The Rowland Institute at Harvard

Fish Laboratory

Ava Chase

Fish Film Footage

In the Fish Lab at the Rowland Institute, we explore perceptual and cognitive capabilities in fish.  Recently completed work used koi carp as subjects in investigations of high-level stimulus control involving complex auditory stimuli, including music.  As mentioned, a new paper describes this work (Ava Chase.  Music discriminations by carp (Cyprinus carpio).  Animal Learning & Behavior, 29(4), 336-353, 2001).

The paper describes a series of four experiments: 1. Discrimination and Categorization Based on Musical Genre, 2. Iterated Reversals, 3. Transition to Synthetic Keyboard Music and Polyphonic Discrimination with Identical Timbre, and 4. Discrimination of Melodies Without Timbre Cues.

Here are links to movie files of one of the fish, Oro, performing in the melody experiment.  There is one movie each of a positive and negative trial.  Each movie is about 45 seconds long and includes sound and a follow-along musical score.  As they are long files (6.5 and 7.7 MB in length), it's probably best to download them to your computer's disk and play them from there. Download Oro with a [positive stimulus] and with a [negative stimulus].  Right-click and choose  Save Target As  or  Save Link As  being sure to note where you saved the file in your computer.

To view the movies you'll need Media Player (included in Windows), QuickTime or another mpg viewer.  QuickTime is available from Apple's Web Site.

While you're waiting for the downloads (get them both underway at once; it will take about 5 minutes on a 350k/sec connection) here's a short explanation of what you'll see in the movies. 

The melody experiment employed two very similar melodies and the purpose was to determine if the fish could tell them apart.  The positive stimulus (S+) was a familiar 84-note sequence from Rachmaninoff's "Rhapsody on a theme by Paganini," while the negative stimulus (S-) was the same 84 notes played in reverse order.  (By this means all the available cues (timbre, rhythm, note-duration sequence, etc.) were the same for both melodies; only the note sequence was changed.  As explained in the paper, the positive (S+) melody was played by a musician on a Kurzweil keyboard and electronically recorded as a MIDI synthesizer score.  The negative (S-) melody was created from the positive MIDI score by editing it to reverse the note pitch sequence.)

The melody experiment consisted of 74 sessions, each generally lasting nearly two hours.  During each session a computer conducted trials by randomly selecting either the positive or negative stimulus and playing it, with 20-second inter-trial quiet periods.  The full 74-session experiment took about 6 months and contains over 10,000 trials. 

In both movies Oro is in the left experimental half of the tank; Pepi and Beauty can be seen on the right watching through a clear partion.  An underwater speaker is visible at the left end of the tank and the black Chase Fish Button can be seen on the gravel at the bottom.  A pellet dispenser can be seen mounted on the top left of the tank, with a hose leading into the tank.  A nipple on the end of the hose keeps the reward pellet in place just below the surface, where the fish can suck it out.

If Oro presses the Fish Button during a silent period or while S- music is playing no food reward will be forthcoming.  But if Oro presses the button during the S+ music a bright light comes on outside the tank (this is the conditioned reinforcer) and the dispenser drops one pellet into the hose.  When Oro sees the reinforcer light he knows a food pellet will be available and he goes up to get the food.  (Of course, if Oro thinks he's made the correct decision, he'll be on his way to get the reward regardless of the light.) 

As the movie starts a silent period is underway and Oro is swimming around waiting for a trial to begin... 

Please read the paper to see how nicely this experiment turned out.  But the graph below can give you a quick preview.  Note how Oro's results were relatively unaffected by changing the timbre, lowering the scale of S+ by a fifth, or removing eleven (the first 4 and last 7) notes of the melody.  This would seem to indicate Oro wasn't using simple pattern recognition to recognize the correct melody.

Edited by Winfield Hill     Last modified Wednesday, February 14, 2002.