Unity and disunity in evolutionary sciences: process-based analogies open common research avenues for biology and linguistics

Abstract

Background: For a long time biologists and linguists have been noticing surprising similarities between the evolution of life forms and languages. Most of the proposed analogies have been rejected. Some, however, have persisted, and some even turned out to be fruitful, inspiring the transfer of methods and models between biology and linguistics up to today. Most proposed analogies were based on a comparison of the research objects rather than the processes that shaped their evolution. Focusing on process-based analogies, however, has the advantage of minimizing the risk of overstating similarities, while at the same time reflecting the common strategy to use processes to explain the evolution of complexity in both fields.

Results: We compared important evolutionary processes in biology and linguistics and identified processes specific to only one of the two disciplines as well as processes which seem to be analogous, potentially reflecting core evolutionary processes. These new process-based analogies support novel methodological transfer, expanding the application range of biological methods to the field of historical linguistics. We illustrate this by showing (i) how methods dealing with incomplete lineage sorting offer an introgression-free framework to analyze highly mosaic word distributions across languages; (ii) how sequence similarity networks can be used to identify composite and borrowed words across different languages; (iii) how research on partial homology can inspire new methods and models in both fields; and (iv) how constructive neutral evolution provides an original framework for analyzing convergent evolution in languages resulting from common descent (Sapir's drift).

Conclusions: Apart from new analogies between evolutionary processes, we also identified processes which are specific to either biology or linguistics. This shows that general evolution cannot be studied from within one discipline alone. In order to get a full picture of evolution, biologists and linguists need to complement their studies, trying to identify cross-disciplinary and discipline-specific evolutionary processes. The fact that we found many process-based analogies favoring transfer from biology to linguistics further shows that certain biological methods and models have a broader scope than previously recognized. This opens fruitful paths for collaboration between the two disciplines.

Reviewers: This article was reviewed by W. Ford Doolittle and Eugene V. Koonin.

Keywords: Constructive neutral evolution; Incomplete lineage sorting; Language evolution; Lateral transfer; Process-based analogies; Protein assembly; Similarity networks; Word formation.

Fig. 1

Timeline of early tree- and network diagrams in linguistics (top) and biology (bottom). Schottel’s branching table of Germanic languages from 1663 is the earliest we could identify. The three following early diagrams in linguistics by Stiernhielm (1671) [7], Hickes (1689), [9], and Gallet (1800) [8] all contain reticulation, real trees only start with Ćelakovský and Schleicher (1853) [4, 15]. The situation is similar in biology, where the two schemas by Leclerc De Buffon (1755) [12] and Rühling (1774) [13] allow for reticulation, in contrast to Lamarck (1809) [17] and Darwin (1837, 1859) [5, 16]

Fig. 2

Contrasting purely linguistic, purely biological, and analogous processes in linguistics and biology. For Process-Based Analogies, we contrast the biological term with the linguistic term, if both disciplines address the processes in their terminology. See the text for further clarification

Fig. 3

Sequence and word similarity networks. a In sequence similarity networks, sequences and similarities between sequences are represented in a network. Sequences are represented as nodes, and similarities between sequences are represented as edges if they exceed a certain threshold. Since evolutionary processes leave certain traces in the topology of these networks, they can be identified by applying standard network techniques. b Since words can be modeled as sequences of sounds, it is straightforward to create networks which represent the similarity among words. Due to the peculiarities of language evolution, however, similarity measures need to be specifically adapted to linguistic needs. As in biology, linguists start from alignments, as illustrated for words meaning ‘sun’ in five Germanic languages, but specific scoring functions are used

Fig. 4

Polymorphisms in language evolution. a Synonymy: languages have many nearly synonymous words (German Hals and Nacken both mean ‘neck’ in English). They can be interchangeably used to express one and the same concept. Near synonymy is often resolved by dropping one of the two words. b Analogy: languages with complex morphology (case systems, etc.) often have irregular paradigms which consist of different stems (like good, better, best in English). These paradigms are often resolved retaining only one form and adapting the other forms to this model (e.g., good, gooder, goodest). b Derivation: words can be slightly modified by adding affixes (word derivation) or merging to words with each other (compounding). Often, both the modified or merged forms can still be interchangeably used with the original forms. They can also replace the original forms. d Incomplete lineage sorting: if rapid divergence occurs before the polymorphisms are resolved, they may yield patterns that seem to be in contradiction with tree-like divergence

Fig. 5

Similarity networks applied to linguistic data. a Similarity networks are reconstructed from global alignments for words meaning ‘person’ in Germanic, Romance, and Slavic languages (data taken from [101]). Five large connected components are identified. While three of them are homogeneous regarding the language family and show true cognate sets common in the respective branch of Indo-European, the top-left cluster contains words from all three branches. This cluster mainly shows Romance reflexes of Latin persona ‘person’. Slavic and Germanic words occurring in this cluster are all borrowed. b Similarity networks are reconstructed from local alignments for dialect words meaning ‘face’ in 20 Chinese dialect varieties (data taken from [132]). The data contains three variants, two simple words liǎn and mián, two words of different origin, and one fused form liǎn-mián. Numbers in the alignment reflect tone patterns, which are characteristic for South-East Asian languages. Edges colored in black differ in their local and global alignments, edges colored in gray show identical alignments for local and global analyses. The fused form serves as a hub connecting the two components. Data and code to reproduce the networks is available from the data and material accompanying this article (Additional file 1)

Fig. 6

Word formation processes in the German language. Word derivation and word compounding are the major processes underlying word formation. Word derivation involves the combination of bound morphemes (suffixes, prefixes, and infixes) with free morphemes (regular words of a language). The graphic shows how the German Krankheitsverlauf ‘disease progression’ has been created in multiple stages by which the adjective krank ‘sick’ was nominalized with help of the suffix -heit and later compounded with the nominal form of the complex verb verlaufen ‘to progress’. Note that free morphemes may easily turn into suffixes during language change