The Fog Index

Thank you, Al. For my part, apology accepted.

Brian Rush said:

Not really. Remember, we're not talking about every step of the vetting, editing, and publishing process being automated; there's no way to do that, at least not using current technology, and so if your work passes the first hurdles it will still be considered by a human professional. It's only the first step of the process that could possibly be automated.

Click to expand...

And so I can't really see the value in it. It's surely much cheaper to let an intern weed out the absolute rubbish, and forms a useful part of the education of a publisher.

The errors weeded out would be those that show a basic inability to write

Click to expand...

- that's such a difficult concept to spec out. How do we define 'ability to write'?

and/or inability to follow correct query procedures: incorrect formatting

Click to expand...

- we'd have to set it to only weed out egregiously bad formatting, because I don't want to filter out good things that have come out slightly off. The bar would have to be low.

, incorrect genre,

Click to expand...

- Impossible for a computer to assess in the foreseeable future - I can think of a lot of books it'd have a problem with even if you could work out a way of doing it. How can you algorithmically determine whether a book is SF or horror? Furthermore, how do you end up with new genres or with books that break genre rules?

certain tell-tale grammatical errors (the algorithm would probably need to make an exception when those errors occur inside quotes),

Click to expand...

- Again, I have no idea what grammatical errors you could call 'tell-tale', and the ability to spot them implies, to my mind, some ability to parse the structure of natural language. And there are, again, plenty of books that make grammatical errors part of their style. (Trainspotting springs to mind. Do we have to write a Scots argot module for the filter?)

It's hard enough to write a regex that can 100% accurately determine which parts of a manuscript are inside quotes.

- excessive misspellings (ditto),

Click to expand...

- Would have to set the bar pretty low again, I think, and it'd be hell on neologisms.

- If you can give instructions to a human being to do this, a programmer can give those instructions to a computer.

Click to expand...

Well, I think there are a couple of assumptions there that just aren't borne out by the facts. The first is that the first filter is about stuff that an algorithm could possibly pick up. I don't read the first page of a manuscript looking for misspellings or grammar errors or the wrong genre. The second is that most of the things we can instruct a human to do cannot currently be emulated by software.

There's no guarantee of course that what passes the program's muster will be publishable, and no guarantee either that something good won't be weeded out by mistake on rare occasions, but that happens now, too.

Click to expand...

True, but at least it happens according to human responses. We're trying to produce stuff that people will like, not robots.

I've seen programs that will write articles. Seriously. If that can be automated, so can the first steps of vetting a story. Also, so can the most basic editing.

Click to expand...

I think the programs that write articles aren't working off an understanding of content, they're data mining tools. What you get is a summary of other articles

Torgo said:

- Would have to set the bar pretty low again, I think, and it'd be hell on neologisms.

Click to expand...

Yeah, China Mieville and many other SFF writers would never get through to the human readers.

Al Stevens said:

I apologize for what I wrote and for the offense I've given.

Click to expand...

Thank you, Al. That's much appreciated.

For those who work in publishing, if you could have a piece of software installed on your PC on Monday morning that could analyse manuscripts, is there anything you can think of that it could do for you, with today's technology, that would be genuinely useful and save having to hand the manuscript over to a human reader?

RobJ said:

For those who work in publishing, if you could have a piece of software installed on your PC on Monday morning that could analyse manuscripts, is there anything you can think of that it could do for you, with today's technology, that would be genuinely useful and save having to hand the manuscript over to a human reader?

Click to expand...

No.

No, because I don't want manuscripts analysed: I want to find one which sings to me.

As a former slushpile reader, I would guess that I could actually identify things faster than a computer could analyze a partial. So it's way more cost and time-effective to employ a cheap or free intern who has read a lot of books than to pay for software.

If diamond mines could extract huge diamonds for low or no cost, do you think they would still chop all the ore into tiny bits? Probably not.

However, I'm just chiming in to say, this program sounds like a horrible idea in for real use, but also sounds like a fabulous future dystopian plot bunny. A world where all fiction now has to pass filters for grammaticality, repetitiveness, grade level, etc. And then in our inciting incident the protag stumbles across some great work of the Western canon that has long since been thrown out...

Sounds like a great premise to me!

JSSchley said:

As a former slushpile reader, I would guess that I could actually identify things faster than a computer could analyze a partial. So it's way more cost and time-effective to employ a cheap or free intern who has read a lot of books than to pay for software.

Click to expand...

I agree. Again, the things like spelling and grammar are all fixable, if there's story there. Slush-reading, or really, ms. reading in fiction is all about "Does this text incite narrative lust?"

A really good reader can spot a story that she might not particularly like, but know that enough others will to make it worth publishing. A really good intern can be better than a Beowulf cluster.

This is a great idea for the fact that, it probably will happen. I wouldn't be surprised if this technique isn't already in use.

I'm far from an engineer, but my day job entails using software to analyze social media data (millions of short conversations), and some tools are surprisingly cheap and good at this already.

I don't think it would necessarily be a system looking for a certain type of grammar, or mastery of punctuation. I think it would have examined thousands of published manuscripts in a particular genre and it would look for similar characteristics. So, it's not really even looking for things to reject, but looking for certain special manuscripts to pass on that show promise. In that case, each editor could essentially train the tool to look for what they want.

Social media analysis platforms do this already. Some allow the user to analyze a small number of examples of content, and then the tool can recognize similar types.

And... of course as a wannabe writer myself, I would be scared something incredible gets missed. But the agency or editor could also train the tool to automatically promote something very different to be read by human just as easily.

Even though I think this type of thing will (could be already) happen, I do think the shame of it is gaming the system will become as much an art as writing the book itself for some.

BTFC said:

This is a great idea for the fact that, it probably will happen. I wouldn't be surprised if this technique isn't already in use.

Click to expand...

I'm curious how you came to this conclusion, particularly the part in bold, when all the industry professionals who have replied in this thread have said they don't and never would rely on AI prescreening.

What I can't see is how the software will answer the question, "Is it interesting?"

Agents or editors won't want this proposed software, but I can imagine thousands of wannabee authors hocking the kids to buy it.

Terie said:

I'm curious how you came to this conclusion, particularly the part in bold, when all the industry professionals who have replied in this thread have said they don't and never would rely on AI prescreening.

Click to expand...

Ok, well let me rephrase: I "wouldn't have been" surprised (I don't claim to work in publishing).

I do work in an industry that is similar enough that I realize this technology is very possible. Plus, the point of my post is that the best application of this kind of software is to pick out manuscripts that a reader might miss. Of course, if I'm on these boards I'm not hoping for machines to choose the next great American novel. But I can see a program like this being adopted widely.

A program that took any text received and chucked it into the house's style might be useful, as long as it did so with human oversight (And MS word almost--ALMOST--lets you make a button that does this... it almost always screws one or two things up, though, just to remind me that it was made by people). I'd hate to be in a world where the autotrons stopped House of Leaves from getting published.

~~

A program that searched the slush for easily changeable keywords that re-sorted the pile by relevance might be inherently valuable, though. Finding gems in the slush is as much about mood as quality, once the obviously bad things are gone. This wouldn't throw anything out; it would just sort by content filters instead of submission dates.

~~

Slush-spinning might be amusing; write a small program that presents a random paragraph from the slush pile. If it's interesting enough, you can open the source file. If its not, you can click for the next paragraph. But this would require agents and editors to have the free time to turn slushing into a game, since rejecting out of this mode would be patentedly silly in 99% of cases.

Al Stevens said:

I apologize for being misunderstood.

Click to expand...

hmm. Wondering if you see the logic problem in this sentence, Al.

Layla, that's already been dealt with and Al has apologised very nicely. Let's move on, please.

sure. I saw after I posted that I came in a bit late. Sorry about that.