search wiktextract `etymology_text` for reliable patterns

[jmviz]

Wiktextract etymology_text can be searched in conjunction with the templates with the aim to pick up reliable etymology patterns that may be missed from processing the templates solely. E.g.:

• from {{}}, chains -> impute der(s)
• initial From {{}} and {{}} and From {{}} + {{}} -> impute an affix (e.g. druwits)
• final Ultimately from {{}}. -> impute a der (maybe a root?)
• final Equivalent to {{}} + {{}}. -> impute a surface analysis, which generally should not be preferred but could be used in case there is/are no valid preceding template(s).

In each case, the template(s) may be any der-like template or m.

In etymology_text the templates are all fully expanded, so this would involve progressively scanning through the text to find the nth etc. template expansion.

Consider also:

• https://arxiv.org/pdf/1906.03158.pdf, https://github.com/plkmo/BERT-Relation-Extraction
• https://arxiv.org/pdf/1909.07755.pdf, https://github.com/lavis-nlp/spert
• phi-2 candle, ollama