I have been studying natural semantic metalanguage (NSM).
It has been studied and cross-referenced fom 30+ languages.
the core 65 changes several times, in the beginning it was 13 core concepts.
It is as objective as it can to study human language, but of course there is still some subjectivity in it. Some words they excluded like world.
If its models all the way down, these are probably some of the core language concepts, but there will be many more other core concepts learned first. The first words babies first learn is mom and dad.
40-50 classes of verbs?!?! – To a large extent, this outcome is due to the work of Beth Levin. Her most influential book is English Verb Classes and Alternations (Levin 1993).
They often use templates for definitions. Templates are how they can describe some words.
template used for some action verbs like run/swim:
someone does something somewhere for some time
because of this, this someone is moving in this place during this time as he/she wants
words that can’t be broken down more.
Many nouns are converted directly to verbs, butter can be morphed into “he buttered the bread” and a knife can be morphed into “she knifed him”.
Fun and funny are related morphologically, but not semantically, there are many other words like this.
Some of the definitions are kind of weird, but great since not circular.
it really makes me think about variables in intelligence. someone, something, somewhere, sometime, this, do, etc.
questions : ask NSM team are there any research papers that show the order of concepts of NSM babies learn first?
why are many definitions so different, for example swim?
speech-act-verbs : want, think, know, many others.
word2vec is circular, every “word definition” is references back to other words. Other language dictionaries are also circular. NSM is not circular in that the the NSM primes have no definition, and then all other words can be built on top of them + NSM molecules.