Holy machine learning, Batman
Researchers from Tel Aviv University have published a study in Scientific Reports that explains how they used machine learning to better understand what Egyptian fruit bats are “saying” to each other. With machine learning, it becomes possible to carry out large scale analysis of animal calls, and understand “both the behavioral context as well as the identities of the emitter and the addressee” communicating with one another.
Focusing on “aggressive” communications, what the researchers found was that they’re arguing about 60% of the time – eating, sleep, mating, and personal space.
They were also able to sometimes predict (with 41% accuracy) the outcomes of the exchanges, and noted that the bats implicitly recognized different genders when calling out at each other.
A better understanding of animal communications requires this increasing level of specificity, as what we generally do know about their vocalizing is based on understanding “statements” made to an entire group, like warning “this area is dangerous, run” or “stay back, this is my territory.” It also makes clear that each communication is more detailed than a simple “go away,” since it depends on what the bats are fighting over.
The team of Yosef Prat, Mor Taub, and Yossi Yovel (who runs the Bat Lab at TAU) spent 75 days recording the bat colony using modified voice recognition software originally built for human speech. In this time, 14,863 female bat calls were recorded, as was a video record of these bats interactions during this period. The machine learning software then used visual and auditory cues to sort their behaviors into different categories, as well as note the outcomes of the bats’ fights with one another.
From this, using modified voice recognition software, the algorithm was able to identify the bat “speaking” 71% of the time, and what that bat was “saying” 61% of the time.
The researchers told The Guardian that the system could potentially be used for other animals, and would probably be most valuable in discerning what other kinds of social animals are saying to one another, like dolphins or monkeys, rather than just their addressing a group.