A Lexical Family Tree

Etymologies, such as those provided by a Google search, sometimes provide a tree-like depiction of the various historical antecedents of a word with the headword as the root. What a tree for the entire lexicon look like?

The Topology of Etymology

Without actually constructing such a network, we can describe its topology.

Imagine a network where each node is a “definition” of a word. If a word A is one of the most recent ancestor of B, A points to B. For example, OFr. “hercier” would point to to “hearse” and “rehearsal.” This relation is irreflexive, as a word cannot descend from itself. Further we may forbid loops, as a word may not descend from one of its descendants. (n.b This implies that the relation is asymmetric) Thus, we can succinctly say that

Def. An etymological network on a corpus C is a directed acyclic graph on a superset of C with the binary relation “”

Because it seems likely that there were fewer words in the ancient languages, I’m guessing that this graph will look vaguely like a polytree, with its roots being roots in the appropriate proto-language. (For English, mostly Proto-Indo-European) The graph will probably have a maximum path length of less than ten (average maximal path length probably around five) judging my my experience with the “number of steps” in most etymologies, although I would be interested to see what the maximal paths looks like.

I’m also interested in the connectedness of the graph. If there were two disconnected components, for example, this would tell us that there are two families of words which share no common roots or intermediaries. There are like two lexical families. It’s hard for me to say whether such lexical isolates exist. If there are minimally connected families (for example, families which share a single intermediary ancestor) what words are the bridges? These are some of the topological questions that would be answered by the construction of such a network.

Practical Challenges

While the necessary information is publicly available, it would be prohibitively tedious to compile it by hand. Probably better would would be to scrape it from some standardized etymology source. Unfortunately, it seems that no such source exists. Most etymologies are verbal descriptions, brief histories of words. Another problem is the resolution of conflicting etymologies, and the procuring of sources for the more difficult dead intermediary languages.

Another problem is the tracing of the lineace of “direct” descendants and the existence of multiple concurrent versions of a word, especially before modern spelling reforms. Consider also how to take into accound the fact taht first attestatations of a word often long precede their entrance into the standard lexicon. Such a network would not capture this dynamic aspect of the lexicon.

Nonetheless, if we only care about investigating the general structure of such a network and restrict ourselves to a fairly small corpus of “modern” words, it should be possible.

As I proof of concept, I plan to compile such a tree by hand. I hope at some point to have time to write a script to scrape and format the data appropriately for the construction of a more substantial network.