The fishing village of Siribinha (11º48'49"S, 37º36'38"W) is part of the municipality of Conde and is located in the Itapicuru estuary in Bahia, in the northeast of Brazil . The area consists of freshwater alluvial wetlands, mangroves, beach vegetation, and shrubby thicket-like forests (locally known as restingas) growing on sand dunes. Coconut plantations and cattle ranches also make up part of the land use tenure of the region.
Siribinha is a community of artisanal fishers comprising ca. 500 inhabitants. The community was relatively isolated up to the 1990s, since prior to that there were no roads connecting it to nearby villages and cities. Despite the emergence of small-scale tourism starting from the mid-1990s, Siribinha is still predominantly a fishing community, where fishing and shellfish gathering constitute the main economic activity of the majority of the community members.
Artisanal fishing in the north coast of Bahia is characterized by family work, where members of the family are variously involved in the activity of catching and processing the catch, especially shellfish. In Siribinha, as well as in other Brazilian fishing communities, fishing is typically a male activity, while shellfishing, which comprises the activity of gathering mollusks and crustaceans, is carried out primarily by women and children .
Most of the information provided on the community results from our own interview data and participant observation in the larger project in which the present study is included.
Throughout the paper, we indicate the community members by the initial letter of their names followed by a dot and their age (e.g. E.68), for confidentiality reasons. The Portuguese transcripts were translated by the first author and the translation was revised by the other authors. In the quotes from community members’ interviews, pauses are indicated with a slash (/), and the end of a speech turn, with a period (.). The transcripts are shown in italics and, if we need to comment or add something, this is done using parentheses, without italics. For each transcript included in the paper, we provide the Portuguese original excerpts in the Supplementary Material 1. To carry out the research reported here, we combined two methods: free listing and triad tasks.
For our specific purposes of understanding fish ethnotaxonomy, we performed a free listing task to determine the most salient species of fish, that is, the fish that the community members consider to be the most important. Free listing interviews were performed either in their houses, during door-to-door visits, or in the shared social spaces in the village. Most interviews happened in their free time, i.e., during the day when they were at home or sitting on their porches, but some of them were also done when they were repairing their nets or landing fish. Their consent to be interviewed was obtained in audio recordings after stating the terms of an Informed Consent Form.
The free listing was carried out in April 2018 with 91 community members (approx. 20% of the community), comprising a total of 49 men and 42 women, all aged 18-89 years old. The interviewees consisted of 38 fishermen and 25 fisherwomen (shellfish gatherers), but also another 11 male and 17 female community members dedicated to other activities (students, teachers, traders, accommodation providers, among others). We considered fishermen or fisherwomen those who engaged in fishing/shellfish gathering activities ≥ 3 days a week or, in the case of retirees, if they had engaged in these activities at such an intensity before retiring.
Each participant was asked “What fish do you know?”. We then let the interviewees speak freely so as not to influence their train of thought, and we noted the cited fish in the order they were mentioned. In instances where interviewees cited ethnogenera of fish such as arraia (stingray), bagre (catfish), cação (small shark), pescada (hake), and robalo (snook), we sought further clarification by asking later if there is more than one type of each of them. If that was the case, these would be annotated just after each ethnogenus mentioned.
To understand how members of the community categorize fish and to what extent the categories are shared across the community, we carried out triad tasks (or triad tests) [24, 14]. The triad task allows us to derive a consensual cultural model without assuming that such a model exists beforehand . For this purpose, we randomly selected 45 people that took part in the free list task (15 fishermen, 15 shellfish gatherers/fisherwomen, and 15 other community members) and solicited their participation on the triad tasks. The triad tasks were conducted between October 2018 and January 2019.
During the triad task, a series of 10 sets of three photographs (a triad) of fish were presented to each participant. They were then asked which ethnospecies was the most different among the three shown in the photographs. It was then assumed that the two other ethnospecies were being considered more similar to each other by the interviewee. If the participant had difficulty with a specific triad, that triad was postponed until the end of the task. If the participant still could not provide an answer after this second round of questioning, he or she was asked whether the difficulty of making the requested judgment was due to the ethnospecies being very similar or very different. For each attempt, participants could choose an item, therefore, as “different” (codes 1–3), “very different” (code 0) or “very similar” (code 4). We followed this procedure to avoid situations where participants felt forced to produce an answer, thereby choosing items randomly and biasing the data .
The triad task was performed with a Lambda 2 design [24, 25], which means that each pair of ethnospecies was compared exactly twice. Using 10 ethnospecies, this design generates 30 triads, a number that was deemed to be manageable for the triad tasks, without taking too much of interviewees’ time or tiring them out and thereby compromising data quality.
To generate the triad tasks, we randomly selected ten ethnospecies of fish (Table 1) among the 33 most salient ones to the community, according to the findings from the free listing interviews (see later).
Ethnospecies of fish selected for the triad tasks (in alphabetical order) in Siribinha, northeast Brazil.
Academic scientific species (family)
Bagre bagre (Ariidae)
Eugerres brasilianus (Gerreidae)
Scomberomorus cavala (Scombridae)
Micropogonias furnieri (Sciaenidae)
Mugil liza (Mugilidae)
Cynoscion acoupa (Sciaenidae)
Cynoscion leiarchus (Sciaenidae)
Centropomus parallelus (Centropomidae)
Mugil curema (Mugilidae)
Caranx hippos (Carangidae)
The photographs of fish were taken with the same camera approximation to ensure that fish body size proportions were maintained. All the photographs used in the triad task are presented in the Supplementary Table S1.
To obtain an idea of how the participants classified the fish they saw in the photographs, we also asked them at the end of the triad tasks which criteria they would use to differentiate one fish from the others.
During the free listing tasks, we observed that some interviewees provided two or more different names or synonyms for some fish (28 of the cited ethnospecies corresponded to 13 academic scientific species). Small phonetic differences were common, like arraia jamanta and arraia jalamanta (Mobula sp.) or bagre upemba and bagre urupemba (unidentified species). In some cases, different names were also cited for the same fish species at different ontogenetic phases, like saúna (smaller) and tainha (bigger) (Mugil curema) or pescada amarela (smaller) and pescada selvagem (bigger) (Cynoscion acoupa). Therefore, when running the analyses, we selected the most frequently used name by the community for each fish to which there were synonyms, in order to avoid artificially inflating the number of fish mentioned, thus biasing the salience estimation. The academic scientific species names for the ethnospecies and general fish types were provided by a fish specialist, José Amorim dos Reis Filho, on the basis of the ethnospecies’ names. A rarefied ethnospecies-interviewee curve (Supplementary Figure S1) indicated that the number of participants engaged to generate our free lists was adequate, as the number of ethnospecies captured approached an asymptote.
To determine whether the gender and activity of the interviewees had any influence on the number of ethnospecies (or general fish types) cited, we performed two-way ANOVAs with gender (male, female) and activity (fishers, other activities) as factors (α = 0.05).
We calculated the Salience Index (S) of each fish ethnospecies following Chaves et al. , using the formula: S=Σ((L–Rj + 1)/L))/N, where L is the length of a list, Rj is the rank of item j in the list, and N is the number of lists in the sample. The index takes into account not only the frequency of occurrence of each item, but also the order in which they were mentioned in the interviews . In cases where an interviewee mentioned a general type, for instance, cação (small shark), as the first item but clarified more specific ethnospecies later, we substituted the general fish type for the mentioned ethnospecies. However, in some cases in which the interviewee did not provide any ethnospecies for a general fish type when questioned at the end of interview, we maintained the general fish type as he or she listed.
The Salience Index of each ethnospecies cited is calculated by the probability of occurrence of these values in a null scenario , and varies between 0 and 1, which denote ethnospecies with extremely low or high salience, respectively. The p-values of salience show the probability that the salient values occur in a null scenario, calculated from simulated populations with similar characteristics to the real one, using Monte Carlo techniques . Following Chaves et al. , we accepted a threshold p-value < 0.05 to denote significance. Using this methodology, it is possible to establish a cut-off point in a free list and select only the most salient items, i.e., the ones showing high salience index and p-value < 0.05.
To visualize how fish knowledge varies between interviewees (i.e. how their knowledge composition varies), we used a non-metric multidimensional ordination to summarise all the ethnospecies citations by the interviewees. The fish citations were coded as presence-absence data and used to compute a Bray-Curtis dissimilarity matrix. In the ordination graph, interviewees were represented using different markers to denote different genders and activities [male - fisher, male - other (non-fisher), female - fisher, female - other (non-fisher)]. A permutational multivariate analysis of variation (PERMANOVA; α = 0.05) was run to determine if the composition of knowledge differed between the four groups. A Bonferroni correction was applied in order to reduce the chances of obtaining false-positive results.
To analyze the triad task data, we used the Anthropac 4.98 software . Anthropac generates an “aggregate proximity matrix”, which is a similarity matrix showing the percentage of times each pair of ethnospecies were considered more similar within a triad (agreement = matched responses/30). Using these similarity matrices generated by Anthropac, we performed a non-metric multidimensional scaling ordination to determine how the participants were categorizing the fish and to visualize the degree of similarity between them according to the participants.
For each triad, Anthropac also provides the agreement rate, which is the proportion of triads in which each participant agreed with the modal response (i.e. that pair of fish in which most participants considered the most similar within each triad). Using a principal components analysis (PCA) on the interviewee vs. interviewee matrix also enabled us to determine the presence of a cultural consensus model . Widely shared information would be reflected in a “cultural consensus” or high agreement among individuals. To achieve this, each interviewee’s fish–distance matrix was correlated with that of every other interviewee, yielding a 45 x 45 matrix in which entries correspond to the observed agreement among interviewees on pairwise fish distances. A PCA was then performed on the inter-participant correlation matrix. Following Romney et al. , we deemed that a strong group consensus existed if (1) the ratio first and second factor eigenvalue was high, (2) the first eigenvalue accounted for a large portion of the variance, and (3) all individual first-factor scores were positive and relatively high. If these criteria are met, the structure of the agreement can be explained by a single factor solution, namely, the consensual model; otherwise, we may reach reliable conclusions about inter-participant differences .
We also carried out PERMANOVAs on the interviewee fish-distance matrices to determine if gender, activity and their interactions had any bearing on fish ethnotaxonomic classification. To determine if the agreement rate of participants with the modal response was related to gender, age and activity, we fitted a generalized linear model of the participants’ modal response (dependent variable) with their respective ages, genders, and activities as independent variables, using α < 0.05. All analyses were performed in R (R Core Team 2017). The R script for calculating Salience indices was provided by L. Chaves upon request and ordinations and PERMANOVAs were performed using the adonis function in the vegan package.