Experimental evidence for compositional syntax in bird calls

Abstract

Human language can express limitless meanings from a finite set of words based on combinatorial rules (i.e., compositional syntax). Although animal vocalizations may be comprised of different basic elements (notes), it remains unknown whether compositional syntax has also evolved in animals. Here we report the first experimental evidence for compositional syntax in a wild animal species, the Japanese great tit (Parus minor). Tits have over ten different notes in their vocal repertoire and use them either solely or in combination with other notes. Experiments reveal that receivers extract different meanings from 'ABC' (scan for danger) and 'D' notes (approach the caller), and a compound meaning from 'ABC-D' combinations. However, receivers rarely scan and approach when note ordering is artificially reversed ('D-ABC'). Thus, compositional syntax is not unique to human language but may have evolved independently in animals as one of the basic mechanisms of information transmission.

Figure 1. Sound spectrograms of call treatments played to Japanese great tits.

(a) ABC call is composed of single A, B and C notes. (b) D call is composed of seven to ten D notes. (c) ABC–D call is the combination of ABC and D calls. (d) D–ABC call is a reversed combination of ABC and D calls. These calls were digitally edited using Raven Pro 1.3 software.

Figure 2. Usage of D calls in a non-predatory context in Japanese great tits.

(a) Effect of the presence of a mate on the production of D calls (n=187 observations, n=40 individuals): tits produced D calls more often when they visited the nest alone than when they did following their mate (generalized linear mixed model: χ²=5.00, df=1, P=0.025), after controlling for the nonsignificant influence of sex of the callers (χ²=1.16, df=1, P=0.281). (b) Effect of D calls on the recruitment of their mate (n=136 observations, n=34 individuals): tits that produced D calls were more likely to subsequently attract their mates than tits that did not produce D calls (generalized linear mixed model: χ²=35.37, df=1, P<0.0001), even after controlling for a significant influence of the responding mate's sex (males were more likely to approach D calls given by their partners than were females; χ²=9.32, df=1, P=0.002).

Figure 3. Responses of Japanese great tits to playbacks of ABC, D and ABC–D calls, and background noise (BN).

(a) Number of horizontal scans made by tits in 90 s (generalized linear mixed model: χ²=62.58, df=3, P<0.001). (b) Percentage of trials in which tits approached within 2 m of the loudspeaker (generalized linear mixed model: χ²=34.56, df=2, P<0.001). The box and whisker plots display the median value and 25 and 75% quartiles; the whiskers are extended to the most extreme value inside the 1.5-fold interquartile range. Sample size: n=21 individuals. Each individual was exposed to all four treatments in varied orders, giving n=21 samples per treatment.

Figure 4. Responses of Japanese great tits to playbacks of ABC–D and D–ABC calls.

(a) Number of horizontal scans made by tits in 90 s (generalized linear model: χ²=27.09, df=1, P<0.0001). (b) Percentage of trials in which tits approached within 2 m of the loudspeaker (generalized linear model: χ²=6.03, df=1, P=0.014). The box and whisker plots display the median value and 25 and 75% quartiles; the whiskers are extended to the most extreme value inside the 1.5-fold interquartile range. Sample size: n=34 individuals. Each individual was exposed to only one treatment, giving n=17 samples per treatment.