Skip to content

sound poem 4

March 7, 2010

thus took after due grows to divert
and dignity best bred that dost faculty hurt
bear makes beauty and thy thee be that
but badges to her best thoughts and what at

that said then in how thy thus herself
with heart hath how thou heaven himself
on from fading grief this faith though
that thy thereof bud their therein throw

when sick sin love dost stol prayers
show sightless things so constant speak affairs
the hideous presence is closure even scorn so
looking so sweet plagues by be o

like lover in i love beauty to abuse
beauty which works to willing please use

March 5-6 2010, supervised generation from class-based n-grams of shakespeare’s sonnets using stochastic beam search and phonemic evaluation

Funny what a little rhyme can do. The last couple sound poems I wrote weren’t that satisfying – vowel shifts alone weren’t enough to carry them. So I added a simple rhyme model and fixed the alliteration feature, and now things look a lot more promising.

If I highlight a word in the output, it looks for a line or word that rhymes with that. If I don’t select anything, it looks for a line or word that rhymes with the last word in the textarea!

Anyways, I generated this poem one couplet at a time by:
1. generating 1 line at a time until I found a line I liked
2. generating lines that rhymed with the previous one, 1 at a time, until I found one I liked. if line 1 had no rhyme available, I started over
Note that I generated and selected lines, but I didn’t edit them.

For the first stanza I minimized all phoneme evals except Alliteration (10) and the Plosives (1 each). I like line 3 best:
bear makes beauty and thy thee be that
Note that ‘th’ is a non-sibilant fricative rather than a plosive; the line gained no evaluation score boost for that.

For the second stanza I minimized all except alliteration (10) and non-sibilant fricatives (10). Line 4 of this stanza shows the danger of running too many search iterations:
that thy thereof bud their therein throw
I was running this stanza on beam search with 10 individuals over 10 iterations. The danger is that it does such a good job finding a high-scoring line that it sounds like a toungue-twister. (With stanza 1 I was running 6 iterations of 5 individuals, which is why those ‘th’es could get through.) This suggests that running higher-iteration searches doesn’t always get you better verse, at least with my current approach to evaluation.
BTW, running phoneme evaluations at 10 rather than 1 risks lines without alliteration – selected for the value of consonant phonemes not at the end of the line.

For stanza 3 I focused on alliteration (10) and sibilant fricatives(1). It sounds kind of Lovecraftian! Finally, for the last stanza I focused on alliteration (10) and approximants (1).

As before, my sketchbook is here.

No comments yet

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: