Sperm whales (Physeter macrocephalus) are unusual in that there is good evidence for sympatric populations with distinct culturally determined behaviour, including potential acoustic markers of the population division. In the Pacific, socially segregated, vocal clans with distinct dialects coexist; by contrast, geographical variation in vocal repertoire in the Atlantic has been attributed to drift. We examine networks of acoustic repertoire similarity and social interactions for 11 social units in the Eastern Caribbean. We find the presence of two socially segregated, sympatric vocal clans whose dialects differ significantly both in terms of categorical coda types produced by each clan (Mantel test between clans: matrix correlation = 0.256; p ≤ 0.001) and when using classification-free similarity which ignores defined types (Mantel test between clans: matrix correlation = 0.180; p ≤ 0.001). The more common of the two clans makes a characteristic 1 + 1 + 3 coda, while the other less often sighted clan makes predominantly regular codas. Units were only observed associating with other units within their vocal clan. This study demonstrates that sympatric vocal clans do exist in the Atlantic, that they define a higher order level of social organization as they do in the Pacific, and suggests that cultural identity at the clan level is probably important in this species worldwide.
The formation of social boundaries based on culture was critically important to the evolution of humans. Cultural boundaries are often the limits of cooperative and altruistic exchanges; among humans, language helped to identify these social boundaries in order to solve the dilemma of with whom to cooperate [1–4]. These variations in vocal repertoire marked cultural structures in human society which in turn affected reproduction, survival and selection, and paved the way for large-scale cooperative societies [5,6]
Among mammals, variation in vocal repertoire between sympatric and parapatric groups which could potentially interbreed has generally been referred to as dialects, while differences between populations separated by long distances which do not interbreed has been referred to as geographical variation . Geographical variation of vocal repertoires is widespread across mammalian taxa, including small terrestrial mammals , bats , monkeys (e.g. ), apes (e.g. ) and marine mammals (e.g. [12,13]). This type of variation is often the result of long-term isolation between populations and may be driven by genetic distinctiveness. Alternatively, if vocal learning does occur and cultural segregation mirrors geographical separation, vocal repertoires can diverge over time through spatial or demographic factors with copying errors and random cultural drift creating variation between isolated communities [14–16]. Once either of these occurs, the behavioural differences between the two communities can itself lead to genetic distinctiveness and reproductive isolation .
Sympatric dialects, however, appear to be quite rare and are thought to be the result of selection to advertise the distinction between groups. Along with high levels of philopatry and the consequent learning biases [18,19], vocal learning is thought to be the primary mechanism in the evolution and maintenance of dialects among local, interacting groups. While common in songbirds, vocal learning is rare among mammals. The cetaceans are an interesting exception with advanced vocal imitation and learning abilities [20–22].
The sperm whale (Physeter macrocephalus) provides an interesting case study for the evolution of culturally transmitted dialects as the pattern of variation in vocal repertoires differs between ocean basins. In both the Pacific and Atlantic Oceans, social units of female sperm whales produce repertoires of ‘codas’ [23,24], which are stereotyped patterns of three or more broadband clicks . In the Eastern Tropical Pacific, Rendell & Whitehead  described over 70 different coda types and used differing coda production repertoires to define distinct ‘vocal clans’ in which dozens of social units shared a similar coda repertoire. Units exclusively associate with units within their vocal clan, even though clan habitats are sympatric, suggesting that the dialects, and/or possibly other signals, function as markers of a shared cultural heritage and may act as barriers to cooperative and altruistic behaviour [5,6,23]. Recent work shows that acoustic variation among clans can neither easily be explained by genetic variation , nor is it the product of stochastic processes such as cultural drift . Modelling reveals that biases in vocal learning of coda types, specifically conformity (preferentially learning the most common coda types, ) and homophily (learning from behaviourally similar individuals due to social modularity, ), are required to generate the cultural segregation observed in the Pacific .
The current understanding of vocal variation among sperm whales in the North Atlantic Ocean is quite different. Coda repertoires in the Atlantic vary geographically and there is a significant negative correlation between repertoire similarity and spatial distance between populations . This finding suggests that cultural drift could have played a role in the evolution of divergent repertoires. Gero et al.  found no evidence of sympatric dialects in the Eastern Caribbean. However, one particular coda type (‘1 + 1 + 3’) was found to be produced with very high levels of stereotypy across the entire community, to have only ever been recorded in the Caribbean and to have remained the dominant coda type in the region for at least the last 30 years. The 1 + 1 + 3 coda's stability over this timeline and ubiquity across a population divided into disparate social units who range widely provides a rare example of cultural transmission maintaining high levels of conformity of a behaviour. This finding would indicate that biased vocal learning is also occuring in the Atlantic and that the conformity in the production of this coda type might function as a marker of clan membership in the Eastern Caribbean . Nonetheless, given the lack of sympatric dialects, it begs the question of what might be the function of this level of conformity—why signal clan membership if there is only one clan present, and is social assorting not necessary or possible?
A potential solution to this conundrum was suggested by the patterns observed in sympatric clans in the Pacific. Units from different clans used the same waters, but tended to appear in temporal waves, associating with other units of their own clan . Perhaps, some of the rarely identified Eastern Caribbean units, which were generally encountered apart from the common units, were members of one or more separate clans. During the longitudinal research project off the island nation of Dominica, we frequently encountered nine social units (units A, D, F, J, N, R, T, U and V from ; mean sightings days from 2005 to 2015: 89, range: 42–184) and these have been the subjects of intense social and acoustic analysis [24,31–40]. Here, we compare the vocal repertoires for two rarely seen units (units P and K in ) which have been identified only on 16 and 7 days, respectively, across the 10-year study, testing the hypothesis that they are from a separate clan or clans. We demonstrate that the two rare units produce a shared repertoire which is distinct from that of the nine common units and provide the first evidence of socially segregated, sympatric vocal clans in the Atlantic Ocean.
2. Material and methods
2.1. Field methods
Social units of female and immature sperm whales were located and followed in an area that covered approximately 2000 km2 along the entire western coast of the island of Dominica (15.30° N, 61.40° W). Unit membership was designated based on previous photo-identification analysis in which whales were unit members if they associated in sequential years and in which association was defined as identified within 2 h of each other . The multi-year requirement places emphasis on the long-term bonds among unit members and will also be conservative by including only those members for which long-term spatio-temporal coordination is evident . Research was conducted in the winters of 2005 through 2015 for a total of 3660 h with whales encountered on 402 days across 472 days of effort at sea (see electronic electronic supplementary material, table S1).
Acoustic recordings were made of sperm whale codas in two contexts: (i) at the onset of deep foraging dives and (ii) when whales were socializing in groups at the surface. Recordings were made with various systems that all had flat frequency responses across ranges of at least 2–20 kHz and sampling rates of 44.1 kHz or higher (see the electronic supplementary material for details). The variation in the recording systems did not affect our ability to record clear signals for both the coda and echolocation clicks produced by sperm whales, and as a result, the temporal patterning of clicks used in this analysis.
2.2.1. Measuring and testing of similarity between repertoires
To define the temporal structure of the codas recorded, we measured absolute inter-click intervals (ICI, the time between the onset of one click and the onset of the next click in the coda; see the electronic supplementary material for methodological details). We calculated the absolute ICI measures for recordings from units P and K and used the existing dataset of absolute ICI measurements from the nine common units collected by Gero et al.  using identical methods. To quantify similarity between unit repertoires, we used two different approaches: one categorical and one classification-free continuous measure as in previous work [24,41].
In the categorical approach, two codas were given a similarity of 1 if they were assigned to the same type and were given a similarity of 0 if they were assigned to different types. We assigned codas of similar click length to a categorical type using a hierarchical clustering algorithm called OPTICS  run on the absolute ICI measures for all 11 units combined. We used a ξ-value of 0.04 for all coda lengths (which defines a 4% drop in point density as the criterion for defining a new cluster) as this best defined the clusters visually evident in plots of the first two components of a principal components analysis (PCA) run on the same data. OPTICS allows for points that are outliers or located in sparse areas between dense clusters to be labelled as noise, rather than being forced into defined clusters. All codas which were not classified into clusters and labelled as noise by OPTICS were omitted from the categorical analysis but retained in the classification-free, continuous measure of similarity described below. We used the OPTICSxi module in the ELKI framework (http://elki.dbs.ifi.lmu.de/, ) to run these analyses.
Coda types were given names based on the mean temporal click pattern for all codas included in that cluster, following previous nomenclature [23,24,39,44]. For example, a ‘5R’ coda is one in which five clicks are regularly spaced, while a ‘1 + 1 + 3’ coda sounds like ‘click-[PAUSE]-click-[PAUSE]-click-click-click’ with longer gaps between the first two clicks followed by three clicks in quick succession. We refer to coda types that make up greater than 10% of a unit's coda production as ‘predominant’ coda types.
For the classification-free, continuous measure, the multivariate similarity of two codas with the same number of clicks was measured as a Euclidean distance between the ICI vectors using absolute ICI measures for all 11 units combined using methods as in previous work . Similarities were calculated using custom-written routines in MATLAB 7.12 (The Mathworks, Inc., MA, USA).
Matrix correlations and Mantel tests with 10 000 permutations [45,46] were used to examine repertoire variation between units. We tested whether pairwise similarities between repertoires were higher for repertoires of the same unit on different days (same unit, different day—SUDD) compared with those from different units (different unit, different days—DUDD). Each day's recordings were treated as independent in an attempt to account for any autocorrelation in coda production within a recording day. To do so, we tested the matrix of pairwise similarities of each day's recordings against a 0/1 matrix with 1 coding for SUDD and 0 coding for DUDD. If units produced distinct repertoires, then the expectation is a significantly positive correlation between these matrices. This Mantel procedure was repeated with a 0/1 matrix of the same clan/different clan in which the expected alternative hypothesis is a significant positive correlation between matrices if clan repertoires differ. All matrix correlations and Mantel tests were carried out using SOCPROG 2.5  in MATLAB 7.12 (The Mathworks, Inc., MA, USA).
We then used similarity matrices between repertoires to construct average-linkage clustering dendrograms and measured their robustness using 1000 bootstrap replicates. The cophenetic correlation coefficient (CCC) was also calculated to indicate how well the dendrogram represented the data. A CCC of over 0.8 is considered a ‘good’ representation of the associations .
To visualize social and acoustic similarity of the units among clans, we plotted a social network of associations among individual members of the 11 social units overlaid with an acoustic network among units. The edges of the social network were weighted using a half-weight index (HWI) of association . We defined an association between whales when they were photographically identified within a cluster at the surface (40 m chain rule) within a 2 h sampling interval using photo-identification data from 2005 to 2015, as in previous work . The acoustic network was built based on coda repertoire similarity and weighted by the measures of multivariate similarity between unit repertoires, as above. The social network was plotted using Gephi 0.8.2beta (https://www.gephi.org) and a Force Atlas 2 layout algorithm (http://bit.ly/1kmVfe5). This is a force-directed layout in which nodes repel each other, while edges attract nodes they connect. A buffer was given to prevent overlap of different units in order to allow for overlay of larger nodes and the acoustic network.
We identified 388 codas from 21 recordings on 3 days in 2 years for unit P and 426 codas from 19 recordings on two different days in 1 year for unit K. These were added to the 4116 codas assigned to the nine common units during previous analyses , for a total of 4930 codas in the complete dataset for all 11 units. A total of 324 codas (6.6%) were omitted as noise by OPTICS. The remaining 4606 codas were classified into 22 different coda types (see the electronic supplementary material, figure S1 for a plot illustrating rhythm of all coda types). Codas of five clicks in length made up 70.9% of all codas recorded. The unit repertoires were dominated by three prevalent coda types which each made up greater than 10% of all codas recorded: ‘1 + 1 + 3’ (32.7%), ‘5R1’ (21.9%) and ‘5R3’ (12.4%). Mantel tests confirmed that recordings of the same unit on different days are more similar than recordings of different units on different days (table 1). While discovery curves are not asymptotic, they do suggest that repertoires of all units were adequately sampled given that all units have more than approximately 250 codas (electronic supplementary material, figure S2).
|multivariate similarity||categorical similarity|
|social level||within||between||matrix correlation||p-values||within||between||matrix correlation||p-values|
The 11 social units were divided into two evident clades by a single bifurcation at the root of the hierarchical clustering dendrogram (figure 1). Units P and K were distinguished from the nine common units by distinctly clustering in 100% of the 1000 bootstrap replicates giving confidence that the division is not dependent on sampling. Units P and K share a distinct repertoire from that of the nine commonly observed units, as indicated by both categorical and continuous measures (figure 1 and table 1). Vocal repertoires between the two clades differed significantly based on Mantel tests (table 1). Therefore, using the same justification as in the Pacific , we will divide the social units observed off Dominica into the common EC1 vocal clan and the rare EC2 vocal clan.
The most obvious distinction between the two vocal clans is the production of different predominant coda types. In the case of the EC1 clan, the 5R1 and 1 + 1 + 3 codas are predominant types, while the EC2 clan never produced the 5R1 and only unit P produced the 1 + 1 + 3 very infrequently (figure 1; electronic supplementary material, table S2). Instead, the 5R3 coda dominates the repertoires of the EC2 clan and is only produced infrequently by three of the nine common units and in the case of unit P, the 6R coda is also a prevalent type (only two common units infrequently produce the 6R coda). These differences produce the strong division in the structure of the dendrogram in figure 1. Furthermore, the EC2 clan differ consistently from the more common EC1 clan in their production of the non-predominant types. The EC2 clan produced other ‘regular’ class coda types with more clicks (7R, 9R and 10R highlighted in the red dashed boxes in figure 1) which are long in duration (approx. 1.5 s; rhythms plotted in electronic supplementary material, figure S1) and have constant ICIs throughout the coda, while the EC1 clan units produced ‘increasing’ class codas (6I, 7I, 8I, 9I and 10I highlighted in green dashed boxes in figure 1) which are short in duration (less than 0.5 s; rhythms plotted in electronic supplementary material, figure S1) and have increasing ICIs throughout the coda. These distinctions between the two vocal clans are highlighted using plots of the first two principal components by coda type and unit in the electronic supplementary material, figure S3.
The social network in figure 2 demonstrates the social segregation between clans as there are many associations among units within the common EC1 vocal clan (green circles) but a complete absence between vocal clans. The only social connection between unit P and the EC1 clan is via a mature male who was observed consorting with two units in different clans. The overlaid acoustic network highlights the dissimilarity between clan repertoires and reflects the division in the dendrogram from figure 1.
This study supports the hypothesis generated by the social and cultural patterns observed in the Pacific: in the North Atlantic, as in the Pacific, there are sympatric clans of social units of female sperm whales with distinctive repertoires, who tend to appear in a study area in temporal waves. This finding refutes the contention that repertoire diversity in the Atlantic Ocean is a simple geographical variation caused by cultural drift (genetic drift is an unlikely cause of the dialects as nuclear genes show little geographical structure, ). Modelling by Cantor et al.  suggests that two factors are needed in order to partition Pacific sperm whale units into vocally marked cultural groups, conformism and homophily. Previous research of the Eastern Caribbean sperm whale dialects and sociality has found evidence of both conformism to prevalent coda types  and high social modularity . Taken together, these findings demonstrate that social boundaries are structured along cultural lines and support the contention that cultural identity is important in this species in both the North Atlantic and the Pacific.
While our discovery of sympatric clans in the waters off Dominica narrows the gap between our understanding of the socio-cultural systems of sperm whales in the North Atlantic and the Eastern Tropical Pacific, there are still substantial contrasts. In the Pacific, sympatric coda repertoires are evident from each analysis of the vocalizations of female sperm whales based on samples spanning more than a few months in a particular study area ([23,51]; M. Cantor 2015, unpublished data). However, it took many years of data and a determination to analyse the repertoires of the rarer social units before sympatric clans were identified in the Caribbean. While samples for the EC2 clan units are small in terms of number of days on which recordings were made, it is unlikely that additional sampling would yield different results given the clear division between clans maintained by bootstrapping, as well as the different prevalent codas types between the clans which are produced in the majority of one clan's recordings and virtually none in that of the other clan (see the electronic supplementary material for details on sampling and discovery curves). Nonetheless, the image that appears is of social units from one clan, EC1, predominating in our study area and neighbouring waters, with members of the EC2 clan, whose core range is elsewhere, making occasional incursions. The relatively high resighting rates and residency times of the EC1 member units off islands in the Eastern Caribbean , taken with the fact that no whales in this study have been identified outside of the Caribbean Sea despite longstanding effort in the neighbouring waters of the Bahamas, Gulf of Mexico, and the western North Atlantic [31,35], suggest that EC1 member units' home ranges, as well as their wider movements, are probably smaller than those described for the Pacific [23,52,53]. Thus, as suggested in the work of Antunes , Atlantic clans may overlap in range much less than in the Pacific; nonetheless, we have demonstrated that there are areas in which they are sympatric. Furthermore, the repertoire similarity between clans in the Pacific is an order of magnitude lower than the one between the EC1 and EC2 clans (mean between clan similarity this study: 0.002, s.e. = 0.0001 versus 0.0004, s.e < 0.0001 between sympatric clans in the Pacific; Pacific values re-calculated using Euclidean distance and absolute ICIs, as in this study, from data reported in ). While it is likely that the absolute similarity values are less meaningful when compared across populations based on distinct datasets, than relative ones within a given area's dataset, it appears as though the repertoires of sympatric Pacific clans are more distinct than those of sympatric clans in the Caribbean. Variation in the degree of sympatry probably affects the variation in repertoire similarity by altering the selective forces acting to distinguish repertoires among clans. Certainly, this is the case among humans, in which the correlation between symbolic markers and behaviour is the strongest when cultures are close in space ; and birds, in which songs in sympatry are more divergent than those in allopatry (e.g. ). With increased habitat overlap, one would expect more divergent coda repertoires, should they function in cultural recognition and identity. Our results then follow from the social identity hypothesis proposed by Rendell & Whitehead  for the function of coda dialects among Pacific clans. Furthermore, this study answers the question of why there exists such high levels of conformity in the production of the 1 + 1 + 3 coda by members of several disparate social units raised by Gero et al. . With the presence of two sympatric clans, highly conserved, clan-characteristic codas (such as the 1 + 1 + 3 coda in the case of the EC1 clan) may function as a marker of cultural identity.
Whitehead & Gero  proposed that the Eastern Caribbean, some of the most urban habitat for this species in the North Atlantic Ocean, is an attractive sink (or ecological trap, [56,57]) such that immigration from surrounding waters probably explains the conflicting results of high levels of mortality in this community but an increasing population trend based on mark–recapture data. This may support the scenario outlined in the previous paragraph, whereby the increasing incursions by members of a peripheral EC2 clan would impact mark–recapture estimates, but it would not affect the measures of mortality, recruitment and fecundity which are predominantly based on individual-level long-term study of the most common units in the EC1 clan. Alternatively, units P and K may be recent immigrants from a clan elsewhere in the Atlantic where resources are not as abundant. Electronic supplementary material, figure S4 contrasts the Caribbean coda types with those of neighbouring waters, including the Gulf of Mexico, Azores and the Sargasso Sea from the results of Antunes . It suggests that the vocal repertoire of the EC2 vocal clan may share some similarities with the repertoires in the Sargasso Sea and the Azores in which long ‘Regular’ class codas also predominate. Further quantitative analysis will be required to conclusively match the EC2 clan's repertoire to a specific region in the North Atlantic or determine whether it is a distinct repertoire from those which have been previously documented.
As in the Pacific, units in the Eastern Caribbean seem to be socially segregated based on vocal clan. Identifiable vocal dialects create population structure in this species as they do among killer whales (Orcinus orca, ), birds (summarized in ) and humans. Among human cultural groups, a socially learned vocal marker, language, helps to solve the dilemma of with whom to cooperate by reliably identifying those who share similar behaviours [1,2]. When in-group favouritism of this kind occurs, it can dramatically decrease within-group variation and accordingly increase between-group heterogeneity of other behaviours . Previous work has demonstrated this to be the case among the sympatric clans in the Pacific, which differ in their movement patterns, habitat use and foraging success , as well as diet , reproductive success  and social behaviour . Killer whale ecotypes also show differences across a myriad of behaviours including social interactions, diet and foraging behaviour, and movement patterns and diving behaviour (summarized in ). This study opens the door to studies on the cultural content of the two clans in the Caribbean.
Overall, it would appear that sperm whales in the Eastern Caribbean live in a much more individualized society than their counterparts in the Eastern Tropical Pacific. Social relationships between unit members are dynamic , long-term relationships between units may be based on social preferences through direct personal knowledge  and, at the highest level, sympatric cultures are segregated based on coda dialect. By contrast, the society in the Pacific appears to be more structured at the cultural level, in an ‘Us versus Them’ fashion, such that units associate with other units which share vocal dialect but do not appear to show strong preferences at lower levels of social organization. This may be the result of a society devastated by whaling, such that the social structure in the Eastern Caribbean reflects what might have existed prior to whaling [53,63]. These differences could also be a response to differing environments in which broad recognition of clan membership is important in the Pacific to facilitate frequent group formation in a high predator risk environment in which individual units may range more widely and are not able to form preferred associations between units . Nonetheless, this study demonstrates that sympatric vocal clans do exist in the Atlantic, that they define a higher order level of social organization as they do in the Pacific, and suggests that cultural identity at the clan level is probably important in this species worldwide.
Our data were collected in Dominica under scientific research permits from the Fisheries Division of the Ministry of Agriculture and Fisheries: SCR 013/05-02, RP-2/12 IW-1, RP-09/014 IW-1, RP-01/079 W-2, RP-03/059 W-4, P-122/4 W-2 and P-40/2 W-7. Over the course of the long-term study, the field protocols for approaching, photographing and recording sperm whales were approved by the University Committee on Laboratory Animals of Dalhousie University and the Animal Welfare and Ethics Committee of the University of St Andrews.
The ICI and association data used in this study are available from Dryad Digital Repository (http://dx.doi.org/10.5061/dryad.53g73).
S.G. collaborated in the conception and design of the study, funded and coordinated the field operations of the study, undertook all of the fieldwork to collect data, completed the acoustic and statistical analysis and wrote the manuscript; A.B. collaborated in the conception and design of the study, collaborated in the acoustic and statistical analysis and edited the manuscript; H.W. collaborated on the design of the analysis, funded the fieldwork, participated in the collection of the field data and edited the manuscript; P.T.M. collaborated in the conception and design of the study, participated in the collection of the field data, funded the fieldwork and edited the manuscript. All authors gave final approval for publication.
We have no competing interests.
S.G. was supported by an NSERC Postgraduate Scholarship (PGS-M), an NSERC Canadian Graduate Scholarship (CGS-D), the Izaak Killam Memorial Scholarship, the Patrick F. Lett Fund, the Dalhousie's Presidents Award, the Marine Alliance for Science and Technology for Scotland (MASTs) pooling initiative, the Carlsberg Foundation and an FNU fellowship for the Danish Council for Independent Research from the Ministry of Higher Education and Science supplemented by a Sapere Aude Research Talent Award. P.T.M. contributed funding from an FNU large frame grant. H.W. was funded through operating and equipment grants from the Natural Sciences and Engineering Research Council of Canada (NSERC) and the Whale and Dolphin Conservation Society.
We thank Chief Fisheries Officer Mr Sebastien and the Dominica Fisheries Division, as well as the staff at Al Dive, Anchorage Hotel, Dive Dominica and W.E.T. Dominica for logistical support while in Dominica; all the crews of the whale watch vessels and all of R/V Balaena's crew members. Some of these data were collected using LOGGER software and analyses used Rainbow Click software, both developed by the IFAW to promote benign and non-invasive research. This manuscript emanates from The Dominica Sperm Whale Project: http://www.thespermwhaleproject.org follow: @DomWhale.
© 2016 The Authors. Published by the Royal Society under the terms of the Creative Commons Attribution License http://creativecommons.org/licenses/by/4.0/, which permits unrestricted use, provided the original author and source are credited.
Nettle D. 1999Language variation and the evolution of societies. In The evolution of culture (eds Dunbar RIM, Knight C, Power C), pp. 214–227. Piscataway, NJ: Rutgers University Press. Google Scholar
Van den Berghe PL. 1981The ethnic phenomenon. New York, NY: Elsevier. Google Scholar
Barth F. 1969Introduction. In Ethnic groups and boundaries: the social organization of cultural difference (ed. Barth F), pp. 9–38. Boston, MA: Little, Brown and Company. Google Scholar
Payne RS, Guinee LN. 1983Humpback whale, Megaptera novaeangliae, songs as indicators of ‘stocks’. In Communication and behaviour of whales (ed. Payne RS), pp. 333–358. Boulder, CO: Westview Press. Google Scholar
Boyd R, Richerson PJ. 1985Culture and the evolutionary process. Chicago, IL: Macmillan Publishers Limited. Google Scholar
Antunes R. 2009Variation in sperm whale (Physeter macrocephalus) coda vocalization and social structure in the North Atlantic Ocean. PhD thesis, University of St Andrews, St Andrews, UK. Google Scholar
Gero S, Gordon JCD, Carlson C, Evans P, Whitehead H. 2007Population estimate and inter-island movement of sperm whales, Physeter macrocephalus, in the Eastern Caribbean. J. Cetacean Res. Manag. 9, 143–150. Google Scholar
Bridge PD. 1993Classification. In Biological data analysis (ed. Fry JC), pp. 219–242. Oxford, UK: Oxford University Press. Google Scholar
Torres AMA. 2006Distincion de clanes de cachalote (Physeter macrocephalus) en el Golfo de California, mediante comparacion de repertorios de codas. Master's thesis, CICIMAR, Instituto Politécnico Nacional, Mexico. Google Scholar
Catchpole CK, Slater PBJ. 2013Bird song: biological themes and variations. Cambridge, UK: Cambridge University Press. Google Scholar
Tønnessen JN, Johnsen AO. 1982The history of modern whaling. Berkeley, CA: University of California Press. Google Scholar