People today are truly lucky they can talk to their smartphones, tablets, or computers and receive appropriate responds. People can choose to write questions into these machines and still get the appropriate answer. It is great. The ultimate powerful computer, IBM’s Watson showed its major accomplishments, abilities, and vast knowledge on Jeopardy. This raises a question about Watson’s ability to transcribe song lyrics. Watson is so powerful that some may assume he can perfectly transcribe song lyrics. Watson matched up against two expert transcriptionists to determine which group would perform better. Any bets on which group won; well it will surprise everyone but the transcriptionists won with a perfect score! Poor Watson utterly failed on the few lines of lyrics from four different songs by four very different artist sung acapella. To be fair to Watson how many people misunderstand the lyrics to songs probably everyone has to some song.
The results for Watson were truly sad. Watson had trouble with the lyrics to “Ticks” by Brad Paisley with twelve errors, missed five words, and did no better with the lyrics of “Running with the Devil” by Van Halen with nine errors and fifteen missed words. He also failed with the lyrics to the third and fourth songs. He had eight errors and three missed words with the lyrics of “Tiny Dancer” by Elton John and did no better with the lyrics to Taylor Swift’s “I Knew You Were Trouble” with four errors and six missed words. Moreover, with all of Watson’s missed words and errors, you might think he would produce some mondegreens not just letters such as “m..m….m”. Watson created no mondegreens. Mondegreens are nonsensical words, at times amusing but do not have any meaning and created when someone misunderstands lyrics. Similar to the child’s game of “Telephone” where one person whispers a sentence to the next person and so forth until the last child. The results are usually very amusing and not at all like the original message. This is because in whispering the children do not always understand what was whispered to them.
This leaves an interesting question of whether Watson or any artificial intelligence would perform better with just the spoken word. Since we have proven from smartphones, computers, and tablets, we could take the leap that yes there is some difference in hearing and understanding lyrics from the spoken word at least for artificial intelligence. It maybe that singing of the lyrics with a different beats and tempos influenced how Watson heard and understood the sung lyrics. It would be interesting to see Watson’s or any artificial intelligence’s results if the lyrics were spoken instead of being sung. From Watson’s performance, one can also presume that if you have lyrics that you need transcribed, your best bet is to pick a human transcriptionist. Guess that is one job safe from Artificial Intelligence at least at this moment.