Females of many species choose mates using multiple sensory modalities. Multimodal noise may arise, however, in dense aggregations of animals communicating via multiple sensory modalities. Some evidence suggests multimodal signals may not always improve receiver decision-making performance. When sensory systems process input from multimodal signal sources, multimodal noise may arise and potentially complicate decision-making due to the demands on cognitive integration tasks. We tested female túngara frog, Physalaemus (=Engystomops) pustulosus, responses to male mating signals in noise from multiple sensory modalities (acoustic and visual). Noise treatments were partitioned into three categories: acoustic, visual, and multimodal. We used natural calls from conspecifics and heterospecifics for acoustic noise. Robotic frogs were employed as either visual signal components (synchronous vocal sac inflation with call) or visual noise (asynchronous vocal sac inflation with call). Females expressed a preference for the typically more attractive call in the presence of unimodal noise. However, during multimodal signal and noise treatments (robofrogs employed with background noise), females failed to express a preference for the typically attractive call in the presence of conspecific chorus noise. We found that social context and temporal synchrony of multimodal signaling components are important for multimodal communication. Our results demonstrate that multimodal signals have the potential to increase the complexity of the sensory scene and reduce the efficacy of female decision making.

Field Collection:

We conducted all experiments at the Smithsonian Tropical Research Institute (STRI) in Gamboa, Republic of Panama, between 4 June and 7 August 2019. We collected female and male túngara frogs as amplectant pairs nightly between 1930 and 2100 h. We brought the pairs to our laboratory at STRI facilities and placed them in total darkness in a cooler for at least 1 h before testing. This hour allowed their eyes to dark adapt after collection with flashlights (Fain et al. 2001; Taylor et al. 2008). On a typical night, we collected 6–14 pairs of frogs with roughly 70% of those females responding in a phonotaxis trial. Sample size for all treatments in this study was n=32; a total of 288 frogs texted in the nine treatments. 

Experimental Design:

We tested all females from 2200 to 0330 h on the same night we collected them. The temperature of the sound attenuation chamber (Acoustic System, ETS-Lindgren, Austin, TX, USA), where we tested the frogs, was kept ca. 27°C, similar to nighttime temperatures in the field. We illuminated this chamber with a GE nightlight (ca. 5.8 x 10^-10 W/cm², model no. 55507; Fairfield, CT, USA) to resemble a nocturnal light level within the range of natural light experienced by túngara frogs (Taylor et al. 2008). For each trial, we placed a female under an acoustically and visually transparent plastic funnel (ca. 10 cm diameter) at the center of this chamber. The plastic funnel had a cutout in the front where we put black mesh, so that females could evaluate the visual cues while under the funnel. The funnel initially held the females equidistant at 80 cm from two speakers (Mirage Nanosat Black, Klipsch Audio, Indianapolis, IN, USA) separated in azimuth by 60°. In all experiments, we antiphonally broadcast amplified (NAD C-316BEE, Pickering, Ontario, Canada) stimuli using Adobe Audition 2.0. The stimuli were natural calls, “Od” and “Sc,” with a known preference function (75% of females prefer Od to Sc; Ryan and Rand 2003; Coss et al. 2021). To avoid confounding stimulus call property differences in background noise, we chose to use a single call pair. In this study, we were not testing general female preference functions. Instead, we were interested in testing the influence of various noise types (visual and acoustic) on the ability of females to express their mating preference. We adjusted the stimuli so that their peak amplitudes measured 82 dB SPL (fast, C weighting, re 20 µPa) from the female’s release point using a Larson Davis 831 Class 1 SPL meter (Larson Davis, Depew, NY, USA). All sound files played the stimuli on a 3 s loop with an inter-stimulus interval of 1.1 s. Between trials, we alternated the stimuli broadcast from the speakers to avoid potential side bias. The 1.1 s interval provided enough time to add an asynchronized robofrog vocal sac inflation in visual noise treatments (see below).

For each trial, we separated a female from her male and placed her under the funnel. We then played the treatment files for a 2 min acclimation period. Females under the funnel could see and hear the stimuli during this time. We then released a female by raising the funnel with a pulley system outside the chamber. If the female entered a 5 cm radius around a speaker and remained there for at least 3 s, we scored her as making a choice. Alternatively, if the female failed to leave the initial funnel zone within 2 min or wandered around the chamber for 10 min without making a choice, we removed her from the chamber and retested her once more later that evening. If a female failed to make a choice on her second attempt, we did not include her in the dataset. We also recorded the female’s latency to choose and noted her path to the chosen stimulus. All trials were video recorded using an infrared video camera mounted on the ceiling of the chamber directly above the funnel zone and EthoVision (Noldus) software. Following these general procedures, we conducted a series of experiments (Table 2) to test female responses in the presence of unimodal noise (acoustic- or visual-only) compared to multimodal noise. Given that female túngara frog mate preferences are consistent (Ryan et al. 2019a), female responses in phonotaxis experiments provide a robust measure for testing how background noise from multiple modalities affects female perception of males.

Acoustic Noise Trials: Previous research demonstrated that higher densities (duty cycles) of chorus noise, reflective of nights with bustling choruses, can impair female ability to discriminate between males, while lower densities of chorus noise have no significant effects on discrimination (Coss et al. 2021). These results revealed the chorus density threshold at which females can no longer distinguish males in heterospecific noise: gladiator (Hypsiboas rosenbergi) and hourglass (Dendropsophus ebraccatus) tree frogs, and conspecific acoustic noise. In this study, we compared the acoustic, visual, and multimodal noise treatments to the control to examine the impact of increasing noise complexity. As a summary of how the acoustic noise treatments were conducted in our previous study, we broadcast either low-density conspecific or mid-density heterospecific noise, corresponding to nights with low- and mid-level calling activity (Coss et al. 2021), from two speakers mounted on the back wall (behind the speakers playing the acoustic stimuli) in each of the corners (125 cm above ground; 210 cm from the female release point; Fig. 1A). Mounting these two noise-broadcasting speakers above the floor of the chamber and on the corners generated an even, chorus-like distribution of noise within the chamber. As with the acoustic stimuli, peak amplitudes of the acoustic noise (both conspecific and heterospecific) were calibrated to 82 dB SPL from the funnel zone. This noise level is typical at frog ponds.

Visual Signals/Noise Trials: To simulate a visually noisy environment, we placed three robotic túngara frogs (robofrogs) with inflatable vocal sacs in the chamber (one in front of each stimuli speaker and one in between these speakers; Fig. 1B). Robofrog models were designed and 3-D printed by Savi Made (Windsor, Ontario, Canada). We made the vocal sacs using silicone rubber (Dragon Skin^TM 10 Medium) and solvent (NOVOCS^TM Gloss; Smooth-On, Macungie, PA, USA). We added a pastel purple dye to the mixture to replicate the color of a túngara vocal sac. After casting the molds, we placed a thin line of white silicone pigment on front of inflated vocal sacs (10 cc) to resemble the white stripes present on the túngara vocal sac. We attached the opening of the mold to silicone tubing (Tygon® SPT 3350; United States Plastic Corporation®, Lima, OH, USA) using a silicone rubber adhesive (Sil-Poxy®; Smooth-On, Macungie, PA, USA). This tubing was connected to one of three pneumatic pumps located outside the test chamber. These pumps were controlled by a computer, allowing the robofrogs to be triggered to inflate and deflate either synchronously or asynchronously with the acoustic stimuli (see Taylor et al. 2008 for additional details). The middle robofrog (not associated with a speaker) inflated during the 1.1 s silent gaps between the acoustic stimuli and completely deflated before the onset of the next call at either of the two stimulus speakers, thereby acting as visual noise. In one treatment, the robofrogs associated with the stimulus speakers inflated synchronously with their respective acoustic stimulus (generating a multimodal stimulus). In the second treatment, the robofrogs inflated asynchronously with the acoustic stimuli, where inflations began 130 ms after the closest acoustic stimulus ended, providing additional visual noise. Thus, the robofrog inflation overlapped some background noise but was fully asynchronous with the call at its stimulus speaker. This created visual noise at the stimulus speaker. In both treatments, the acoustic stimuli and robofrog combinations on the left and right alternated, such that there was no overlap between them. The asynchronous visual noise treatment was designed to mimic a complex chorus, where a female may be able to see a male’s vocal sac inflate but not detect his call due to acoustic masking. Males in nature are constrained to inflate their vocal sac synchronously with their call and females do not respond to the inflating vocal sac unaccompanied by a call (Taylor et al. 2008). Thus, we considered any vocal sac inflation that was not in-phase with a call as visual noise. Also, because females preferentially respond to a synchronous inflating vocal sac, we considered any robofrog inflating synchronously with the call at a speaker to be a multimodal signal.

Multimodal Noise Trials: We merged acoustic and visual noise to make several combinations of multimodal noise treatments (Table 2; Fig. 1C). These treatments contained either the mid-density heterospecific or low-density conspecific acoustic noise (Coss et al. 2021). Visually, treatments had: a) synchronously inflating robofrogs at the stimulus speakers, with only the independent, middle robofrog acting as visual noise, or b) all three robofrogs inflating out-of-phase with the calls at the two stimulus speakers. In the synchronous multimodal noise treatments, like in the synchronous visual treatments, the robofrogs on the right and left were synced to their respective stimuli with an extraneous middle robofrog (visual noise) not synced to either the stimuli or the acoustic noise. Multimodal noise treatments were primarily used to examine if the integration of acoustic and visual noise alters female perception of the stimuli or cognitively overwhelms females. Again, we consider synchronous inflation of the vocal sac as a signal and asynchronous inflation as noise.

Statistical Analyses:

We tested for differences in female phonotactic preferences between the control and noise treatments using a binomial distribution (SISA binomial calculator; Uitenbroek 1997). We report significance levels for preferences as mid-P-values, previously recommended for use in categorical data for smoothing radical changes in P-value generated by discrete data (Agresti 2001; Hwang and Yang 2001). For all statistical analyses comparing preferences in experimental treatments to the control, expected values were set to a 0.75 probability value. We chose this value based on previous experiments confirming a consistent 75% preference for the Od over Sc call (Ryan and Rand 2003; Coss et al. 2021). After transforming the data with a natural log function to satisfy assumptions of normality and homogeneity of variance, we performed an Analysis of Variance (ANOVA) on the latency data from this study. All statistical analyses on latency data (α = 0.05) were conducted in R software (R Core Team 2018).