Following once again the view of a language as a system that interacts with its environment, the concept of linguistic Phylogenetics is not ungrounded within the general phylogenetic framework 8. Indeed in the last 15 years there has been a steady increase of papers where language development and biological speciation have been treated as quite similar [320]. Pagel in his eponymous review paper “Human language as a culturally transmitted replicator” [230] not only argues on the similarity of genes’ and languages’ evolutionary behaviour but offers an extensive catalog of analogies between biological and linguistic evolution as well.
Interestingly one might even argue that linguistic phylogenetic studies pre- ceded biological ones at least in theWestern world. While Aristotle (382-322 BC) was probably one of the first to cluster different animal species in terms of com- parative methods [10], Socrates (469-399 BC), among other philosophers, actually realized that language “changed” (or at least “decayed”) as time passed [49]. The reason for this observation was relatively simple: Homer’s (∼8th century BC) writ- ings while revered as accounts of heroic tradition, they were already at least 350 years old at the time of Socrates. People simply realized that Achilles did not speak like them. One of the first to formulate an actual “connection” though between Linguistics and Biology was Gottfried W. Leibniz (1646-1716)9. Leibniz advocated
8
See section 3.5 for a short introduction in Phylogenetics.
9The reader will note a significant chronological gap. Aside the obvious need for significant
the idea ofNatura non facit saltus(“nature does not make jumps”), gradual change. In addition to that he also advocated the ideas of Monadology: fundamental imma- terial units that are eternal were “the grounds of all corporeal phenomena” [195]. Those ideas proved fundamental both in Biology and Linguistics. Therefore it is not surprising that the father of modern Biology, Charles R. Darwin (1809-1882) also made similar assertions regarding language in his landmark workThe Origins of Species. Leaving historical remarks aside, the seminal paper of Cavalli-Sforza et al. [52] changed the way linguistic and genetic information are combined within a single analysis framework in modern times. There the authors focused on the reconstruction of a human phylogeny based on maximum parsimony 10 principles but importantly, after pooling genetic data geographically in order to account for heterogeneity, if heterogeneity persisted, they added an “ethnolinguistic criterion of classification”. That allowed the synchronous derivation of a genetic and a linguistic phylogeny that displayed significant overlap and emphasized that the two fields not only could share methods but also results.
Up until relatively recently maximum parsimony trees [321; 107] and com- parative methods[320] stood as the state-of-the-art in Linguistic Phylogenetics. And while comparative methods were already employed rather broadly within the con- text of glottochronology [105] the question of computational reconstruction of pro- tolanguages started to emerge [227; 268]. Importantly people began to incorporate the phonetic principals in their tree reconstructions. Research came to the realiza- tion that exactly because language was just a human-bounded characteristic, direct analogies with generic Phylogenetics were not only possible, but actually strength- ening the theoretical framework used. Language acquisition being associated with children (founder effects), parallel development of characteristics being not as un- common as originally thought (convergent evolution), insertion-deletion-reversals being usual “units of changes” (the same operations being used in the changes of genetic code) showed that even qualitative linguistic phenomena could be en- capsulated within a phylogenetic framework. Evidently the inherent problems of Phylogenetics such as having (2N−3)!!11 rooted-trees for N leaves and being pre- sented with a relatively small amount of data compared to the number of candidate trees [138] did remain, but linguists were nevertheless able to validate a very cru- cial insight from sociolinguistics: “while most linguistic structures can be borrowed between closely related dialects, natively acquired sound systems and inflections are resistant to change later in life” [268]. That meant that essentially a sound system was resistant to change and therefore presented a “good” character for phylogenetic
the Biblical story of the Tower of Babel made the question of linguistic phylogenies somewhat heretic.
10See Sect. 3.5.2 for an overview of tree reconstruction methodologies. 11
The double or odd factorial where (2k−1)!! =(22kkk)!! or more generally: (2k−1)!! = Π
k
studies. Insights like the positive correlation between rates of change and speaker population size [12] and the coherence of rule-based changes [221; 39] were estab- lished. Importantly almost all these techniques rely on binary features [73] or at a best case scenario multi-state ones [231; 106]. While computational linguists rec- ognized the importance of phoneme sequences [39; 38] they do not act on premises of continuous data. As it will be shown in following chapters, the current work is not qualitatively comparable with state-of-the-art multi-state implementations [38] where the number of available training data is significantly larger. In addition, even excluding training sample size issues, the current methodology also acts almost ag- nostically in relation with semantic information by only using phonetic information. Undoubtedly the choice of discarding semantic information is a strong (yet not un- common [105; 231]) assumption from a linguistic point of view but it presents itself as a definite progress in the phonetic literature because up until now comparative methods focused on scalar characteristics only.
As a final note we draw attention to the notion of themolecular clock of a biological phylogeny [165] and its significance in a linguistic phylogeny. As Gray et al. note, absolute dates in Linguistics are notoriously hard to get [106]. Disregarding the issues that relate to tree uncertainty and lack of concrete evidence, this difficulty is rooted with the quite restrictive assumption of (in this case) lexical clock (or a
glottoclock). Exactly because one asserts that changes “occur in a more-or-less clocklike fashion, so that divergence between sequences should be proportional to the evolutionary time between the two sequences” [165], this assumption is hard to evaluate experimentally [105; 76]. Nevertheless we need this assumption for standardMaximum Likelihood methodology to be applicable. It would be therefore interesting to explore a possible application of methodologies that act under the assumption of a non-universal time- continuum. The current work does explore this in terms of the observational time on the phylogeny’s leaves but leaves the question of actual phylogenetic time (and its potential time-distortion) for future work.