Multi-criteria group decision making (MCGDM) deals with the process of making decision among a set ofdecision makers who evaluate alternatives over several criteria. MCGDM problems evolve in tandem with theprogress of society and thus, their complexity is growing. Such complexity has given rise to the large-scalegroup decision making (LS-GDM) problems in which hundreds of decision makers may participate in thedecision process and there are challenges to face such as groups formation and polarization opinions. In anyMCGDM problem, the elicitation of decision makers' preferences is a key task, because the solution of theproblem is determined by this information. Pairwise comparison is a widely used technique for this task but,a large number of comparisons might lead to erroneous solutions, since the consistency of the decision makers'preferences could be a ected. The best-worst method (BWM) was proposed in order to reduce the numberof comparisons and consequently, the apparition of inconsistency in decision makers' opinions. However,this proposal deals with numerical assessments, which are not enough to model uncertainty that commonlyappears in MCGDM problems. To face the latter limitation, an extension to the fuzzy environment of theBWM was proposed, taking advantage of the use of linguistic information and of its well performance inuncertainty modelling. Nevertheless, the latter proposal represented the results by means of triangular fuzzymembership functions, which are hard to understand from decision makers' point of view. Therefore, in thisstudy, we extend the classic BWM in the 2-tuple linguistic environment to model uncertainty associated withthe pairwise judgments/comparisons via fuzzy linguistic terms and to enhance the accuracy of computationover linguistic terms and interpretability of the results. Moreover, we apply our proposal to LS-GDMscenarios in which polarization opinions and sub-groups identi cation that, so far, have not been addressedfrom any of BWM proposals, are taken into account.