Re there is no grammar: this has been a perennial issue in CL/NLP. Different people have grown up with different relations to language. Some take some habits more seriously than others. Some give some habits more value/authority/status than they deserve, so they become rules. And some obey rules more than others, so rules get internalized and one ends up believing that there is something "magical" about language. (Some also try to exploit rules enough to make others' lives miserable --- this is how language/grammar can be used as a weapon [language attitudes]!). But all in all, we just communicate in whichever way we end up doing so.
On Wed, Jul 26, 2023 at 5:15 PM Ada Wan adawan919@gmail.com wrote:
Dear all The primary reason I got onto this thread has to do with what I sensed might be an attempt to promote certain methodology, one that direly needs some re-evaluation, much like many other in the space of language and computing (and/or CL/NLP, digital humanities... etc.). I know that many practitioners in this space have computed using "words" as a representation and therefore might have had many hypotheses as to what kinds of textual relations is to behave how in the vector space etc., many might even have related grammatical relations to certain spatial relations --- but what is one to make of e.g. different grammatical relations having the same statistical representations, or different statistical representations having the same grammatical relations? And as any trained linguists could inform one honestly, there is really no "grammar". There are no "grammatical relations" that are "intrinsic" to language.
@Peratham: many of the statements that you made don't really make sense or lack clarity, if you think about them, e.g. "[t]ensor arrays are just ER diagrams most of the time" --- this depends on the data and how it is being represented. (I assume "ER" here refers to "entity relationship".) Re "I don’t feel them as a very powerful framework for every system.": the matter is not about having "a very powerful framework for every system" but to understand the limit (and the lack and irrelevance) of "words" (esp. in computing). Re "And tensor methods do not protect lots of people living under illegal and crime circumstances. This is probably off-topic but it is possible for many people to be not protected by laws and polices. As you may know.": I don't understand this statement of yours. Would you please clarify?
@Ibrtchx: Re "characters, words, phrases, sentences, ... all the way to whole books are always intra- and intertextually relational" --- I agree, except for the inclusion of "words" and "sentences" as these are, at least, obsolete, unreliable, and non-universal. We can do better in this regard. Anything we examine can be relational, assuming we have established or understood the connection. But note that the connection may be in us, instead. Re "being 'relational' has a measurably tractable meaning brought about by the dot product in a vector space ;-)": this depends.
On Wed, Jul 26, 2023 at 4:00 AM Albretch Mueller lbrtchx@gmail.com wrote:
On 7/25/23, Peratham Wiriyathammabhum peratham.bkk@gmail.com wrote:
Luckily, words are often relational. Nice having some dialogue with you.
characters, words, phrases, sentences, ... all the way to whole books are always intra- and intertextually relational and, once again, being "relational" has a measurably tractable meaning brought about by the dot product in a vector space ;-)
Other people stumbling onto this thread will certainly notice the context in which it was framed.
lbrtchx