My name is Parry and I’m here to say / Natural language processing is worth a play / There’s technology out there making rhymes / It makes you money all the time.
OK, that was the worst rap ever. I won’t quit my day job, I wasn’t destined to be a rapper. Even though I grew up on the mean streets of Vancouver’s suburbs, the ancestral home of gangsta rap.
Luckily, there is an emerging field within artificial intelligence that will prevent me every having to punish you with my sick rhymes ever again.
Which verse is by the famous rapper?
One of the lyrics below was written by a famous rapper. One was written by a computer. And one was written by me.
I gots big love for you, fa sho
I know, you know, we know
Truth be told I loves u for real yo
Our love blows up like volcanos
I know, you know, we know
Cause I slow ya roll and you know it bro
Let me really tell you how it should be told
You think not? You could see though
What we have is real
Wanna shout it from the mountains baby
That I love you, but if they don’t hear me
I know, you know, we know
Give those three verses a read, and… let’s take a poll. Which one is the real rapper?
Let’s not deconstruct the deep, philosophical meaning of these lyrics. We’ll leave that for the unemployed English graduates of the future.
Let’s look at the linguistic construction of the phrases:
- Versions A, B and C all rhyme, or at least follow a rhyming pattern.
- Versions A, B and C all follow a clear rhythmic pentameter.
- Versions A, B and C are all plausible verses in a rap song.
But only one of them was written by a famous rapper.
Now – to avoid spoilers and cheaters – here’s a brief musical interlude to make sure you don’t look at the answers before you fill out the poll above.
This interlude is brought to you by an amazing rapper whose name (Childish Gambino) was computer generated, which seems fitting…
Here’s my prediction of the results:
Bear in mind the results are tallied in real time and I’ve no idea what the results will be at the time of writing this blog.
But, here’s what I predict the results will say: none of the 3 verses will be a clear winner.
I reckon all three will get some votes, unless this vote gets tainted by rap nerds who look the lyrics up. Assuming that won’t happen, there will be no clear winner. So don’t be a rap nerd, let’s keep this an untainted poll based upon your own gut response to the lyrics.
OK, so, which one was written by a real rapper?
Click here to see the poll results in real time
Version A was created by a straight gangsta homeboy who is a devilishly handsome, self-described digital anarchist. Yep, that verse was my own amazing attempt at being a rapper. I’m the next Macklemore, that’s clear.
Version B was created by an algorithm devised by Finnish computer scientists.
Version C was created by Talib Kweli in his song, “We know.”
How did the Finnish scientists generate their verse?
They used “Natural Language Processing,” a field of artificial intelligence that has incredibly exciting commercial potential.
To quote their paper:
“Our work lies in the intersection of the areas of computational creativity and information retrieval. In our approach we assume that users have a certain concept in their mind, formulated as a rap line, and their information need is to find another line, or a set of lines composing a song.”
In other words, they take existing rap lyrics. They interpret themes within the text. Then, they determine the number of syllables, and the rhyming and pentameter constructs. Then they computationally select relevant lines to make an effective verse.
Here’s an example of another lyric they created, alongside the source artist:
|For a chance at romance I would love to enhance||Big Daddy Kane – The Day You’re Mine|
|But everything I love has turned to a tedious task||Jedi Mind Tricks – Black Winter Day|
|One day we gonna have to leave our love in the past||Lil Wayne – Marvin’s Room|
|I love my fans but no one ever puts a grasp||Eminem – Say Goodbye Hollywood|
Note how each lyric is from a different artist. Yet, they all share the same theme (love), the same “flow,” or pentameter, and they all rhyme. And they don’t just rhyme at the end of each phrase like a junior-high love letter – the entire stanza has a human-sounding rhythmic pentameter.
The scientists measured their success on a metric they call “rhyme density.” This is a statistic that measures how many rhymes exist both within a stanza, but also within a given line, and is a metric for overall flow of a rap verse.
Their algorithm out-performs the rhyme density of regular rappers by 21%. Sheeeeeit.
By the way, the rapper in the research with the highest rhyme density was Inspectah Deck, one of the original Wu’s who was never very commercially successful on his own. Which is a shame, as he is a fantastically talented rapper. Listen to this song and take note of his rhyme density. The algorithm even out-performed him.
The full research paper can be found here (PDF.)
That’s “dope,” as the kids say, but what does it mean for marketers?
Human language is an incredibly complex problem to solve computationally. I sincerely doubt a computer will be able to write better lyrics than Big L (RIP), for example, notwithstanding his helpful translation of Ebonics.
But within small language spaces, for example AdWords, email subject lines, or calls-to-action on websites, there is HUGE potential for the artificially intelligent optimisation of language.
Here’s a real marketing use case: email subject line optimisation
Email subject lines are, in essence, a set of words, phrases and sentiments that have a specific, measurable outcome – that is, making someone open and act upon your email.
For any given subject line, there are trillions (for real, yo!) of possible ways to convey your message. The human brain, however, is creatively limited… and is fundamentally not objective. Humans can’t think of even a minor fraction of all the possible variants.
Using artificial intelligence you can consider huge volumes of data – exponentially more than a human brain can – and then apply artificial intelligence to optimise marketing language generation.
There is software out there that does this right now. Full disclosure: my company Phrasee does this, so I’m obviously biased, but there are other companies who are commercialising this emerging technology as well. It’s a nascent but fast growing market. Early adopters of this technological innovation are reaping the benefits.
The future of copywriting is artificial
Many marketers are precious about their words being used, like they’re the next 2Pac of the marketing world (NSFW)… and are unwilling to believe that artificial intelligence can out-perform them.
Perhaps it’s job security worries. Or perhaps it’s because they don’t understand the power of natural language processing technology. Or perhaps it’s just good old inertia and risk-aversion.
The main point is this:
If you’re precious about your own words being used in your marketing, regardless of the quantitative results, then carry on as you are. No worries, this blog ain’t about you.
If you would rather use artificial intelligence to improve your marketing language and increase your response rates, then you could be the next straight marketing gangsta.
If you take away one thing from this blog post, take away these wise words, which were (obviously) written about natural language processing being a nascent technology with incredible potential.
Uh 1-2-3 that’s how it be
Somebody gotta be doubtin’ it
It was me Michael T.T. handing out freebies
Let’s party everybody bounce with me
BTW: this was a computational combination of Outkast, KRS One, Ghostface Killah and 50 Cent. Just in case you were curious. I’ve no idea who Michael T.T. is, but I’m sure he’s a nice guy.