leonardcohenforum.com

Hey fellow Leonard fans! I used to be a bit active here in the old days
Please help me It may look like spam but I swear it isn't!

Short version - please rank the following Leonard Cohen *lyrics* from the saddest/most negative (first place) in your opinion to the happiest (last place). Try to refer to lyrics only and not the music. It only takes a minute and you'll be helping a study! Thanks

Please answer here:

https://surveys.enalyzer.com?pid=meqg6dum

In order to vote: click "select ranking items" and rate the songs in the popup. You can freely change your order/answer until you submit.

If you're wondering why I need this, read on...

I'm a Computer Science student and I'm conducting a research on Sentiment Analysis - which is basically an algorithm's ability to "understand" emotions in text.
I gave the algorithm every song Leonard (and several other artists I love) ever wrote and asked it to give it a score of how happy/sad it is. Part of the study is to compare the algorithm "opinion" to human opinion. Which is where you come in! I want to know what's the fans opinion on saddest/happiest songs, and compare it with the algorithm, to see how good it is.

So please help me! If there's interest I'll update you with the results

Thanks a lot!
Yaniv

If I may be so bold, I would suggest that the main problem with the premise of your study is the severe limitations of its parameters. To characterise Leonard's lyrics on a continuum with "happy" at one end and "sad" at the other is to completely miss the vast depth and profundity of his poetry and lyricism. In fact, "happy" and "sad" seem altogether ineffective adjectives to describe Leonard's lyrics (for the most part).

Thus, I cannot see how any relevant or useful information can be derived from forcing Leonard's lyrics into such a limited and honestly rather unfitting framework. The answers you would get on your survey could not possibly accurately capture the emotional (and intellectual) impact of Leonard's lyrics, since true feelings on them would likely be far more complex and multifaceted than simply "happy" or "sad". Leonard's lyrics also have a strong intellectual component -- they "mean" something (or, several things), even if the meanings cannot be readily derived due to the depth and complexity of Cohen's writing.

I assume that your algorithm uses things like keywords, and that certain words are associated with certain emotions, such that if certain words are present in a lyric, then the lyrics are deemed either happy or sad, is that how you've designed it? That may work (somewhat) for very simplistic lyrics, but, again, I cannot see how it would or could adequately reflect the emotions of poetically complex lyrics like Leonard Cohen's.

Hey,
I agree the definition is very simplistic - we're not looking for the algorithm to completely analyze the songs. Leonard is also quite an ambitious targets - in other artists it was much easier to make the decisions (like The Cure - quite a difference between "Disintegration" and "Friday I'm in Love"). But nevertheless, we wanted to check the results in very poetic and difficult lyrics.
We don't expect the algorithm to actually succeed - but there's still academic interest in seeing it's results and analyzing where it failed.
Thanks!

It's more than just a matter of interpretation. I don't think even the most basic affective response to some of these songs can be captured by the categories (sad/happy). Just to take one example, I wouldn't use either of those words to characterize my basic emotional response to "First We Take Manhattan." Others on the list are designed to evoke ambivalence or a somewhat tense, conflicted affect ("Hey, That's no Way to Say Goodbye," for example, is a song about the sadness of parting in which the singer is urging his partner not to be sad--I really have no idea of how to rank the purity or intensity of the sadness that's part of that song next to, say, the sadness of "It Seems so Long Ago, Nancy," which is sadness mixed with anger, irony, more than a little guilt, and a certain gratitude, as well).

So, beyond a very general sense that some of these songs ("Nancy," I guess, for example) are sadder than others, or have more to do with sadness than others, I don't know how to rank their relative sadness. It's not something that I think makes sense as a thing to do (for a human listener, let alone a computer).

Maybe it would make more sense to see if the computer would agree with human beings about what the core or dominant emotion of a given song might be? That might be a more natural thing to ask of people and a more interesting thing for a machine to try doing. Ranking just doesn't make any sense to me.

Just my thoughts after looking at the list and seeing if I could do what you've asked me to do.

Best,

Louis

yaniv297 wrote: ↑Thu Jan 09, 2020 1:05 pm ...We don't expect the algorithm to actually succeed - but there's still academic interest in seeing it's results and analyzing where it failed....

The problem remains that you are using the wrong tool to measure something that requires a different tool (if such a tool even exists). Trying to measure the supposed sadness/happiness quotient of a Leonard Cohen song with a sad/happy algorithm is rather like trying to measure how far the moon is from the earth using a kitchen weigh-scale -- it is not a matter of analysing where it fails, but simply recognising beforehand that you're using the wrong kind of tool, such that any results you get will be irrelevant (you don't need to spend all of your time and money on a study to come to that conclusion!).

As Louis also said, above, it doesn't make sense to try to use this algorithm with Leonard Cohen's songs. Like him, I found myself at a complete loss as to how to rank the songs on your scale, because they simply do not fit into the framework in which you are asking us to put them. Any ranking I would give them, with your criteria, would be totally meaningless, since I'd just be making up something that I don't actually feel or think.

Okay, I will do that. I know this thread is quite old but I just want to help you somehow.

I did the ranking. Interesting idea to see how an iterative learning algorithm would classify lyrics vs. humans who would tend to classify based more on intuition*. Something like computer chess vs human. I'd say classification of "positive emotion" vs "negative emotion" could be broader definition rather than sad/happy.

It would be interesting to see how the algorithm fares with lyrics that are clearly one or the other (i.e. positive or negative) vs ones that are more ambiguous and complex, as others have noted regarding LC's lyrics.

*although neuroscientists may tell you humans also use a form of iterative learning in drawing conclusions based on previously existing experiences (of LC's lyrics). So another control might be to use people well familiar with LC's lyrics and another group who have never seen them before.