Evidence of social learning across symbolic cultural barriers in sperm whales

Reviewer #2 (Public review):

Summary:

The current article presents a new type of analytical approach to the sequential organisation of whale song units.

Strengths:

The detailed description of the internal temporal structure of whale songs is something that has been thus far lacking.

Weaknesses:

The conceptual and terminological bases of the paper are problematical and hamper comparison with other taxa, including humans. According to signal theory, codas are indexical rather than symbolic. They signal an individual's group identity. Borrowing from humans and linguistics, coda inter-group variation represents a case of accents - phonologically different varieties of the same call - not dialects, confirming they are an index. This raises serious doubt about whether alleged "symbolism" and similarity between whale and human vocal behaviour is factual. The same applies to the difference between ICIs (inter-click interval) and IOIs (inter-onset interval). If the two are equivalent, variation in click duration needs to be shown so small that can be considered negligible. This raises serious doubt about whether the alleged variation in whale codas is indeed rhythmic in nature and prevents future efforts for comparison with the vocal capacities of other species. The scope and relevance of this paper for the broader field is limited.

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public review):

Summary:

This manuscript presents evidence of ’vocal style’ in sperm whale vocal clans. Vocal style was defined as specific patterns in the way that rhythmic codas were produced, providing a fine-scale means of comparing coda variations. Vocal style effectively distinguished clans similar to the way in which vocal repertoires are typically employed. For non-identity codas, vocal style was found to be more similar among clans with more geographic overlap. This suggests the presence of social transmission across sympatric clans while maintaining clan vocal identity.

Strengths:

This is a well-executed study that contributes exciting new insights into cultural vocal learning in sperm whales. The methodology is sound and appropriate for the research question, building on previous work and ground-truthing much of their theories. The use of the Dominica dataset to validate their method lends strength to the concept of vocal style and its application more broadly to the Pacific dataset. The results are framed well in the context of previous works and clearly explain what novel insights the results provide to the current understanding of sperm whale vocal clans. The discussion does an overall great job of outlining why horizontal social learning is the best explanation for the results found.

Weaknesses:

The primary issues with the manuscript are in the technical nature of the writing and a lack of clarity at times with certain terminology. For example, several tree figures are presented and ’distance’ between trees is key to the results, yet ’distance’ is not clearly defined in a way for someone unfamiliar with Markov chains to understand. However, these are issues that can easily be dealt with through minor revisions with a view towards making the manuscript more accessible to a general audience.

I also feel that the discussion could focus a bit more on the broader implications - specifically what the developed methods and results might imply about cultural transmission in other species. This is specifically mentioned in the abstract but not really delved into in detail during the discussion.

We are grateful for the Reviewer’s recognition of the study’s contributions to understanding cultural vocal learning in sperm whales. In response to the concerns regarding clarity and accessibility, we have revised the manuscript to improve the definition of key concepts, such as the notion of “distance” between subcoda trees. This adjustment ensures clarity for readers unfamiliar with the technical details of Markov chains. Additionally, we have expanded the discussion to highlight broader implications of our findings, particularly their relevance to understanding cultural transmission in other species, as suggested.

Reviewer #2 (Public review):

Summary:

The current article presents a new type of analytical approach to the sequential organisation of whale coda units.

Strengths:

The detailed description of the internal temporal structure of whale codas is something that has been thus far lacking.

Weaknesses:

It is unclear how the insight gained from these analyses differs or adds to the voluminous available literature on how codas varies between whale groups and populations. It provides new details, but what new aspects have been learned, or what features of variation seem to be only revealed by this new approach? The theoretical basis and concepts of the paper are problematical and indeed, hamper potentially the insights into whale communication that the methods could offer. Some aspects of the results are also overstated.

We appreciate the Reviewer’s acknowledgment of the novelty in describing the internal temporal structure of whale codas. Regarding the concern about the unique contributions of this approach, we have further emphasized in the revised manuscript how our methodology reveals previously uncharacterized dimensions of coda structure. Specifically, our work highlights how non-identity codas, which have received limited attention, play a significant role in inter-clan acoustic interactions. By leveraging Variable Length Markov Chains, we provide a nuanced understanding of coda subunits that complements existing studies and demonstrates the value of this analytical approach.

Reviewer #3 (Public review):

Summary:

The study presented by Leitao et al., represents an important advancement in comprehending the social learning processes of sperm whales across various communicative and socio-cultural contexts. The authors introduce the concept of ”vocal style” as an addition to the previously established notion of ”vocal repertoire,” thereby enhancing our understanding of sperm whale vocal identity.

Strengths:

A key finding of this research is the correlation between the similarity of clan vocal styles for non-ID codas and spatial overlap (while no change occurs for ID codas), suggesting that social learning plays a crucial role in shaping symbolic cultural boundaries among sperm whale populations. This work holds great appeal for researchers interested in animal cultures and communication. It is poised to attract a broad audience, including scholars studying animal communication and social learning processes across diverse species, particularly cetaceans.

Weaknesses:

In terms of terminology, while the authors use the term ”saying” to describe whale vocalizations, it may be more conservative to employ terms like ”vocalize” or ”whale speech” throughout the manuscript. This approach aligns with the distinction between human speech and other forms of animal communication, as outlined in prior research (Hockett, 1960; Cheney & Seyfarth, 1998; Hauser et al., 2002; Pinker & Jackendoff, 2005; Tomasello, 2010).

We thank the Reviewer for recognizing the importance of our findings and their appeal to broader audiences interested in animal cultures and communication. In response to the suggestion regarding terminology, we have adopted a more conservative language to align with distinctions between human and non-human communication systems. For example, terms like “vocalize” and “vocal repertoire” are used in place of anthropomorphic terms such as “saying”. This ensures consistency with established conventions while maintaining clarity for a broad readership.

Reviewer #1 (Recommendations):

Comment 1

Lines 11-13: As mentioned above, the implications for comparing communication systems and cultural transmission in other species isn’t really discussed much and I think it’s a really interesting component of the study’s broader implications.

Thank you for the comment.

Action - We added a few more sentences to the discussion regarding this.

Comment 2

Figure 1: More information on the figure of these trees would help. What do the connecting lines represent? What do the plain black dots and the black dot with the white dot represent? Especially since the ”distance between trees” is a key result, it’s important that someone unfamiliar with Markov chains can understand the basics of how this is calculated and what it represents. It is explained in the methods, but a brief explanation here would make the results and the figure a lot clearer since the methods are the last section of the manuscript.

These were omitted as we believed that attempting to introduce the mathematical structure and the methodology to compare two instances, in a figure caption, would have caused more ambiguity than necessary.

Action - Added an informal introduction to these concepts on the figure caption. Also added a pointer to the Supplementary Materials.

Comment 3

Table 1: A definition of dICIs should be included here.

Added the definition of discrete ICI to the table.

Comment 4

Figure 2: The placement of the figures is a bit confusing because they are quite far from the text that references them.

We thank the reviewer for pointing this out, we tried to edit the manuscript to improve this issue, but this part of the editing is more within the journal’s powers than our own.

Action - Moved images closes to the corresponding text in manuscript.

Comment 5

Line 117: Probabilistic distance needs to be briefly explained earlier when you first mention distance (see Lines 11-13 comments).

Action - Clarifications added in the caption of figure 1. as per comment on Lines 11-13

Comment 6

Figure 4: Is order considered in these pairwise comparisons? It looks like there are two dots for each pairwise comparison. Additionally, why is the overlap different in these two comparisons? For example, short:four-plus has an overlap of 0.6, while four-plus:short has an overlap of 0.95.

The x-axis of the plots in Figure 4 is geographical clan overlap. This is calculated as per (Hersh et al., 2022) and is described in our Methods (see “Measuring clan overlap” section). Given two clans—for example, the Four-Plus and the Short clan—spatial overlap is calculated twice: as the proportion of the Four-Plus clan’s repertoires that were recorded within 1,000 km of at least one of the Short clan’s repertoires, and as the proportion of the Short clan’s repertoires that were recorded within 1,000 km of at least one of the Four-Plus clan’s repertoires.

Order is important in these pairwise comparisons and generates an asymmetric matrix because the clans have different spatial extents. A clan found in only one small region might overlap completely with a clan that spans the Pacific Ocean, while the opposite is not true. For example, the Short clan spans the Pacific Ocean while the Four-Plus clan has been documented over a smaller area (but that smaller area overlaps extensively with the Short clan range). That is why the value is smaller (0.6) when considering how much of the Short clan’s range is shared with the Four-Plus clan, and larger ( 0.95) when considering how much of the Four-Plus clan’s range is shared with the Short clan.

Action - We have now added a reference to that section of the Methods in our Figure 4 caption and include the clan spatial overlap matrix as a supplemental table (Table S5).

Comment 7

Figure 4: I think the reference should be Hersh et al. [11].

Thank you for catching this.

Action - Reference corrected

Comment 8

Line 227: What aspect of your analysis looked at how often codas were produced? You mention coda frequency, but it is unclear how this was incorporated into your analysis. If this is included in the methods, the language is a bit too technical to easily parse it out.

Indeed here we are referencing the results of the paper mentioned in the previous line. We do not look at coda production frequency.

Action - Added citation to paper that actually performs this analysis.

Comment 9

Lines 253-255: I think you could dig into this a little more, as ”there is currently no evidence” is not the most convincing argument that something is not a driver. Perhaps expanding on the latter sentence that clans are recognizable across oceans basins would be helpful. Does this suggest that clans with similar geographic overlap experience diverse environmental conditions across ocean basins? If so, this might better strengthen your argument against environmental drivers.

Thank you for pointing this out. We feel that the next sentence highlights that clans are recognizable across environmental variation from one side to the other of the ocean basin, which supports the inductive reasoning that codas do not vary systematically with environment. However, we have edited these sentences for clarity.

Comment 10

Lines 311-314: It would also be interesting to look at vocal style across non-ID coda types. Are some more similar to each other across clans than others? Perhaps vocal style can further distinguish types of non-ID codas.

In supplementary Materials 3.4.2 and 3.5 we highlight our results when the codas are separated by coda type summarized in Table S4. We do compare the vocal style across non-ID coda types across clans and within the same clan. The results however are aggregated to highlight the differences in style between the clans and a a coda type-only comparison is not shown.

Comment 11

Lines 390-392: I’m assuming this is why pairwise comparisons were directional (i.e., there was both an A:B and a B:A comparison)? Can you speak to why A:B and B:A comparisons can have such different overlap values?

Given two clans—for example, the Four-Plus and the Short clan—spatial overlap is calculated twice: as the proportion of the Four-Plus clan’s repertoires that were recorded within 1,000 km of at least one of the Short clan’s repertoires, and as the proportion of the Short clan’s repertoires that were recorded within 1,000 km of at least one of the Four-Plus clan’s repertoires.

Order is important in these pairwise comparisons and generates an asymmetric matrix because the clans have different spatial extents. A clan found in only one small region might overlap completely with a clan that spans the Pacific Ocean, while the opposite is not true. For example, the Short clan spans the Pacific Ocean while the Four-Plus clan has been documented over a smaller area (but that smaller area overlaps extensively with the Short clan range). That is why the value is smaller (0.6) when considering how much of the Short clan’s range is shared with the Four-Plus clan, and larger (0.95) when considering how much of the Four-Plus clan’s range is shared with the Short clan.

Action - We now include the clan spatial overlap matrix as a supplemental table (Table S5).

Comment 13

Line 56: Can you briefly explain what memory means in the context of Markov chains?

We provide an explanation of the meaning of memory in the Methods section on ”Variable length Markov Chains”. Briefly, the memory in this case means how many states in the past of the Markov chain’s current state are required to predict the next transition of the chain itself. Standard Markov chains “look” back only one time step, while k-th order Markov chains look back k steps. In our case, there was no reason to assume that the memory required to predict different sequences of states (interclick intervals) should be the same across all sequences, and thus we adopted the formalism of variable length Markov chains, that allow for different levels of memory across the system.

Comment 14

Supplementary Figure S3: Like in the main manuscript, briefly explain or remind us what the blank nodes and the yellow nodes are.

Action - Clarified that the orange node represents the root of the tree in the figures.

Comment 15

Supplementary Figure S7: Put the letters before the dataset name.

Action - Done.

Comment 16

Supplementary Figure S10: Unclear what ’inner vs outer’ means.

One specifies comparisons across clans (outer) and the other within the same clan (inner)

Action - Added clarification on the caption of Figure S10

Comment 17

Supplementary Figure S14: Include a-c labels in the figure itself.

Action - Labels added to figure

Comment 18

Supplementary Figure S14: The information about the nodes is what needs to be included earlier and in the main body when discussing the trees.

Action - Added the explanation earlier in the text and in the main body

Reviewer #2 (Recommendations):

Comment 19

Line 22: ”Symbolic” and ”Arbitrary” are not synonyms. Please see the comment above.

We agree. Here, we make the point that the evolution of symbolic markers of group identity can be explained from what are initially arbitrary, and meaningless, signals (see [L1, L2]). Our point being that any vocalization, any coda, could have become selected for as an identity coda, and to become symbolic, and evolve to play a key role in cultural group formation and in-group favoritism because they enable a community of individuals to solve the problem of with whom to collaborate. The specific coda itself does not affect collaborative pay offs, but group specific differences in behavior can, as such the coda is arguably symbolic; as it is observable and recognizable, and can serve as a means for social assortment even when the behavioural differences are not. This can explain the means by which the social segregation which is observed among behaviorally distinct clans of sperm whales. However, in this manuscript, we do not extend this discussion of existing literature and have attempted to concisely describe this in a couple of lines, which clearly do a disservice to the large body of literature on the evolution of symbolic markers and human ethnic groups. We have added some citations to this section so that the reader may follow up should they disagree with out brief introductory statements.

Action - Added citations and pointers to the literature.

Comment 20

Line 24: The authors’ terminology around ”markers”, ”arbitrary”, ”symbolic” is unnecessarily confusing and mystifying, giving the impression these terms are interchangeable. They are not. These terms are an integral and long-established part of key definitions in signal theory. Term use should be followed accordingly. The observation that whale vocal signals vary per population does not necessarily mean that they function as a social tag. The word ”dog” varies per population but its use relates to an animal, not the population that utters the word. ”Dog” is not ”symbolic” of England, English-speaking populations or the English language. Furthermore, the function of whale vocal signals is extremely challenging to determine. In the best conditions, researchers can pin the signal’s context, this is distinct from signal’s function and further even for the signal’s meaning. How exactly the authors determine that whale vocal signals are arbitrary is, thus, perplexing given that this would require a detailed description and understanding of who is producing the song, when, towards whom, and how the receivers react, none of which the authors have and without which no claim on the signals’ function can be made. This terminological laxness and the sensu latu in extremis to various terms in an unjustified, unnecessary and unhelpful.

We use these terms as established in Hersh et al 2022 and the works leading up to it over the last 20 years in the study of sperm whales. These are often derived from definitions by Boyd and Richerson’s work on culture in humans and animals along with evolution of symbolic markers both in theory and in humans. We agree with the reviewer that these are difficult to establish in non-humans, whales or otherwise, but feel strongly that the accumulating evidence provides strong support for the function of these signals as symbolic markers of cultural groups, and that they likely evolved from initially arbitrary calls which were a part of the vocal repertoire (similar to the process and selective environment in Efferson et al. [L1] and McElreath et al. [L2]). We feel that we do not use these terms interchangeably here, and have inherited their use from definitions from anthropology. The work presented here uses terminology built across two decades of work in cetacean, and sperm whale, culture. And do not feel that these terms should be omitted here.

Comment 21

Lines 21-27: Overly broad and hazy paragraph.

We hope the replies above and our changes satisfy this comment and clarify the text.

Comment 22

Figure 1 legend: What are ”memory structures”? Unjustified descriptor.

The phrase was chosen to make draw some intuition on the variation of context length in variable length markov models.

Action - Re-worded from memory structures to statistical properties

Comment 23

Line 30: Omit ”finite”.

Action - Omitted.

Comment 24

Line 31: Please define and distinguish ”rhythm” and ”tempo”. Also see comment above, rhythm and tempo definitions require the use of IOIs.

We disagree with the reviewer’s claims here. In our research specifically, and for sperm whale research generally, coda inter-click intervals (ICIs) are calculated as the time between the start of the first click and the start of the subsequent click. This makes ICIs identical to inter-onset intervals (IOIs) under all definitions we are aware of. For example, Burchardt and Knornschild [L3] define IOIs as such: “In a sequence of acoustic signals, the time span between the start of an element and the next element, comprising the element duration and the following gap duration”. We now include a sentence making this point.

Regardless, we disagree on a more fundamental level with the statement that unless researchers quantify inter-onset intervals (IOIs), they cannot make any claims about rhythm. There are many studies that investigate rhythmic aspects of human and animal vocalizations without using IOIs [L4–L7]. If the duration of sound elements of interest is relatively constant (as is the case for sperm whale clicks), then rhythm analyses can still be meaningfully conducted on inter-call intervals (the silent intervals between calls).

For sperm whales, coda rhythm is defined by the relative ICIs standardized by their total duration. These can be clustered into discrete, defined rhythm types based on characteristic ICI patterns. Coda tempo is relative to the total duration of the coda itself. This can also be clustered into discrete tempo types across all coda durations as well (see [L8]).

Action - We added a sentence specifying that in this case we can use both ICIs and IOIs because of the standardized length of a single click.

Comment 25

Line 36: Are there non-vocalized codas to require the disambiguation here?

No, we have omitted for clarity.

Comment 26

Line 44: ”Higher” than which other social group class?

Sperm whales live in a multi-level social organization. Clans are a “higher” level of social organization than the social “units” which we define in line 40. Clans are made up of all units which share similar production repertoire of codas.

Action - We have added ’above social units’ on line 44 to make this clear.

Comment 27

Line 47: The use of “symbolic” continues to be enigmatic, even if authors are taking in this classification from other researchers. In signal theory (semiotics), not all biomarkers are necessarily symbols. I advise the authors to avoid the use of the term colloquially and instead adopt the definition used in the research field within which the study falls in.

There is ample examples of the use of ”symbolic” when referring to markers of in-group membership both in human and non-human cultures.Our choice to use the term “symbolic” is based on a previous study [L9] that found quantitative evidence that sperm whale identity codas function as symbolic markers of cultural identity, at least for Pacific Ocean clans. The full reasoning behind why the authors used the term “symbolic markers” is given in that paper, but briefly, they found evidence that identity coda usage becomes more distinct as clan overlap increases, while non-identity coda usage does not change. This matches theoretical and empirical work on human symbolic markers[L1, L2, L10, L11].

Action - We retain the use of the term here, as defined in the works cited, and based on its prior usage in the study of both human and non-human cultures.

Comment 28

Line 50: This statement is not technically accurate. The use of a signal as a marker by individuals can only be determined by how individuals ”interpret” and react to that signal - e.g., via playback experiments - it cannot be determined by how different populations use and produce the signals.

We respectfully disagree. While we agree that the optimal situation would be that of playback, the contextual use can provide insight into the functional use of signals; as can expected patterns of use and variation, as was tested in the papers we cite. However, this argument is not the scope nor the synthesis of this paper. These statements are supported by existing published works, as cited, and we encourage the reviewer to take exception with those papers.

Comment 29

Line 69: ”Meaningful speech characteristics”??? These terms do not logically or technically follow the previous statement. Why not stay faithful to the results and state that the method used seems to be valid and reliable because it confirms former studies and methods?

Action - Reworded to better underline the method’s results with previous studies

Comment 30

Lines 72-74: This statement doesn’t seem to accurately capture/explain/resume the difference between ID and non-ID codas.

We are not sure what the reviewer is referring to in this case. The sentence in this case was meant to explain the different relations that ID/non-ID codas have with clan sympatry.

Comment 31

Line 75: The information provided in the few previous sentences does not allow the reader to understand why these results support the notion that cultural transmission and social learning occurs between clans.

We conclude out introduction with a brief summary of our overall findings, which we then use the rest of the manuscript to support these statements.

Comment 32

Table 1: So far, the authors refer to their analyses as capturing the ”rhythm” of whale clicks. Consequently, it is not readily clear at this point why the authors rely on ”ICIs” (inter click intervals) instead of the ”universal” measure used across taxa to capture the rhythm of signal sequences - IOIs (inter onset intervals). If ICIs are the same measure as IOIs, why not use the common term, instead of creating a new term name? Alternatively, if ICIs are not equivalent to IOIs, then arguably the analyses do not capture the ”rhythm” of whale clicks, as claimed by the authors. Any rhythmic claim will need to be based on IOI measures. In animal behaviour, stereotyped is primarily used to describe pathological, dysfunctional behaviour. I suggest the use of other adjective, such as ”regular”, ”repetitive”, ”recurring”, ”predictable”. Another deviation from typical terminology: ”usage frequency” -¿ ”production rate”. Why is a clan a ”higher-order” level of social organization? This requires explanation, at least a mention, of what are the ”lower-order” levels. To the non-expert reader, there is a logical circularity/gap here: Clans are said to produce clan-specific codas, and then, it is said that codas are used to delineate clans. Either one deduces, or one infers, but not both. This raises the question, are clans confirmed by any other means than codas?

We are not creating a “new term name”: inter-click interval (ICI) is the standard terminology used in odontocete (toothed whale) research. We take the reviewer’s point that some readers will not be coming to our paper with that background, however, and now explicitly point out that ICI is synonymous with IOI for sperm whales. Please see our response to your earlier comment for more on this point.

Comment 33

Line 92: Unclear term, ”sub-sequence”. Fig. 1B doesn’t seem to readily help disambiguate the meaning of the term.

In fact reference to Fig. 1B is misplaced as it does not refer to the text. A sub-sequence is simply a contiguous subset of a coda, a subset of it.

Action - Removed ambiguous reference to Fig. 1B

Comment 34

Line 94: How does the use of ”sequence” compare here with ”sub-sequence” above?

In fact its the same situation although the previous comment highlighted a source of ambiguity.

Action - Reworded the sentence to be less confusing.

Comment 35

Line 95: Signal sequences don’t ”contain” memory, they require memory for processing.

Action - Rephrased from “sequences contain memory” to “states depend on previous sequences of varying length”.

Comment 36

Lines 95-97: The analogy with human language seems forced, combinatorics in any given species are expected to entail different transitions between unit/unit-sequences.

Thank you for the comment. Indeed, the purpose of the analogy is to illustrate how variable length Markov Chains work (which have been shown to be good at discerning even accents of the same language). We used human language as an analogy to provide the readers’ with a more intuitive understanding of the results.

Action - Revised paragraph to read: “Despite we do not have direct evidence of unitary blocks in sperm whale communication, on can imagine this effect similarly to what happens with words (e.g., a word beginning with “re” can continue in more ways than one starting with “zy”).”

Comment 37

Line 97: Unclear which possibility is this.

Action - Made the wording clearer.

Comment 38

Line 99: Invocation of memory, although common in the use of Markov chains, in inadequate here given that the research did not study how individuals perceived or processed click sequences, only how individual produced click sequences. If the authors are referring to the cognitive load imposed by producing clicks sequences, terms such as ”sequence planning” will be more accurate.

Here, we use the term “fixed-memory” in relation to the definition of a variable length Markov model. We feel that, in this section of the manuscript, the context is clear that it is a mathematical definition and in no way invokes the biological idea of memory or cognition. It is rather standard to use memory to describe the order of Markov chains. Swapping words in the definition of mathematical objects when the context is clear seems to cause unnecessary ambiguity.

Action - We clarified this in the manuscript (see comments above).

Reviewer #3 (Recommendations):

Comment 39

Line 16: Add ”broadly defined” as there are many other more restricted definitions (see for example Tomasello 1999; 2009). Tomasello M (1999) The cultural origins of human cognition. Harvard University Press, Cambridge Tomasello M (2009) The question of chimpanzee culture, plus postscript (chimpanzee culture 2009). In: Laland KN, Galef BG (eds) The question of animal culture. Harvard University Press, Cambridge, pp 198-221.

Thanks for the clarification.

Action - We added the term “broadly” and added the last reference.

Comment 40

Line 22: Is all stable social learned behavior that becomes idiosyncratic and ”distinguishable” considered symbolic markers? If not, consider adding ”potentially.”

No, but the evolution of cultural groups with differing behavior can reorganize the selective environment in such a way that it can favour an in-group bias that was not initially advantageous to individuals and lead to a preference towards others who share an overt symbolic marker that initially had no meaning and a random frequency in both populations. That is to say, even randomly assigned trivial groups can evolve arbitrary symbolic markers through in-group favouritism once behavioural differences exist even in the absence of any history of rivalry, conflict, or competition between groups. See for example [L1, L2].

Comment 41

Table 1: Identity codas are defined as a ”Subset of coda types most frequently used by a sperm whale clan; canonically used to define vocal clans.” Therefore, I infer that an identity coda is not exclusively used by a specific clan and may be utilized by other clans, albeit less frequently. If this is the case, what criteria determine the frequency of usage for a coda to be categorized as an identity or non-identity coda? Does the criteria used to differentiate between ID and non-ID codas reflect the observed differences in micro changes between the two and within clans?

The methods for this categorization are defined, discussed, and justified in previous work in [L9, L12]. We feel its outside the scope of this paper to review these details here in this manuscript. However, the differences between vocal styles discussed here and the frequency production repertoires which allow for the definition of identity codas are on different scales. The differences between identity and non-identity codas are not the observed differences in vocal style reported here.

Comment 42

Table 1: The definition of vocal style states that it ”Encodes the rhythmic variations within codas.” However, if rhythm changes, does the type of coda change as well? Typically, in musical terms, the component that maintains the structure of a rhythm is ”tempo,” not ”rhythm.” How much microvariation is acceptable to maintain the same rhythm, and when do these variations constitute a new rhythm?

Thank you for raising this important point about the relationship between rhythmic variations and coda categorization. In our definition, ”vocal style” refers to subtle, micro-level variations in the rhythmic structure of codas that do not alter their overarching categorical identity. These microvariations are akin to ”tempo” changes in musical terms, which can modify the expression of a rhythm without fundamentally altering its structure.

The threshold at which microvariations constitute a new rhythm, and thus a new coda type, remains an open question and is a limitation of current analytical approaches. In our study, we used established classification methods to group codas into types, treating variations within these groups as part of the same rhythm. Future work could refine these thresholds to better distinguish between meaningful rhythmic variation and the emergence of new coda types.

Comment 43

Table 1: Change ”say” to ”vocalize” (similarly as used in line 273 for humpback whales ”vocalizations”).

Thanks.

Action - Done.

Comment 44

Lines 33-35 and Figure 1-C: Can a lay listener discern the microvariations within each coda type by ear? Consider including sound samples of individual rhythmic microvariations for the same coda type pattern (e.g., Four plus, Palindrome, Plus One, Regular) to provide readers/listeners with an impression of their detectability. If authors considered too much or redundant Supplemental material at least give a sound sample for each the 4 subcodas modeled structures examples of 4R2 coda variations depicted in Figure 1-C so the reader can have an acoustic impression of them.

We do not think that human listeners would be able to all of the variation detected here. However, this does not mean that it is not important variation for the whales. Human observers being able to classify call variation aurally shouldn’t be seen as a bar representing important biological variation for non-human species, given that their hearing and vocal production systems have evolved independently. Importantly, ’Four Plus’,’Palindrome’, etc are names of Clans; sympatric, but socially segregated, communities of whale families, which share a distinct vocal dialect of coda types. These clans each have have distinguishable coda dialects made up of dozens of coda types (and delineated based on identity codas), these are not names/categorical coda types themselves.

Action - We now provide audio samples of all coda types listed in Figure 1B in the paper’s Github repository.

Comment 45

Line 69: As stated above, it may be confusing to refer to it as ”speech.” I suggest adding something like: ”Our method does capture one essential characteristic of human speech: phonology.” Reply 45.—Thank you for drawing our attention to this.

Action - We removed the word “speech” from the manuscript, using “communication” and/or “vocalization” depending on the context.

Comment 46

Line 111-112: Consider adding a sound sample of the variation of the 4R2 coda type that can be vocalized as BCC but also as CBB as supplementary data.

What the reviewer has correctly observed is that the traditional categorical coda type ’names’ do not capture the variation within a type by rhythm nor by tempo.

Action - We have added samples of all coda types listed in Figure 1B in the paper’s Github repo.

Comment 47

Figure 3: Include a sound sample for each of the 7 coda types in Figure 1B (”specific vocal repertoires”) to illustrate the set of coda types used and their associated usage frequencies, or at least for each of the 7 coda types in Figure 3 and tables S1 and S2.

Sperm whales in the Eastern Caribbean produce dozens of rhythm types across at least five categorical tempo types [L8, L13]. The coda types represented in Figure 1B do not demonstrate all the variability inherent in the sperm whales’ vocal dialect. Importantly, Figure 3, as well as table S1 and S2, refer to clan-level dialects not specific individual coda types.

Action - We added sound samples for each coda rhythm type listed in Figure 1B to the Github repository.

Comment 48

Lines 184-190: It is unclear what human analogy term is used for ID codas. This needs clarification.

We are not making an analogy in humans for the role of ID vs non-ID codas, but only providing the example of accents as changes in vocalization (style) without a change in the actual words used (repertoire).

Action - We tried to make it clearer in the manuscript.

Comment 49

Line 190: Change ”whale speech” to ”whale vocalizations.”

Thanks.

Action - Done.

Comment 50

Figure 4: Correct citation number Hersh ”10” to Hersh ”11.”

Thanks.

Action - Fixed the reference.

Comment 51

Lines 224-232: Clarify whether the reference to how spatial overlap affects the frequency of ID codas refers to shared ID codas between clans or the production frequency of each coda within the total repertoire of codas.

The similarity between ID coda repertoires we are referring to there is based on the ID codas of both clans.

More details on the comparison can be found in [L9].

Action - We added a sentence explaining the comparison is made using the joint set of ID codas.

Comment 52

Lines 240-241: What are non-ID codas vocal cues for?

Non-ID codas likely serve as flexible, context-dependent signals that facilitate group coordination, convey environmental or social context, and promote social learning, especially in mixed-clan or overlapping habitats. Their variability suggests multifunctional roles shaped by ecological and social pressures.

Comment 53

Lines 267-268: It’s unclear whether non-ID coda vocal styles are genetically inherited or not, as argued in lines 257-258.

We did not intend to argue that non-ID coda vocal styles are genetically inherited. Instead, we aimed to present a hypothetical consideration: if non-ID coda vocal styles were genetically inherited, one would expect a direct correlation between vocal style similarity and genetic relatedness. This hypothetical framework was introduced to strengthen our argument that the observed patterns are unlikely to be explained by genetic inheritance, as such correlations have not been observed. While we acknowledge that we lack definitive proof to rule out genetic influences entirely, the evidence available strongly suggests that social learning, rather than genetic transmission, is the more plausible mechanism.

Action - Clarified in manuscript.

Comment 54

Line 277: Can males mate with females from different clans?

Yes, genetic evidence shows that males may even switch ocean basins.

Action - We have clarified that we mean the female members of units from different clans have only rarely been observed to interact at sea between clans.

Comment 55

Lines 287-292: Consider discussing the difference between controlled/voluntary and automatic/involuntary imitation and their implications for cultural selection and social learning (see Heyes 2011; 2012). Heyes, C. (2011). Automatic imitation. Psychological bulletin, 137(3), 463. Heyes, C. (2012). What’s social about social learning?. Journal of comparative psychology, 126(2), 193.

Thank you for your insightful comment regarding this. The distinction between controlled/voluntary and automatic/involuntary imitation, as highlighted by Heyes [L14, L15], provides a potentially valuable framework for interpreting social learning mechanisms in sperm whales. Automatic imitation refers to reflexive, often unconscious mimicry driven by perceptual or motor coupling, while controlled imitation involves deliberate and goal-directed efforts to replicate behaviors. Both forms likely play complementary roles in the cultural transmission observed in sperm whales.

This dual-process perspective highlights the potential for cultural selection to act at different levels. Automatic imitation may drive convergence in shared environments, promoting acoustic homogeneity and facilitating inter-clan communication. In contrast, controlled imitation ensures the preservation of clan-specific vocal traditions, maintaining cultural diversity. This interplay between automatic and controlled processes could reflect a balancing act between cultural assimilation and differentiation, underscoring the adaptive value of these mechanisms in dynamic social and ecological contexts.

Action - We have incorporated a short discussion of this distinction and its implications for our findings in the Discussion. Additionally, we have cited [L14, L15] to provide theoretical grounding for this interpretation.

Comment 56

Methods: Consider integrating the paragraph from lines 319-321 into lines 28-35 and eliminate redundant information.

Thanks.

Action - We implemented the suggestion, removing the first paragraph of the Dataset description and integrating the information when we introduce the concepts of codas and clicks.

[L1] C. Efferson, R. Lalive, and E. Fehr, Science 321, 1844 (2008).

[L2] R. McElreath, R. Boyd, and P. Richerson, Curr. Anthropol. 44, 122 (2003).

[L3] L. S. Burchardt and M. Knornschild, PLoS Computational Biology 16, e1007755 (2020).

[L4] A. Ravignani and K. de Reus, Evolutionary Bioinformatics 15, 1176934318823558 (2019).

[L5] C. T. Kello, S. D. Bella, B. Med´ e, and R. Balasubramaniam, Journal of the Royal Society Interface 14, 20170231 (2017).

[L6] D. Gerhard, Canadian Acoustics 31, 22 (2003).

[L7] N. Mathevon, C. Casey, C. Reichmuth, and I. Charrier, Current Biology 27, 2352 (2017).

[L8] P. Sharma, S. Gero, R. Payne, D. F. Gruber, D. Rus, A. Torralba, and J. Andreas, Nature Communications 15, 3617 (2024).

[L9] T. A. Hersh, S. Gero, L. Rendell, M. Cantor, L. Weilgart, M. Amano, S. M. Dawson, E. Slooten, C. M. Johnson, I. Kerr, et al., Proc. Natl. Acad. Sci. 119, e2201692119 (2022).

[L10] R. Boyd and P. J. Richerson, Cult Anthropol 2, 65 (1987). [L11] E. Cohen, Curr. Anthropol. 53, 588 (2012).

[L12] T. A. Hersh, S. Gero, L. Rendell, and H. Whitehead, Methods Ecol. Evol. 12, 1668 (2021), ISSN 2041-210X, 2041-210X.

[L13] S. Gero, A. Bøttcher, H. Whitehead, and P. T. Madsen, R. Soc. Open Sci. 3, 160061 (2016).

[L14] C. Heyes, Psychological Bulletin 137, 463 (2011).

[L15] C. Heyes, Journal of Comparative Psychology 126, 193 (2012).