iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
A study published October 1, 2026, in Scientific Reports found that machine-learning models could predict the broad social context of African savanna elephant rumbles better than chance, whether the rumble was given alone or as part of an overlapping chorus. The same study found that some calls became more acoustically similar as chorused exchanges unfolded. The models did not translate elephant speech, and the authors’ suggested link between that convergence and group coordination remains a hypothesis.
What the study examined
The paper, “Social context information and acoustic convergence in elephant vocal choruses,” was written by L. Mahon, M.A. Pardo, M. Miron and colleagues and appears as Scientific Reports 16, article 27961 (2026). It studied wild African savanna elephants (Loxodonta africana) in Amboseli National Park, Kenya. The recordings came from several periods: 1986–1990, 1998–2009, 2020 and 2022. The final dataset contained 902 rumble recordings, including both single calls and chorused calls. A Nature Portfolio press release dated October 2, 2026 describes the sample as 84 elephants from 39 families.
Most earlier acoustic work on elephant rumbles focused on individual calls that did not overlap with other animals’ calls. Real social exchanges often involve several elephants vocalizing at once, and overlapping signals are harder to separate. The authors therefore analyzed chorus recordings as overlapping multi-caller signals rather than first isolating every caller. That choice is what allowed them to ask whether a chorus keeps the context information found in a single rumble.
The six social contexts
The study sorted rumbles into six labeled social contexts. These labels are the authors’ analytical categories. They describe the situation in which each call was given; they are not dictionary translations of the calls.
#1 Best Overall
| Context label | Situation described in the study | Acoustic convergence reported in chorused exchanges |
|---|---|---|
| Coo | Females calling to calves | Not found in the analyses reported; small sample sizes noted by the authors |
| Contact | Calls between elephants that are separated | Found |
| Let’s-go | Calls associated with attempts to start group movement | Not found in the analyses reported; small sample sizes noted by the authors |
| Cadenced | Rumbles given during feeding, resting or slow movement | Found |
| Little-greeting | Calls between closely bonded elephants | Not found in the analyses reported |
| Greeting | Calls during reunions | Found |
How well the models performed
The researchers trained classifiers on acoustic properties of the calls and asked them to predict the social context. The models performed above chance, but they were far from reliable, and they confused some contexts with one another. The figures below are classifier metrics (weighted average precision and recall, within-family tests). They describe how often the model’s predictions matched labeled contexts in a given test setup. They are not the share of elephant communication that was correctly translated.
| Test setup as reported | Weighted average precision | Weighted average recall |
|---|---|---|
| Single-rumble dataset | 0.47 | 0.44 |
| Chorus dataset | 0.37 | 0.36 |
| Other family-level data | Lower values reported; exact figures not stated in the summaries available | Lower values reported; exact figures not stated in the summaries available |
The most informative result is not the absolute score but the link between the two datasets. Models trained on single rumbles could predict the context of chorused calls above chance. That suggests single and chorused calls share context-linked acoustic features, so a chorus is not simply noise layered over unrelated signals.
Rank #2
Acoustic convergence in choruses
Convergence here means that the acoustic features of calls in an exchange became more similar as the exchange went on. The study reported this pattern in cadenced, contact and greeting exchanges. It did not report it for coo, let’s-go or little-greeting calls. For coo and let’s-go, the authors caution that the absence of a finding should be read carefully because the samples were small. The pattern therefore applies to three of the six contexts in the analyses reported, not to chorusing in general.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Why convergence might matter, and what is still untested
The authors propose that convergence in some exchanges may relate to logistical coordination. Cadenced and contact exchanges can involve several callers over extended interactions, and matching calls could help a group keep track of one another. This is a proposal for further investigation. The study did not include playback experiments, in which researchers play calls to elephants and observe their responses. Without such tests, the study cannot show that convergence helps elephants agree on a plan or move together.
Rank #3
What the researchers say about the approach
Co-author Sara Keen, identified as a behavioral ecologist at Earth Species Project, told Scientific American‘s Jackie Flynn Mogensen, as reported by Sarah Kuta in Smithsonian Magazine on October 8, 2026: “it is pretty exciting that we’re able to train a model to look at an entire chorus and have some interpretation of exactly what’s going on.” The paper’s own summary frames the conclusion this way: “elephant choruses function as group signals encoding context through both their composition and patterns of acoustic convergence.” Both statements describe an analytical approach and a set of findings, not a full translation.
Method notes for readers who want to check the work
- Equipment changed over time. The methods list reel-to-reel and digital recorders and Earthworks microphones across the recording periods. These describe how the historical and newer recordings were made, not recommendations for field gear.
- Audio processing. The authors pitch-shifted audio for analysis and standardized files before analysis. The shifted frequencies are an analysis step; they do not mean the elephants vocalized at those frequencies.
- Data and code. The paper states that acoustic features, metadata and analysis code are available in an Earth Species Project repository named “Elephant-choruses-context-convergence.”
- Permissions and funding. The paper says Kenya’s relevant authorities and Amboseli National Park granted permissions. Funding came from the National Geographic Society, Tennessee Elephant Sanctuary, Oakland Zoo, named trusts and foundations, and anonymous donors.
- Affiliations. The authors are affiliated with Earth Species Project, ElephantVoices, Amboseli Trust for Elephants and Colorado State University.
A figure to treat with caution
The Nature Portfolio press release says that “around 60% of rumbles occur in choruses involving two or more elephants.” The summary of the paper available for this article does not identify the source or method behind that population-level percentage. Treat it as the publisher’s approximation rather than a measured estimate, and do not quote it as an independently verified statistic.
Rank #4
What the findings do and do not show
- They show that rumbles from single calls and chorused calls carry information about social context that a classifier can detect above chance, in one well-studied Amboseli population.
- They show that some exchanges, in three of the six labeled contexts, become more acoustically similar over their course.
- They do not show that a particular call means a particular word or sentence, or that a person can now hold a conversation with elephants.
- They do not show that elephants use convergence to coordinate movement. That idea needs playback experiments and further study.
- They do not show that the models are dependable for individual calls. Their performance was modest and confused some contexts.
For readers, the useful takeaway is that “decode” in this context means finding patterns that predict a situation. It is a meaningful step in animal acoustics, but it is not a language translator.
Quick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
