You are again making a sever confusion between "lexemes" (your comment is true about them: it is a form in some orthographic system) and "lemmas" (strictly identical to "senses").
I just said that your schema makes 1-to-many relations between LEMMAS and SENSES where this should be 1-to-1.
there are 1-to-many relations from LEXEMES to LEMMAS=SENSES, I've not contested that. but we cannot use LEXEMES as the base of text abstraction (in an abstract language), we'll use LEMMAS.
We don't need any complex relation like LEXEME --(1-to-N)--> LEMMA --(1-to-N)--> SENSE (the second pair is non-sense it should be 1-to-1, and thus merged).
The abstract text will contain LEMMAS (semantic), from which some rules will decide which lexeme (lexical and very specific to each language) to use according to the target language and other constraints, and then which form of the lexeme (grammatical derivations/inflections/conjugation/contextual mutations or particles, plus capitalizing rules for some syntaxic or presentation forms)