How does it determine where are the stresses? Is it based on just the word itself?

I'm very new to English poetry, I might be entirely wrong here. But when I input some famous poems, the way this app puts stresses is quite a bit different from my understanding.

For instance, for

> Bards of Passion and of Mirth,

It sees _and_ as unstressed, the second _of_ as stressed, and _Mirth_ as unstressed, which doesn't quite make sense to me to be honest.

> Tomorrow and tomorrow and tomorrow

It sees both _and_ as unstressed, while my understanding is that this verse is better read as fully iambic, putting stresses on both _and_.

> And what is love? It is a doll dress'd up / For idleness to cosset, nurse and dandle

https://i.imgur.com/raPMDnc.png

It sees four consecutive unstressed syllables. I'm not sure what is the logic here.

Thanks for the thoughtful feedback and questions! The meter detection algorithm is by far the most complex piece of this code, and you're right that it still sometimes generates unintuitive results. I'm always looking for more examples to add to my test suite to make improvements, so I really appreciate you sharing what you tried.

It tends to behave poorly when there's a word it doesn't recognize (like "doll'd"—I hope to add support for more of these poetic spellings just as I added support for spelling words like "dancing" as "dancin'"), so that might be partly at play with some of what you're seeing. But I think there's also something else going on here—it previously didn't ever generate so many consecutive syllables with the same stress, so I probably introduced a regression at some point. I'll investigate and add more tests for that too!

Lastly, a byproduct of my own background and the corpuses I use is that this does work better with "modern" poetry. For example, Shakespeare's Sonnet 29 starts with this line:

> When, in disgrace with fortune and men’s eyes

In the intended iambic pentameter, it would be:

> when, IN disGRACE with FORtune AND men's EYES

I don't know enough history to know if people ever spoke English like that, but that interpretation is way off to a modern reader. If I didn't know it was supposed to be iambic and was trying to diagram the stresses, I'd do something like:

> WHEN, in disGRACE with FORtune and MEN'S EYES

Meter is hard!

I got stuck on consistently detecting syllable stress when I tried to write a similar tool, so I'm really interested too.

For ridiculously-specialist poetry-writing software, the maker of that pretty Classical Chinese programming language ( https://wy-lang.org/ ) people were talking about had also previously made a similar editor for Tang-era Chinese poetry - https://github.com/LingDong-/cope , which keeps track of tone patterns and rhymes.

If you like this, you may like the Hemingway app: https://hemingwayapp.com/

Hi all! Versepad creator here. I did not expect to see this on HN today but I’m happy to answer any questions and would appreciate any feedback!

Very cool. I have been wanting some of these features for a while and finally got around to my own implementation: https://github.com/mcculley/WordWhittler

I welcome ideas on how to do smarter things around text composition.

I was heavily inspired having benefited from Remembrance Agent back when I lived in Emacs: https://github.com/zzkt/remembrance-agent

Suddenly I invested 30mins in writing some of the worst poetry. Great stuff! :)

Pretty cool, it taught me how to write a Haiku:

> Monkey eats apple

> Monkey prefers banana

> Suck it up monkey

That’d be a senryū. A proper haiku needs to have a kigo – a word hinting at a season – and kireji, a cut-word.

(And English is on average terser than Japanese: I read somewhere that an approximate equivalent of Japanese 5-7-5 morae is 3-5-3 English syllables, so the syllable count requirement is not absolute. Even Bashō’s famous crow haiku かれ朶に烏のとまりけり秋の暮 is 19 morae rather than 17.)

It’d be cool to have it support villanelles, where you can type in the first stanza and it’d pre-fill the patterned lines in the remaining five.

(After typing this, it occurred to me to check if Emacs has a mode that supports this. And, sure enough, it does! http://www.bobnewell.net/filez/poetry.el)

Thanks for the feedback! I would love to eventually support multi-stanza forms like villanelles and sestinas but that requires some more thought. Are there other forms or features you'd use?

The pantoum comes to mind. :)

Tried to write a limerick, surprised at how much smoother this made the process. Always interesting to see tools creatives use.

Thanks for the kind words! Limericks are a little tricky to nail down a "specification" for so I'm glad what I came up with worked well for you. <3

This looks really cool!

It feels like it would be a very small step to "RhymePad", i.e. an editor designed and optimized for writing rap lyrics which I think would be really cool, too.

Perhaps that already exists?

Amazing! Is it just me though, the rhyme scheme detection seems poor? I've been using the examples from their updates page, and even then it seems to miss internal rhymes?

Hi! You're right that rhyme detection can be improved, and rhyme scheme currently only considers words at the end of each line. I'd love to hear how you'd expect/want an internal rhyme feature to work. What would it look like? How would it avoid clutter when also showing syllables or meter?

Hi sorry for the late reply! If you're still checking this, then a UI like this might be useful! https://www.youtube.com/watch?v=k2ah9CtlaEs

It would be super cool to hook up GPT fine tuned on poetry dataset to complete the lines :).

It's impossible to work with it on mobile due to input highjacking. It actively tries to update my input any time I type something.

Thanks for the feedback! I made some changes a few years ago[1] to make Versepad work on my own iOS phone (the only mobile device I have). Can you share more details (or a video would be amazing) about your device and what behavior you're seeing so I can look into this?

[1] https://versepad.com/marginalia

Same