From the department of the self serving, I will note that their focus for the computational NLU micro theory is pragmatics.
They realized that humans shortcut their language use by not dealing with the details but switching meanings. From the book:
“Regarding the latter, in his analysis of the place of formal semantics in NLP, Wilks (2011) reports a thought-provoking finding about a sentence type that has been discussed exten- sively in the theoretical literature, illustrated by the well-known example John wants to marry a Norwegian. Such sentences have been claimed to have two interpretations: John wants to marry a particular Norwegian (de re), and he wants to marry some Norwegian or other (de dicto). When Wilks carried out an informal web search for the corresponding “wants to marry a German” (since marrying a Norwegian was referenced in too many
linguistics papers), the first twenty hits all had the generic meaning, which suggests that if one wants to express the specific meaning, this turn of phrase is just not used. Wilks argues that computational semantics must involve both meaning representation and “con- crete computational tasks on a large scale” (p. 7). He writes, “What is not real Compsem [computational semantics], even though it continues to masquerade under the name, is a formal semantics based on artificial examples and never, ever, on real computational and implemented processes” (p. 7).”
The “marry a German” research illustrates the need to reflect the real world, but that we also paraphrase until we need to select a meaning that the paraphrase meaning just can’t convey in context. Words don’t matter, meaning does - see emojis. I submit that lexeme should define a concept meaning not a word meaning. We use words, but communicate concepts. The NLU computational construct must reflect the real use of language.
Further, for a knowledge representation, words and N-grams are not knowledge carriers, concepts in context are the real world knowledge carriers. We should probably model our NLU solution the same.
Tagged words and word proximity computing are a path that approximates how we choose meaning, but fail in generating language due to the approximating nature and lack of real context. We can, however, use that computational approach to build a mirror of human communication (a pragmatic solution) by vectorizing the real world use and de-duping (paraphrase grouping). The result is also computational. Over time, we maintain meaning vectors and paraphrase groups to better reflect real world human use. We de-emphasize words and concentrate on concepts in contexts, just like human communicators do.
Lastly, a pragmatic computational approach will have a natural language interface - even non developers can read and interact with zero training - just concepts and contexts with their associated knowledge items.
Thank you Amirouche, great read for NLP nerds.