Previous bottlenose dolphin studies suggest that the coarse envelope of an echo spectrum (“macrostructure”) has hierarchical dominance over finer-scale spectral features (“microstructure”) during synthetic echo discrimination tasks. In this study, two dolphins listened to and discriminated between underwater sound stimuli consisting of pairs of clicks with different micro- and macrostructures. After conditioning dolphins to reliably discriminate between two “anchor” stimuli with different micro- and macrostructures, probe stimuli, which contained a macrostructure identical to one of the anchor stimuli and the microstructure of the alternate anchor, were infrequently presented. Dolphins responded to probes in a manner consistent with macrostructure primacy.
I. INTRODUCTION
Bottlenose dolphins discriminate, identify, and classify objects that differ in size, shape, and material composition based on acoustic features of echoes generated by their broadband, impulsive sonar system.1 The mechanisms of object classification used by dolphins for active acoustic sensing in underwater environments are constructive to bio-inspired sonar system design, as dolphins outperform hardware sonars at target discrimination in complicated acoustic conditions.2,3 Dubrovsky et al. (1992)4 performed experiments to examine how echo features, including components of the acoustic power spectrum, are organized into a perceptual hierarchy. The results from a passive listening task with simulated echoes suggested that dolphins preferentially attend to the coarse features of an echo's spectral envelope [macrostructure, or macro power spectrum (MaPS)]. This superseded the importance of the microstructure, or micro power spectrum (MiPS), which consists of ripples within the MaPS resulting from multiple echoes arriving within the temporal integration window of the dolphin (∼260 μs).5 This hierarchy could be interpreted in light of the acoustic cues available in target echoes. Multi-highlight echoes within a ∼260 μs integration window might be identified as a target based on the MaPS matching the spectrum of the dolphin's emitted click, while MiPS features could specify the small-scale “parts” that belong to the target as a perceptual “whole.”6
This concept of hierarchical echo processing by dolphins is a foundation for future work. However, the previous work by Dubrovsky et al. (1992)4 relied on a single subject, and the reports remain outside peer-reviewed sources despite their substantial scientific merit. The present study further examined the echo-processing hierarchy of MaPS and MiPS in common bottlenose dolphins (Tursiops truncatus) utilizing paired-component comparisons, which pit two stimulus features against one another while holding others constant. The hypothesis of this study was based on the results from previous experiments;4 namely, that there is an informational hierarchy in dolphin echo discrimination in which the MaPS features of simulated echoes show primacy to fine-scale MiPS.
II. MATERIALS AND METHODS
This study used the general methods of Dubrovsky et al. (1992)4 in that dolphins were trained to distinguish between two “anchor” stimuli having unique MiPS and MaPS features. Stimuli were paired clicks that simulated basic aspects of echoes a dolphin might receive from a target with two closely spaced reflective surfaces. Following training, dolphins were infrequently presented with novel, unreinforced “probe” stimuli based on hybrids of the MaPS and MiPS features of the anchor stimuli. The responses to the probes indicated how MaPS, MiPS, or a combination of both features were used in echo categorization. Two experiments were conducted: in the first, a go/no-go paradigm was used; in the second, a two-alternative choice (left/right) paradigm was used. This study was conducted between April and November 2019, and followed a protocol approved by the Institutional Animal Care and Use Committee of the Naval Information Warfare Center (NIWC) Pacific and the Navy Bureau of Medicine and Surgery. It followed all applicable U.S. Department of Defense guidelines for the care and use of animals.
Two bottlenose dolphins, APR (female, 34 yr) and APO (male, 4 yr), participated in the study. The upper-frequency limit of hearing, defined as the frequency at which the auditory evoked potential hearing threshold was 120 dB re 1 μPa, was 100 kHz for APR and 80 kHz for APO.7 Thus, APR and APO exhibited some degree of high-frequency hearing loss relative to full bandwidth hearing in this species (upper-frequency cutoff of ∼140 kHz).
All sessions were performed in a 9 m2 netted enclosure located within San Diego Bay, CA. Dolphins stationed on an underwater biteplate attached to an aluminum frame (Ref. 8) that was secured in the center of the enclosure. A response paddle was located approximately 2.3 m to the left of the biteplate and at the same depth. A second, identical paddle was added to the right of the biteplate at an equal distance for the left/right task. The sound projector was positioned approximately 1 m directly in front of the biteplate.
A. Stimuli
The electrical waveforms used to generate acoustic stimuli were paired, 9 or 14 μs duration DC pulses with a 150-μs or 100-μs separation [Fig. 1]. Stimuli were generated using custom software and converted to analog with a PCI-6259 data acquisition (DAQ) card (National Instruments, Austin, TX). The clicks were filtered (200-kHz lowpass, 3 C module, Krohn-Hite Corporation, Brockton, MA), routed through an SR560 preamplifier (Stanford Research Systems, Sunnyvale, CA), and amplified with either a 7600 M or 7602 M power amplifier (Krohn-Hite Corporation). These produced broadband impulses, or clicks, when transmitted from an ITC 5446 directional transducer (International Transducer Corp., Santa Barbara, CA). Stimuli were named Anchor 1, Anchor 2, Probe 1, and Probe 2 [Fig. 1; also see Ref. 4]. Under these conditions, DC pulse duration governed the MaPS of the click pair (i.e., the spectral envelope), and time separation governed the MiPS (i.e., peaks and notches in the spectra were spaced at a frequency interval equal to the inverse of the click separation time). Anchor 1, consisting of 14-μs duration clicks, produced MaPS 1 (also a component of Probe 1), while Anchor 2, consisting of 9-μs duration clicks, produced MaPS 2 (also a component of Probe 2). Similarly, the 100-μs time separation between clicks produced MiPS 1 (Anchor 1 and Probe 2), and the 150-μs time separation produced MiPS 2 (Anchor 2 and Probe 1). Note that the dolphins' high-frequency hearing loss limited the availability of some high-frequency information in the stimuli, but that characteristic MaPS and MiPS features were still available to each dolphin.
(Color online) Simulated echo stimuli. Acoustic spectra and paired click waveforms (black inset) for each stimulus organized according to features of interest (MaPS and MiPS). The duration of each DC pulse and time separation between pulses within a pair are given (scale bars shown for Anchor 1 apply to all panels). Responses to the anchors were trained, whereas probe sounds were the test stimuli. Probe 1 shared macrostructure with Anchor 1 (MaPS 1, solid red outline) and microstructure with Anchor 2 (MiPS 2, blue textured details), while Probe 2 shared macrostructure with Anchor 2 (MaPS 2, blue double outline) and microstructure with Anchor 1 (MiPS 1, solid red details).
(Color online) Simulated echo stimuli. Acoustic spectra and paired click waveforms (black inset) for each stimulus organized according to features of interest (MaPS and MiPS). The duration of each DC pulse and time separation between pulses within a pair are given (scale bars shown for Anchor 1 apply to all panels). Responses to the anchors were trained, whereas probe sounds were the test stimuli. Probe 1 shared macrostructure with Anchor 1 (MaPS 1, solid red outline) and microstructure with Anchor 2 (MiPS 2, blue textured details), while Probe 2 shared macrostructure with Anchor 2 (MaPS 2, blue double outline) and microstructure with Anchor 1 (MiPS 1, solid red details).
Calibrations of all sounds were performed using a TC4013 hydrophone (Reson Inc., Slangerup, Denmark) positioned at the location between the dolphin's two ears when it was on station, but without the dolphin present. Stimuli were calibrated in terms of sound exposure level (i.e., related to stimulus energy per unit area) and were presented at mean levels of 120 dB re 1 μPa2s for the go/no-go experiment, and 110 dB re 1 μPa2s for the left/right experiment. Individual stimulus presentations roved ± 3 dB to eliminate potential loudness cues. Stimuli were presented along with band-limited, spectrally white masking noise, delivered using the same 5446 projector following generation with a National Instruments USB-6251 DAQ card, lowpass filtering using the SR560 preamplifier, and amplification by the 7600 M/7602M amplifier. Masking noise was digitally compensated to correct for the ITC5446 transmitting response and any multipath interference.9 Noise mean pressure spectral densities were set to 70 dB re 1 μPa2/Hz (±1 dB) from 20 to 200 kHz. Masking noise created a flat noise floor to provide a calibrated, consistent signal-to-noise ratio instead of the more variable ambient noise in San Diego Bay.
B. Psychophysical procedures
1. Go/no-go experiment
Dolphins were trained to station themselves on the biteplate and swim to press the paddle (“go”) in response to Anchor 1, and to withhold responses (“no-go”) to Anchor 2. In each session, the anchor type was pseudorandomly chosen for each trial, resulting in a 50% proportion of each anchor type. Each trial began with the dolphin being cued to swim and station itself at the biteplate. Following a random interval of between 0.5 and 3 s, 40 stimuli were projected to the dolphin at a rate of 20/s. For a correct response to Anchor 1, the dolphins were required to leave the biteplate within a 1-s response window after the stimulus ended and touch the response paddle. For a correct response to Anchor 2, the dolphins were required to remain on the biteplate during the full response window. To provide feedback to the dolphins on anchor trials (including during the left/right task, below), a reinforcement indicator sound was played following each correct response, while a different sound was played after an incorrect response (projected by a DAEX25W-8, Dayton Audio, Springboro, OH). Sessions generally consisted of 50–70 trials.
Probe trials were introduced into the sessions after correct responses to anchors were > 90% over 3 consecutive sessions. No feedback sounds were played after probe trials regardless of response, and dolphins were instead recalled to the trainer without fish reinforcement. The first 10–20 trials of a session were conducted without probes and designated as “warmup” trials. If the subject responded correctly to 100% of the first 10 anchor trials or > 85% of the first 20 anchor trials, then each subsequent block of 10 trials contained a single, pseudorandomly chosen probe. If correct performance on anchors was ≤ 78% (i.e., two or more incorrect responses) within any 10-trial block containing a probe, the response to the probe presented therein was discounted. After 16 sessions with APR and 18 sessions with APO, an additional probe was added for each block of 20 trials, increasing the proportion of probe trials from 0.1 to 0.15. The purpose of this increase was to determine whether responses to probes would be extinguished with repeated presentation and consistent absence of a food reward.4 Typically, 5–9 probe trials were presented in a single session throughout the go/no-go experiment.
A total of 94 go/no-go sessions were completed with the two subjects (44 with APR and 50 with APO); however, four of APO's sessions were eliminated for failure to meet warmup criteria, and one was eliminated because post-session calibrations differed from pre-session calibrations by more than 2 dB. A total of 615 probe responses were analyzed. Fifteen probe responses were eliminated because two or more errors were made on anchor trials within the 10-trial block where the probe was presented.
2. Left/right experiment
One of the dolphins (APO) participated in the left/right task, which included a second response paddle located to the right of the biteplate station at a relative distance equal to that on the left. The dolphin pressed the left response paddle for Anchor 1, as in the go/no-go experiment. However, instead of remaining on the biteplate upon presentation of Anchor 2, the dolphin was trained to press the right response paddle. This was done to reduce ambiguity about the categorization of Probe 2 (i.e., did the dolphin truly categorize Probe 2 as being similar to Anchor 2, or was Probe 2 not similar enough to Anchor 1 to elicit a paddle-press response in the go/no-go experiment?). The proportion of probe trials was 0.15 for the first 15 sessions but was reduced to 0.05 for sessions 16–46. Fifty left/right experimental sessions were initiated but only 46 sessions were analyzed; four sessions were eliminated because APO failed to meet the warmup criteria. APO's responses to eight probe trials were also eliminated from analysis due to poor performance on anchor discrimination during the presentation block. This left a total of 187 probes (94 Probe 1 and 93 Probe 2) for analysis.
III. RESULTS AND DISCUSSION
A. Go/no-go experiment
All probe and anchor responses are plotted against session number in Fig. 2(a), and the percentage of paddle press responses for each stimulus are reported in Table I. Both of the dolphins correctly responded to the anchor stimuli at high rates, although there were a few occurrences with the first 20 sessions where more errors were made. Dolphins pressed a paddle in response to 93%–98% of Probe 1 trials, and 20%–29% of Probe 2 trials. In contrast to prior findings,4 neither subject in this study ceased responding to Probe 1 with repeated presentations in the absence of reinforcement. This was unexpected, as there was an energetic cost (albeit small) associated with responding relative to remaining on the biteplate. Since the dolphins' probe responses were not extinguished and Probe 1 response rates were similar to Anchor 1, data from the go/no-go experiment did not provide certainty that the dolphins could discriminate Probe 1 from Anchor 1. The proportion of responses from both dolphins pooled for Probe 1 (0.96, 95% CI [0.93–0.98]) was nearly identical to Anchor 1 (0.97, 95% CIs [0.97–0.98]), in contrast to the response proportions for Probe 2 (0.25, 95% CI [0.2–0.3]), which were appreciably higher than for Anchor 2 (0.02, 95% CI [0.02–0.03]) (see Table I). Furthermore, APR's responses to Probe 2 increased dramatically beginning at session 26 (see Fig. 2(a)), suggesting that dolphins could distinguish Probe 2 from Anchor 2, and that they used the MiPS 1 feature to some degree in categorizing Probe 2 trials. This conclusion is supported by Dubrovsky et al. (1992)4 who suggested that while MaPS was hierarchically dominant to MiPS, dolphins also appeared to perceive the MiPS dimension when classifying the probes into the binary anchor categories.
Responses to probes plotted against session number. Each Probe 1 trial is represented by a black triangle and each Probe 2 trial is represented by an open circle. The gray trace on the top of each plot represents the percentage of correct anchor discriminations (righthand axis “Anchor Percent Correct”) for each session. (A) Go/no-go experiment. Dolphins (APO top; APR bottom) responded to probes by either pressing the paddle (“Response”) or withheld a response and remained stationary on the biteplate (“No response”). (B) Left/right experiment. APO pressed the left paddle in response to Anchor 1 and the right paddle in response to Anchor 2. For probe trials, the dolphin was not forced to choose a paddle and had the option of remaining on the biteplate station (withheld response). Four Probe 2 trials resulted in a withheld response (shown as “X”).
Responses to probes plotted against session number. Each Probe 1 trial is represented by a black triangle and each Probe 2 trial is represented by an open circle. The gray trace on the top of each plot represents the percentage of correct anchor discriminations (righthand axis “Anchor Percent Correct”) for each session. (A) Go/no-go experiment. Dolphins (APO top; APR bottom) responded to probes by either pressing the paddle (“Response”) or withheld a response and remained stationary on the biteplate (“No response”). (B) Left/right experiment. APO pressed the left paddle in response to Anchor 1 and the right paddle in response to Anchor 2. For probe trials, the dolphin was not forced to choose a paddle and had the option of remaining on the biteplate station (withheld response). Four Probe 2 trials resulted in a withheld response (shown as “X”).
Paddle-press responses to each stimulus type in the go/no-go experiment. Number of responses and the corresponding percentage of total trials for each trial type are shown. Rows subdivide the results by the proportion of probe trials, and overall number and percentage of responses are shown in bold. For Anchor 2, paddle-press responses were incorrect while Anchor 1 paddle-press responses were correct.
Subject . | Proportion of probe trials . | Probe 1 (% of total) . | Probe 2 (% of total) . | Anchor 1 (% of total) . | Anchor 2 (% of total) . |
---|---|---|---|---|---|
APO | 0.10 | 28 (82%) | 5 (14%) | 464 (92%) | 10 (2%) |
0.15 | 104 (96%) | 25 (22%) | 765 (98%) | 3 (0%) | |
Overall | 132 (93%) | 30 (20%) | 1229 (95%) | 13 (1%) | |
APR | 0.10 | 39 (93%) | 3 (8%) | 520 (98%) | 33 (7%) |
0.15 | 117 (100%) | 45 (36%) | 873 (99%) | 7 (1%) | |
Overall | 156 (98%) | 48 (29%) | 1393 (99%) | 40 (3%) |
Subject . | Proportion of probe trials . | Probe 1 (% of total) . | Probe 2 (% of total) . | Anchor 1 (% of total) . | Anchor 2 (% of total) . |
---|---|---|---|---|---|
APO | 0.10 | 28 (82%) | 5 (14%) | 464 (92%) | 10 (2%) |
0.15 | 104 (96%) | 25 (22%) | 765 (98%) | 3 (0%) | |
Overall | 132 (93%) | 30 (20%) | 1229 (95%) | 13 (1%) | |
APR | 0.10 | 39 (93%) | 3 (8%) | 520 (98%) | 33 (7%) |
0.15 | 117 (100%) | 45 (36%) | 873 (99%) | 7 (1%) | |
Overall | 156 (98%) | 48 (29%) | 1393 (99%) | 40 (3%) |
The dolphins increased their response rate to both probes when the probe presentation rate increased (see Fig. 2(a)). Most dramatic was that APR responded to 8% of Probe 2 trials when the proportion of probe trials was 0.10, and 36% of Probe 2 trials when the proportion of probe trials was 0.15 (see Table I). This pattern, in addition to the high Probe 1 response rates, suggests that the dolphins were biased toward a paddle press response for both probes. Many repetitions and the increase in the proportion of probes per session led to an overall increase in paddle-press responses to probes rather than the expected decrease. This is potentially due to the low cost of a paddle response that had an extensive history of reinforcement.
B. Left/right experiment
APO's responses to probes and overall performance on anchor stimuli are plotted against session number in Fig. 2(b) and summarized in Table II. A correct response to Anchor 1 was a left paddle-press and a correct response to Anchor 2 was a right paddle-press (see Sec. II). In the first seven sessions and in sessions 19–27, APO responded to all Probe 1 trials by pressing the left paddle, consistent with MaPS primacy. APO's responses to Probe 2 were more variable; during some sessions APO adopted a strategy of responding to all probes with the same paddle, despite maintaining consistent anchor performance [e.g., session 14, Fig. 2(b)]. After observing this behavior, the proportion of probe trials in a session was reduced to compensate for the effects of probe recognition. APO appeared to favor the left paddle in early sessions and the right paddle in later sessions. The proportion of left paddle press responses for Probe 1 (0.76, 95% CI [0.66–0.84]) was appreciably less than for Anchor 1 (0.94, 95% CI [0.93–0.94]). This suggests that Probe 1 was indeed discriminable from Anchor 1, although the go/no-go task was unable to definitively demonstrate this. The proportion of right paddle press responses for Probe 2 (0.68, 95% CI [0.57–0.77]) was much lower than for Anchor 2 (0.99, 95% CI [0.98–0.99], which suggests that APO could also discriminate Probe 2 from Anchor 2. That APO chose not to respond by pressing either paddle on four Probe 2 trials in the left/right task also supports this. In contrast, APO always responded to Probe 1 with a paddle-press in this experiment. This was likely because of reinforcement history from the previous go/no-go task, specifically the association between Anchor 1 and a paddle press. Overall, APO's probe responses in this experiment exhibited less fidelity to the MaPS feature than in the go/no-go task, which is potentially indicative of reward-seeking behavior rather than a reflection of changing hierarchical feature perceptions.
Responses of subject APO in the left/right experiment. The number of probe responses to each paddle (left or right) is shown. NR = no response. Rows subdivide the results by the proportion of probe trials (Probe prop.). The number of total responses for each probe that were congruent with MaPS primacy are shown in bold with percentage in parentheses. Number and percentage of overall responses to anchor stimuli are also shown, with correct responses in bold.
Probe prop. . | Probe 1 (% of total) . | Probe 2 (% of total) . | Anchor 1 (% of total) . | Anchor 2 (% of total) . | |||||
---|---|---|---|---|---|---|---|---|---|
Left . | Right . | Left . | Right . | NR . | Left . | Right . | Left . | Right . | |
0.15 | 48 | 10 | 17 | 36 | 0 | 375 | 32 | 6 | 395 |
0.05 | 23 | 13 | 9 | 27 | 4 | 826 | 39 | 5 | 878 |
Overall | 71 (76%) | 23 (24%) | 26 (28%) | 63 (68%) | 4 (4%) | 1201 (94%) | 71 (6%) | 11 (1%) | 1268 (99%) |
Probe prop. . | Probe 1 (% of total) . | Probe 2 (% of total) . | Anchor 1 (% of total) . | Anchor 2 (% of total) . | |||||
---|---|---|---|---|---|---|---|---|---|
Left . | Right . | Left . | Right . | NR . | Left . | Right . | Left . | Right . | |
0.15 | 48 | 10 | 17 | 36 | 0 | 375 | 32 | 6 | 395 |
0.05 | 23 | 13 | 9 | 27 | 4 | 826 | 39 | 5 | 878 |
Overall | 71 (76%) | 23 (24%) | 26 (28%) | 63 (68%) | 4 (4%) | 1201 (94%) | 71 (6%) | 11 (1%) | 1268 (99%) |
IV. CONCLUSIONS
This study supported the earlier findings of Dubrovsky et al. (1992)4 in showing that bottlenose dolphins attended to the acoustic power spectrum envelope, or MaPS, over the MiPS when performing synthetic echo categorization tasks. While the hierarchy is a feature of dolphin perception of pulse-pair stimuli in passive listening tasks, how it translates to the perception of the spatial features of biosonar targets remains unknown. The primacy of MaPS features in biosonar suggests that they are the most salient cue that dolphins use in attending to a target, while a subdominant MiPS dimension might reveal finer structural details. Demonstration of how these acoustic features translate into target perception during biosonar tasks should be an interesting—albeit challenging—topic of future research.
ACKNOWLEDGMENTS
We acknowledge funding from the Office of Naval Research Code 32 (Mine Countermeasures, Acoustics Phenomenology and Modeling Group). This is scientific contribution No. 272 of the National Marine Mammal Foundation.