An automatic lyrics evocifier.

I got an interesting comment to my previous post on the lyrical analysis (of Nocturnal by The Black Dahlia Murder). The idea that was proposed went as follows:
  • Would it be possible to build a tool that could take a (first draft) of the lyrics to a song and propose more evocative words, or maybe introduce certain figures of speech?
  • Would someone want to use it?
My short answer would be: 
Build such a tool? probably yes. 
Would someone want to use it? it depends.

Let's dig a little deeper and see what such a tool could look like and what kind of features it might have:
  1. A thesaurus: I would probably start with this. Easiest way of getting synonyms. I can envision a tool where you paste in your text, it gets analysed and you get a smart ui for displaying the results. Maybe the original text with possible word replacements in a different colour and when you hover over them, you get a pop-up with possible replacement words.
  2. Smart replacement algorithms: You can't just blindly replace words in a song as the timing of the song needs to be respected. Therefore, you need more or less the same number of syllables when you replace something or you have to replace multiple words. You might need to take phonetics into account as well to make sure replacement words sound "correct" next to the other words.
  3. Figures of speech: Now we get into more tricky territory. Working figures of speech into lyrics will be hard. Some of the more technical ones (like an alliteration) seem doable, even though you will be working on multiple words at once. But it doesn't take long before you need your tool to be able to understand the meaning of words, which brings us to...
  4. Semantics: The joy and trouble with English words is that you usually have several synonyms for a word that mean more or less the same, but convey slightly different nuances of that concept. Unfortunately songs can be about anything, which means your tool needs to know about the general meaning of words as well. There isn't one specific domain that you could try to model. You basically have to do them all.
  5. Automatic translation: There exist a lot of bands whose members do not natively speak English. If they could write in their mother tongue and have tool translate it correctly to English and introduce more descriptive words and figures of speech while all the time maintaining the underlining meaning and atmosphere... Killer feature, I'd say.
Looking over the list, I think it would be possible to build a tool that does 1, 2, and at least part of 3. Entries 4 and 5? years away.
Part of the problem is indeed also who would be using the tool. At least for metal bands, not all of them go the atmospheric way. If you're in a down-to-earth band with songs about real life and society, then you are going to want to use words like "bloody" instead of "sanguinary". And even if you would want to use such a tool, a lot of bands probably would not want to pay for it. You can find a thesaurus online for free, ask friend who knows English well to help out, and save the money for other stuff like instruments...
Nevertheless, I found it a cool thought expirement. Now it's up to you! Did I miss a cool other possible feature? Am I underestimating what is possible linguistically? Or if you're in a band and write lyrics, would you pay for a tool like this?

No comments: