When you install WPML and add languages, you will find the flags here to change site language.

And now speech analytics?

The following is my take on “‘I gave 110%’ and Other Words that Predict Athlete’s Risk and Performance Behavior” by Dr. Roger Hall, presented at the MIT Sloan Sports Analytics Conference.

Yesterday, during the basketball analytics panel at Sloan (here’s a quick recap from Scott Sereday covering over at TrueHoop – Bullets from the basketball analytics panel), Mike Zarren of the Boston Celtics mentioned that the projecting players is still an area where success with analytics is lacking.  Paraphrasing Mike – “Projecting how an 18 year will perform at 25 based off numerical data is still very noisy.  We still have to rely a lot on visual data for that”.  The point?  It is tough to tell how a player is going to turn out once they get to the NBA – do they actually love the game, or are they just a big guy who knew they could make money by playing basketball?  Is someone dedicated enough to develop themselves once they have signed a multi-million dollar contract?  High school and college stats cannot provide that information.  Some skills translate better than others, like rebounding – but it is tough to tell which extraordinary talent is going to turn out like Kobe and which is going to end up like Tracy McGrady.

But I saw a presentation today that provided at least a glimmer of potential for addressing that problem.  And the presentation had nothing to do with basketball.  In fact, it did not even begin with a discussion of sports.
No, what most captured my interest today was a discussion that led off with talk of political science and speeches.  Dr. Roger Hall works for a company that has developed an algorithm to help predict who will win an election based off the words and phrases used in the speeches given by the candidates.  Transcripts are entered in and based off personality traits of the words and phrases that appear in the speech, profiles can be produced.  As a test of the profiling ability of the algorithm, speeches given by both government leaders (presidents, prime ministers, etc) and terrorists were entered without knowledge of who gave what speech.  With a 90% success rate, the algorithm could determine if the speaker was a government leader or a terrorist, based solely off the words used, not inflection, tone, intensity or anything else.

It just so happens so that another coworker of Dr. Hall is a football fanatic and he came up with an interesting experiment:  Can we predict what players will be arrested or suspended going forward?  That would be pretty useful information for a team to consider, would it not seem?  So, Hall and team set about it – but not with canned interviews from the combine or requests made specifically for the project.  No, they used interview clips from during a player’s college career:  Pre and post game interviews as well as midweek interviews, where the player was discussing football and the games played, not personal type stuff.  And the results were striking:  Certain personality traits evident in player’s speech significantly correlated to the rate at which they were suspended or arrested, with a low risk speech pattern having just 9% of the players included having been arrested or suspended during their career to the high risk grouping, where 30% had such an incident.

With those results in mind, other questions were asked:  What does the chart look like when I map the performance of a first round pick (pro-bowler, contributor, underachiever) to their speech?  What about the speech of quarterbacks against their level of success?  Again, the results appeared to show strong correlation between how a player spoke and what they achieved.  The quarterback graph was perhaps most striking, as there wound being three basic groupings – the instinctive speakers/players (active quarterbacks like Michael Vick), the analytic (Peyton Manning), and the neutral.  The striking one was the neutral, as it contained just about every backup quarterback in the league.  Based off interviews given in college, the algorithm could predict who wind up a backup in the pros, indicating a certain mindset evident in their speech patterns.

Now, wouldn’t it be interesting to see similar research done around potential NBA draft picks?  While I am not suggesting this is the end all of player development projection, I will say that based off what I saw today, and the NBA’s ongoing difficulties in projecting draft pick success, NBA teams might want to consider seeking out Dr. Hall and his company and seeing if similar patterns are evident in their own players.  And then just maybe a third type of data can be added to that mentioned by Mike Zarren – visual, numerical, and verbal.  It may not seem plausible – but any edge that can be gained in the risk management of signing teenagers to million dollar contracts would seem like a good investment.