read, this strikes me as low on the machine learning wish list.

Some ideas that popped into my head:
- A novel co-authored between a human and a machine/algorithm
- The ability to pull up every instance of a phrase or parallel scenes across the entirety of written literature
- Near instantaneous plagiarism verification
- Incredibly detailed looks at word and grammar usage changes over time
- New tools/insights enabling understanding and comparisons across literature by languages/culture/region
- Create-your-own literary mashups

[Antonomasia]

I've lost count of how many times I've posted this over the years but I've found Amazon UK's customers also viewed/bought algorithms absolutely invaluable. There are some academic areas in which they can provide a useful overview of recent-ish book publications in a field which are being quite widely read, or which classics from a certain time and place are available in critical editions like Penguin and Oxford, in a much shorter time than trawling each imprint's list.

Digital humanities research papers have occasionally been useful to cite in discussions. I'm not sure everything relevant has yet been scanned, but they can provide much more comprehensive answers to some questions than was previously possible

[Whitney]

Antonomasia wrote: "I've lost count of how many times I've posted this over the years but I've found Amazon UK's customers also viewed/bought algorithms absolutely invaluable. There are some academic areas in which th..."

Interesting comments about the Amazon recommendations. Years and years ago, I found their recommendations more useful. I think then they were solely based on information from people with a similar purchasing history. These days, like Hugh, I find them pretty useless. My assumption is that marketing has been heavily factored into the algorithm. I'm wondering if you find them more useful because in a more obscure area, there isn't the marketing (or people gaming the system) factored in. This is complete speculation, of course.

[Antonomasia]

Yes, I am mostly using them in more obscure areas.

The last thing I used them for was just a few weeks ago, to look through Celtic and Anglo Saxon texts, and also Sagas, in Penguin and Oxford.

But even with something like newly published literary fiction or non-fiction, they include other new books so that it's quicker than looking at many media features, because there are several popular new books all in a small space with no scrolling or clicking required. I have got successful gift ideas from them more than once. The only occasions when I don't find them interesting to browse, to see the connections others' taste is making, is when it's a book on a prize longlist that has been out at least a couple of weeks and the other longlisted books have become embedded as connected. (A lot of the specialist areas where I use the recommendations are probably embedded in a similar way, by students and academics.)
And they are better than the GR recs which more and more (since they attempted to improve the algorithm a few months ago) include bizarre stuff. I very rarely find GR recs interesting to browse as an illustration of other people's taste (more as a cringe at the badness of the algorithm), but the Amazon ones are to me.

[Whitney]

Antonomasia wrote: "I very rarely find GR recs interesting to browse as an illustration of other people's taste (more as a cringe at the badness of the algorithm).."

Out of curiosity I just selected 'Browse Recommendations' in GR, yes to your cringe assessment. It looks like, "oh, you've read some books tagged as "classic", here are some random "classics" for you to read. Same with "Adult Fiction" etc... I suspect that this is where Clippy found a job after being fired from Microsoft.

[Carol]

I expect that they’ll try to use machine-learning trained apps to take the initial cut at determining which manuscripts or submissions should be reviewed by humans. That’s the use case across other industries. That, and resume review for open roles. We’ll never know what brilliant talent the apps rejected. But heads will be reduced and money saved.