The study, conducted by Zijian Tian and Kazuo Hiraki of the University of Tokyo, set out to answer a deceptively simple question: does a baby’s assessment of an act of sharing change depending on whether the baby is judging from the viewpoint of the person giving, the person receiving, or a neutral bystander? Decades of research have shown that preverbal infants prefer helpers over hinderers, fair distributors over unfair ones, and protectors over aggressors. But those findings typically tested infants as third-party observers watching scripted puppet shows. The Tokyo team suspected that when the moral stakes are personal—when the evaluator is imagined to be a participant in the interaction itself—the picture might look very different.

To find out, the researchers built an animated scenario around a swing. One agent, the initial possessor, arrives first and claims the swing, riding it upward with sparkling sound effects that mark the experience as rewarding. A second agent, a newcomer, then appears and jumps repeatedly while facing the swing and maintaining eye contact with its occupant—a behavior designed to signal a request for a turn. In the Sharing condition, the occupant dismounts and trades places, letting the newcomer ride. In the Non-sharing condition, the occupant turns back and keeps the swing. The design was deliberately chosen because a swing can be used by only one agent at a time, making it an ideal stand-in for a scarce, desirable resource whose repeated monopolization would be hard to depict with hills or slides.

The team then ran four eye-tracking experiments with a total of 103 infants, all around eleven months old and recruited from the University of Tokyo Baby Lab. The experiments crossed two manipulations. The first varied which agent stayed constant across the Sharing and Non-sharing events: in Condition A, the newcomer was the same green sphere-shaped character in both episodes, while in Condition B, the initial possessor was. The second manipulation varied the test phase: in Experiments 1a and 1b, a central agent—either the newcomer or the initial possessor—chose between two characters, and infants’ anticipatory gaze during a three-second window before the choice revealed whom they expected the agent to approach. In Experiments 2a and 2b, infants watched the two agents side by side as third-party observers, with a swing-seat ring lifting one agent upward to signal a positive outcome.

The results were strikingly asymmetric. In Experiment 1a, where infants predicted the choices of a newcomer requesting a turn, their anticipatory gaze lingered significantly longer on the agent who had shared the swing. The difference scores—fixation time on the sharing agent minus fixation time on the non-sharing agent—were significantly greater than zero during the first block of test trials, with a two-tailed test yielding p = 0.016 and a directional test p = 0.008. In plain terms, the babies expected the newcomer to prefer the character that had helped it get a turn. That preference vanished entirely in Experiment 1b, where infants instead predicted the choices of the initial possessor who had been asked to yield. There, difference scores did not deviate from zero in any trial block. And in both third-party experiments, 2a and 2b, no preference for the sharing agent emerged at all, either before or after familiarization.

Statistical modeling reinforced the pattern. Linear mixed-effects models fitted to the gaze data showed a significant main effect of trial repetition but no main effect of experiment, and post hoc comparisons confirmed that the sharing preference in Experiment 1a was confined to the earliest trials. Looking time after the choice was made—the classic violation-of-expectation measure—showed no systematic differences between conditions in any experiment, a limitation the authors attribute in part to the fixed five-second post-choice window, which may have been too brief to capture effects that infant-controlled paradigms typically allow to run longer.

Why would eleven-month-olds reward sharing only when judging from the newcomer’s side? The authors point to the asymmetry built into giving and taking. For the newcomer, the agent who yields the swing is unambiguously beneficial: it removes the obstacle standing between the newcomer and its goal. For the initial possessor, however, yielding means losing playtime on a resource it controls. Prior work has shown that infants use spatial position, body size, and group size to infer social dominance and expect dominant individuals to prevail rather than yield—so from the possessor’s vantage point, holding onto the swing may not register as a moral failing at all. And as neutral observers, infants may simply lack the motivational stake that makes either interpretation compelling.

The team also ruled out a battery of low-level perceptual explanations. If infants merely preferred whichever character was more novel or had spent less time on the swing, then Experiment 1b should have produced an even stronger preference for the newcomer who never got on the swing at all. It did not. Nor could recency or primacy effects explain the pattern, since the structural differences between the familiarization conditions across experiments controlled for those factors. The authors conclude that the infants’ looking preferences are best explained by the socio-moral structure of the taking-turns interaction itself—specifically, who gave and who received.

The findings carry implications well beyond the swing. If early moral preferences are this malleable, the field may need to rethink how it measures them. Rather than asking whether infants are prosocial, the authors argue, researchers should map when and under what criteria these evaluations emerge. Open questions remain: the study tested only eleven-month-olds raised in Japan in a laboratory setting, and the animated scenario isolated a single non-reciprocal episode in which the giver received no gratitude or social reward—unlike real life, where toddlers as young as sixteen months show more happiness when giving than keeping. Whether infants possess an abstract sense of possession that underpins their judgments also remains untested, though prior research shows preschoolers infer ownership from first arrival and control of permission.

What the study does establish is that the seeds of moral reasoning are planted earlier and grow in stranger shapes than the classic helper-hinderer experiments suggested. A baby not yet able to say a single word already understands that the same act—giving up a swing—can look generous from one side of the interaction and costly from the other. Perspective, it seems, is not a skill that develops later in childhood; it is woven into the very first moral judgments humans make.

Subject of Research: Perspective-dependent socio-moral evaluation of sharing behavior in preverbal infants

Article Title: Preverbal infants' evaluation of sharing behaviors within a “taking turns” context is modulated by the perspectives they adopt

Article References: Tian, Z., & Hiraki, K. (2026). Preverbal infants' evaluation of sharing behaviors within a “taking turns” context is modulated by the perspectives they adopt. Frontiers in Developmental Psychology, 4, Article 1904707. https://doi.org/10.3389/fdpys.2026.1904707

Image Credits: AI Generated

DOI: 10.3389/fdpys.2026.1904707

Keywords: infancy, socio-moral evaluation, perspective-taking, sharing, eye-tracking, developmental psychology, turn-taking, social dominance, fairness, resource distribution, preverbal infants, University of Tokyo

The research, published in Frontiers in Mammal Science, was carried out by Julia Joukhadar, Anton Baotic and Angela S. Stoeger of the Acoustics Research Institute at the Austrian Academy of Sciences and the University of Vienna. Their team recorded a semi-free-ranging herd of seven African savanna elephants (Loxodonta africana) at the Adventures with Elephants conservation and research unit in Bela Bela, South Africa, between 4 and 28 November 2018. Across 18 daytime recording sessions, handlers divided the herd into two subgroups positioned about 200 metres apart with no line of sight between them, a standardized separation designed to elicit frequent contact rumbles while minimizing visual cues. Synchronized microphones and video cameras captured every rumble, yielding 459 annotated calls from three adult females, two adult males and two calves.

The central question was deceptively simple: does the same set of elephant rumbles look like coordinated communication no matter how you measure it? To find out, the researchers applied three previously published analytical frameworks to the identical dataset. The first, a quartile-based approach, sorts calls into overlapping, adjacent, intermittent or single categories using statistical boundaries derived from the latencies between vocalizations. The second, bout criterion interval analysis, uses log-survivorship curves to find the natural breakpoint separating calls clustered within a bout from those separated by longer silences. The third, antiphonal probability analysis, adapted from a 1977 study of white-throated sparrows, asks whether an individual produces more calls after another’s vocalization than would be expected by chance alone.

The results were strikingly divergent. Under the quartile-based method, nearly half of all rumbles — 49 percent — were classified as overlapping, meaning they began before the previous call had ended. Adjacent calls, the category most likely to represent potential antiphonal responses, were the rarest at just 14 percent. The temporal boundaries themselves shifted compared with earlier work: the upper limit for intermittent calls came out at 71 seconds, well below the 120 seconds reported in a foundational 2003 study. Because the boundaries are computed from the data’s own quartiles, a dataset rich in overlapping calls compresses the thresholds, so the same framework can carve the same behaviour into different categories depending on how aroused or urgent the animals happen to be.

The antiphonal probability analysis told a different story again. Using a 25-second response window, the team tested whether each elephant answered others more often than chance would predict. Only two of the seven individuals — the adult male Chova and the adult female Messina — produced significantly more responses than expected. The remaining five, including Shan, the herd’s most prolific caller with 115 rumbles, showed no statistically reliable evidence of turn-taking. The authors caution that this framework is inherently conservative: it assumes that answerable time windows do not overlap, which inflates the number of responses expected by chance during periods of intense calling and biases the analysis against detecting genuine antiphony. Calls from unidentified individuals, which made up 19 percent of the dataset, further complicated individual-level conclusions.

The bout criterion interval analysis produced yet another picture. A Kolmogorov–Smirnov test confirmed that the intervals between calls deviated significantly from the random, exponential pattern expected if calling were uncoordinated. A segmented regression on the log-survivorship curve identified a breakpoint at roughly 124 seconds — dramatically longer than the 1.5-second threshold reported for wild elephant family groups departing a water hole in a 2012 study. By this measure, nearly 90 percent of all recorded rumbles fell within extended group-level calling bouts, suggesting that elephant vocal activity in this context is dominated by sustained chorusing rather than neat, alternating exchanges.

Notably, the 124-second bout criterion and the 71-second quartile boundary for intermittent calls, though derived by entirely different methods, point to a similar transition zone between clustered and isolated vocal activity. The authors are careful to note that this agreement does not constitute independent validation, but it hints that some underlying temporal structure may be real even as the labels attached to it vary. Call duration, meanwhile, told its own story: rumbles ranged from just over one second to 12.44 seconds, but duration differed little across temporal categories. Instead, mixed-effects models revealed that individual identity — not timing — was the strongest driver of how long a rumble lasted, with individual identity explaining considerably more variance than recording session.

The broader implication is unsettling for the field. Terms such as chorusing, turn-taking, duetting, bouts and antiphonal calling are often used interchangeably in elephant research and across taxa, even though they capture partially overlapping aspects of vocal coordination. The 30-second response window introduced in 2003 was adopted by later studies without further testing, and the new results show that such fixed thresholds may not generalize across datasets. Because only a handful of studies have examined elephant vocal timing at all, early methodological choices risk having a disproportionate influence on how temporal coordination is defined and detected — potentially shaping the entire field’s understanding of coordinated vocal behaviour in one of Earth’s most socially complex species.

The authors argue that the way forward lies not in crowning a single universal method, but in combining complementary frameworks. Mixture models, successfully used to estimate bout criteria in humpback whale social calls, could test whether inter-call intervals form distinct temporal populations. Permutation procedures could assess whether observed response timing differs from chance expectations, while mixed-effects and hierarchical models could account for individual identity, dyadic relationships, recording session and context simultaneously. Social bonds matter here: previous work has shown that chimpanzees preferentially join the choruses of closely bonded males, and that bonobo turn-taking tracks relationship strength, so dyadic analyses may reveal patterns that group-level statistics obscure.

For now, the elephants of Bela Bela continue to rumble across their 300-hectare range, indifferent to the statistical debates their calls provoke. But the study’s message resonates far beyond one herd: when scientists ask whether animals converse, the answer may depend as much on the mathematics of the question as on the biology of the animals. In a field where a single threshold, borrowed and never revalidated, can determine whether a rumble counts as a reply or mere noise, methodological humility may be the most important instrument in the bioacoustician’s kit.

Subject of Research: Temporal coordination and antiphonal calling in African savanna elephant vocalizations

Article Title: Methodological frameworks influence the classification and interpretation of temporal coordination in African savanna elephant vocalizations

Article References: Joukhadar, J., Baotic, A., & Stoeger, A. S. (2026). Methodological frameworks influence the classification and interpretation of temporal coordination in African savanna elephant vocalizations. Frontiers in Mammal Science, 5, Article 1893031. https://doi.org/10.3389/fmamm.2026.1893031

Image Credits: AI Generated

DOI: 10.3389/fmamm.2026.1893031

Keywords: African savanna elephant, acoustic communication, antiphonal calling, vocal timing, bioacoustics, temporal coordination, turn-taking, vocal bouts, rumbles, methodology, animal behavior, Loxodonta africana

The study, published in the journal AI & Society by Yasmin A. Carruthers and Johannes M. Heim, compared podcasts generated by NotebookLM with naturally produced human podcasts on the same topics, examining two hallmarks of human conversation: the timing of turn-taking and the way speakers adopt contextually appropriate conversational roles. The researchers argue that naturalness in dialogue requires far more than getting the timings right; it demands a contextually appropriate, addressee-oriented approach that responds to how a conversation actually unfolds.

Human turn-taking is a marvel of predictive coordination. Decades of research, beginning with the landmark work of Sacks, Schegloff and Jefferson in 1974, have established that conversation is governed by an organized system in which only one person typically speaks at a time, transitions are smooth, and gaps or overlaps are minimized. Listeners exploit morphosyntactic, semantic and prosodic cues to anticipate when a speaker’s turn will end, preparing their responses while the previous speaker is still talking. Average turn transitions fall in the range of roughly 200 to 250 milliseconds, far shorter than the approximately 600 milliseconds humans need to plan and produce a spoken response. This means that human interlocutors must be predicting, not merely reacting, and traces of this ability appear remarkably early in development, with infants showing sensitivity to conversational turn structure from birth.

Yet the “no-gap, no-overlap” principle is not absolute. Subsequent research has shown that overlaps can account for a substantial share of speaker transitions, and that brief overlaps often signal engagement rather than disruption. Backchannels, the brief vocal acknowledgments such as “mm-hm” or “right” that listeners produce without claiming the floor, add a dynamic and expressive quality to human dialogue. In podcast settings, hosts and guests also follow genre-specific social scripts: hosts introduce topics and then step back into a novice role to give guests a platform, while guests predominantly adopt expert stances. These role expectations, rooted in the broadcasting heritage of podcasting, make conversational behavior highly predictable despite frequent shifts between expert and novice positions.

Against this backdrop, the researchers annotated approximately ten-minute excerpts from two human podcasts and two NotebookLM-generated podcasts built from their transcripts, using the acoustic analysis software Praat. They coded speaker contributions, role shifts, turn durations, overlaps and transitions, achieving very high inter-annotator agreement. The human podcasts came from a series on productivity, featuring a recurring host and invited guests, while the AI versions were generated by NotebookLM using Gemini-based script writing and Google’s own text-to-speech voices, with no additional training data supplied.

On the surface, the AI podcasts behaved similarly to their human counterparts: questions structured the conversation, topics were expanded through follow-ups, and both hosts and guests provided feedback through backchannels. A statistical model confirmed that guests, whether human or artificial, produced significantly more expert contributions than hosts, in line with the conventions of the podcast genre. But the deeper measurements told a different story. Human turns were dramatically longer, averaging 22.3 seconds for hosts and 49.4 seconds for guests, compared with just 11.3 and 14.8 seconds for the AI agents. Human distributions of turn length were skewed toward long, developed contributions, while AI turns were short and symmetrically distributed.

The most striking divergence appeared in overlapping speech. Human guests overlapped for an average of more than four seconds and human hosts for over two seconds, whereas AI overlaps lasted mere milliseconds. AI backchannels rarely overlapped at all; instead, the current speaker paused, waited for the backchannel to be uttered, and then resumed, producing a mechanical politeness that no human interlocutor would display. The researchers interpret human overlaps as signals of engagement and emotional alignment, made possible by the fact that human listeners continuously predict where turns will end. NotebookLM, by contrast, sidesteps the problem of endpoint prediction entirely by pre-planning the entire dialogue script before converting it to speech, strictly enforcing the avoidance of gaps and overlaps in a way that humans never do.

The study also tested a newer NotebookLM feature that allows a human to join the AI conversation live, and the results were revealing. Every human intervention disrupted the pre-planned script: transitions slowed dramatically, overlaps vanished, audio glitches caused sudden voice switches, and the agents could only acknowledge the human’s contribution after long, silence-based delays before returning to their prepared narrative. Because spontaneous intervention requires resource-intensive, real-time context integration that pre-planning was designed to avoid, the hybrid dialogues lost the fluidity that made the fully scripted versions sound convincing. The authors conclude that pre-determining role behavior and turn-taking only goes so far, and that NotebookLM clearly hits its limits when responding to unpredictable input.

Taken together, the findings suggest that AI-generated dialogue can mimic individual features of natural conversation in isolation but fails to synthesize them into the dynamic negotiation of information and social relation that defines human talk. The AI agents spoke shorter, overlapped far less, and behaved more like two equally informed participants than like a host giving a guest a platform, breaking with genre conventions that listeners intuitively expect. Human conversation, the authors argue, is not optimized for seamless information exchange but for building common ground, an emergent and collaborative process shaped by mutual attention, verbal feedback and shared, growing context. Variable turn-taking timings and adaptive conversational role behavior remain unmatched human skills, and achieving genuine naturalness will require AI systems that can interpret social contexts and respond dynamically to the fluid negotiation of interlocutor relations, not merely imitate their surface elements.

Subject of Research: Turn-taking timings and conversational role behavior in AI-generated podcast dialogues compared with natural human conversation

Article Title: Not your average podcaster: turn-taking timings and conversational role behavior in AI-generated dialogues

Article References: Carruthers, Y. A., & Heim, J. M. (2026). Not your average podcaster: turn-taking timings and conversational role behavior in AI-generated dialogues. AI & SOCIETY. https://doi.org/10.1007/s00146-026-03365-3

Image Credits: AI Generated

DOI: 10.1007/s00146-026-03365-3

Keywords: AI-generated podcasts, NotebookLM, turn-taking, conversational roles, dialogue generation, backchannels, overlaps, knowledge asymmetry, common ground, pragmatics, human-machine interaction, speech technology