Skip to content

Sound Poem 7

March 31, 2010

her reaction then again the same
what a blindness that I’ve heard
wondering what she tries to blame
working when your distance blurred

disturbing and purging my friends from me please
I worked hard for nothing but nothing is wrong
I could never be guilty of red with disease
but time same time and your frustration strong

she shuttered instinctively
that way away
an angels okay

March 29-31 2010, supervised generation from type-based n-grams using stochastic beam search and phonemic evaluation.  Source texts: Joy Division, Minor Threat, Nine Inch Nails and Suicidal Tendencies lyrics.  Generator: epogees

I’m trying to think of ways to quantify how much of a poem “comes from” the source texts as opposed to from the author. For example, in the poem above the following trigrams are common to the source text and the final poem:

  • she tries to
  • I could never
  • a blindness that
  • she shuttered instinctively

and the following 4-grams are common to the source text and the final poem:

  • then again the same
  • I worked hard for
  • disturbing and purging my

So 24 out of 68 common tokens are part of n-grams longer than bigrams.

But how to quantify the “force” that comes from a poem? One of the things I like about the poem above is the phrase “a blindness that I’ve heard” which, althought it contains a common trigram, diverges significantly from the source text (“A blindness that touches perfection” – joy division) It’s clearer that more of the effect of “disturbing and purging my friends from me” comes from the original text (“disturbing and purging my mind” – joy division) and even more in “she shuttered instinctively”, which is a single line in the poem above as well as in the original lyric.

Related to this: I thought about implementing a “track changes” feature similar to that of Microsoft Word, and measuring how many edits the human author makes, but I keep coming up with different ways of editing the text. This time what I did was: generate lines, delete words, then if something was missing I’d look through the language model and non-randomly pick a next-word I thought was good. (I ended up adding a ‘Next Words’ button to make it easier!) But how to quantify the varieties of intentional authoring vs randomness?

If I were writing a research paper about these issue I’d probably begin the title with “Towards…”, and propose some kind of scheme by which poetic techniques ranging from phonemic techniques like alliteration to lexical techniques like repetition to semantic techniques like metaphor were automatically identified (non-trivial!) and traced back to an origin in either 1) the original text, 2) randomly-generated novel text combinations 3) implicitly authored text combinations (i.e. by parameters set by the author) or 4) explicitly authored text combinations (i.e. non-random decisions made by the author). But how to measure which of those poetic techniques provided the “force” to the poem? I’m thinking some kind of dynamic approach: you give the reader a list of the automatically-identified techniques, have the reader weigh them by how much they felt each technique influenced their perception of the poem (criticism by parameterization – HA!) then determined: based on those weights, how much of the force of the poem for that reader came from the computational poet vs the source text vs the randomness of the generation algorithms.

I’m pretty sure I will never implement any of that, but it was fun to think through.  ok, later…

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: