Fig. 2

This figure illustrates three possible network configurations (b–d), when our algorithm detects a hybrid, h, which is neighbour of one of its parents, Nb(h), in the phylogenetic tree (a), e.g., in the IE language phylogeny inferred by Gray and Atkinson (see Fig. 1 in [6]). In configuration b, language h receives the proportion, α, of its lexicon from its closest ancestor in the tree via direct inheritance and the remaining part of its lexicon, (1-α), from a distant parent via word borrowing (e.g., see the case of Penn Dutch in Figs. 4 and 5b). In configuration c, language h is a lexical hybrid of Nb(h) and a distant parent (e.g., see the case of Sranan in Figs. 4 and 5b). In configuration d, language h receives the proportion α (indicated, in this case, in parentheses) of its lexicon from both its closest ancestor via direct inheritance and from its neighbour Nb(h) via word borrowing, and the remaining part, (1-α), of its lexicon from a distant parent via word borrowing (e.g., see the case of Old Armenian in Fig. 4)