Who’s Better at Figuring Out Song Lyrics, Humans or Machines?

Mondegreens, those rhythmically-correct, yet, lyrically-incorrect attempts to hum a song are fun. They are part of the human experience of listening to music. People cannot help but try to attach lyrics to those they enjoy but cannot hear accurately. There Is often that humorous moment when we later learn what the actual artist sang. Nevertheless, the mondegreen, more often than not, remains stuck in our heads.

So, can a machine create the same sort of funny, yet inaccurate, mondegreens? That was what researchers recently tested by placing the IBM Watson Machine against professional transcriptionists. The machine, of course, had never heard the songs previously. Meanwhile, the research team chose songs the humans would be hearing for the first time.

The results were unexpected. Take a look at what happened when researchers pitted a translation machine against a human transcriptionist.

Here is What Happened

The Song List

  • Taylor Swift, “I Knew You Were Trouble.”
  • Van Halen, “Runnin With the Devil.”
  • Elton John, “Tiny Dancer.”
  • Brad Paisley, “Ticks.”

IBM Watson Computer Translation Results

  • Taylor Swift, “I Knew You Were Trouble,” resulted in 4 lyrical errors and 6 missing words.
  • Van Halen, “Runnin With the Devil,” resulted in 9 lyrical errors and 15 missing words.
  • Elton John, “Tiny Dancer,” resulted in 8 lyrical errors and 3 missing words.
  • Brad Paisley, “Ticks,” resulted in 12 lyrical errors and 5 missing words.

Human Transcriptionist Results

  • Taylor Swift, “I Knew You Were Trouble,” resulted in no errors or missing words.
  • Van Halen, “Runnin With the Devil,” resulted in no errors or missing words.
  • Elton John, “Tiny Dancer,” resulted in no errors or missing words.
  • Brad Paisley, “Ticks,” resulted in no errors or missing words.

Research Conclusions

Neither party provided the much anticipated mondegreens. Overall, the IBM Watson Machine proved unreliable in interpreting what it heard. Unlike the average human music lover, Watson could not attach a series of words that would make sense, even if not what the artist sang. Instead, the computer chose random words that sounded similar to an individual lyric. Humans are probably just better at grouping words, even when they do it incorrectly. Watson looks for individual words, not phrases.

On the other hand, the professional transcriptionists proved too perfect to help create any mondegreens. Perhaps next time, the research team should have Watson compete against untrained transcriptionists, who might provide some funny mondegreens.

Watson may have to practice a bit more before his next battle. When it comes to translating songs, humans seem better than machines, at least for the time being.

 

Advertisements

9 Comments Add yours

  1. Robin says:

    Wonderful way to start the education new year. Thanks.

  2. Marge Bartholomew says:

    Hi, thanks for all your helpful posts.

  3. Brianna says:

    Love the website! Thanks for the great ideas!

  4. Sue says:

    I shared this information on my Pinterst board entitled “tech in the classroom.” This is a great offer; thanks for putting it out there for your followers!

  5. Natacha Delgado says:

    Crossing my fingers!!!! 🙂

  6. mdg2 says:

    Good ! Will help my students -Kenya Africa

  7. Susanna Everingham says:

    Your information has helped me greatly and now I am passing it on to my daughter who has just started teaching. Would love to win!

    1. Thanks for the kind words, Susanna!

  8. Rebecca Murphy says:

    Tweeted and following your blog. Thanks for doing what you do!

Leave a Reply

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

WordPress.com Logo

You are commenting using your WordPress.com 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 )

Google+ photo

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

Connecting to %s