Why the old yes-or-no label stopped fitting
The older approach separated vocal usage learning from vocal production learning: usage learning changes when or in what context a sound is produced, while production learning changes the sound itself [2]. That distinction was useful, but later work treated vocal production learning as a continuum from limited to more complex learners, and accumulating data exposed the limits of those early frameworks [2]. The revision is therefore less about adding a new checklist and more about replacing a single label with a set of dimensions.
In the newer classification, one dimension is copying an auditory model—how closely a sound converges on or diverges from a model, and how accurately it is copied. Other dimensions cover the type and breadth of change and the timing: when learning occurs, how long it takes, and how long the result is retained [2]. A mammal can therefore be a strong copier on one dimension and weak on another, which makes 'capable' a matter of degree rather than a yes-or-no verdict.
Which mammals the broader view brings into the conversation
The strongest cases remain the ones where animals copy artificial or human language sounds, or match acoustic models of different sound types. That evidence is clearest in cetaceans, pinnipeds, elephants and bats [1]. These are the species that would comfortably pass a demanding copying test, and they anchor the upper end of the spectrum.
A wider circle of mammals shows vocal convergence, where adjustments within one sound type make individuals sound more alike. This has been reported in primates, mole-rats, goats and mice [1]. The important qualification is that convergence is not proof of production learning: the same pattern could arise from production learning, from usage learning, or from individuals matching physiological states, and the underlying mechanisms are currently unclear [1]. Under revised criteria these species are candidates to examine, not confirmed production learners.
What a mammal now has to show to count
The revised frameworks raise the bar for evidence even as they widen the categories. For experimental studies, learned sounds need quantitative comparison with the animal's vocal repertoire before learning began, or with the species repertoire, to confirm that the sound is genuinely novel [1]. Without that baseline, an apparent imitation may simply be an existing call used in a new situation.
The multi-dimensional framework also asks researchers to rule out vocal changes that happen without learning, and it treats usage learning as its own framework sensitive to context and timing [2]. So the change in criteria cuts both ways: more species can be placed somewhere on the map, but each placement requires showing which dimension is involved rather than assuming that any vocal change is production learning.
Candidate species and the limits of the new labels
One proposed screening approach looks at acoustic allometry—the usual relationship between body size and the frequencies an animal produces. Species in vocal-production-learning lineages deviate from that scaling, and the study suggests that outliers lying above the regression line may have potential vocal production learning abilities [3]. This is presented as a hypothesis-generating signal, not a demonstration that those species actually learn sounds.
Separately, comparisons across 555 species link losses of olfactory receptor genes and expanded vocal-learning brain regions in five lineages—Haplorrhini, Chiroptera, Pinnipedia, Sirenia and Cetacea—where non-terrestrial foraging and vocal learning evolved independently [4]. That work concerns evolutionary trade-offs between smell and vocal communication, not the criteria for classifying learning, so it does not by itself move any species across a threshold. The practical upshot is that revised criteria have made the category more informative but also more provisional: a species can be a strong copying candidate, a convergence candidate, or a statistical candidate, and those are different claims [1][2][3].
Sources used in this answer
Vocal production learning in mammals revisited
The clearest evidence for vocal production learning in mammals comes from cetaceans, pinnipeds, elephants and bats, while vocal convergence appears in primates, mole-rats, goats and mice but may reflect production learning, usage learning or physiological state matching.
The multi-dimensional nature of vocal learning
Building on the distinction between vocal usage and production learning, the authors propose a multi-dimensional classification of vocal production learning covering copying accuracy, degree of change and timing, rather than treating it as a binary trait.
A cross-species framework to identify vocal learning abilities in mammals
Species in vocal-production-learning clades deviate from acoustic allometry, and outliers above the regression line are hypothesized as species with potential vocal production learning abilities.
Convergent Evolutionary Trade-Offs between Olfaction and Vocal Learning in Mammals
Across 555 species, five lineages with independently evolved vocal learning and non-terrestrial foraging show negative shifts in olfactory receptor quantity and diversity, alongside weak associations with lateral cerebellar expansion.
Vocal learning and development in the grey seal, Halichoerus grypus
Grey seals exhibit vocal production learning, and the paper notes that strict documentation criteria and inherent difficulties in demonstrating learning may explain the scarcity of evidence across nonhuman mammals.