David Rose On ChatGPT Recognising Canonical Structure

David Rose wrote to Sysfling on 1 Apr at 10:12:

I’m just looking for the limits of GPT and whatever it is that ‘context’ is supposed be...

These are each instances of a pedagogic register type in which a learner solicits knowledge.

GPT recognises the canonical structure, even when it’s initiated with a statement...

		speech function	exch role	register phase	interact
Mario	Your self is merely semiotic.	statement	K2	propose class	solicit knowledge
GPT	That is correct.	answer	K1	evaluate	approve
	As an AI language model, my "self" or programmed understanding of my existence and capabilities is based purely on symbols and algorithms.	statement	K1	elaborate class	impart knowledge

Keep in mind that these are just labels for structures. We give them functional labels because that’s the relation we interpret between ranks and strata. But GPT doesn't have to interpret them as ‘functions’... just predict appropriate structures in response.

and later that day at 22:01:

But note that GPT is also negotiating affiliation with Mario... positioning itself outside of human communities... no doubt ‘trained’ by its developers.

 

Blogger Comments:

[1] To be clear, ChatGPT produces texts from the lexical collocation frequencies in a reservoir of instances, not from an individuated repertoire of systems that are realised as structures. On this basis, ChatGPT is not an individuated meaner.

[2] To be clear, by 'context' Rose means Martin's misunderstanding of register as a stratum of context. So here Rose is presenting a text (language) as an instance of a type of context, despite context being opposed to language in Martin's stratification model.

[3] To be clear, importantly, it is not ChatGPT that recognises 'the canonical structure' but Rose. See [1] above.

[4] To be clear, in SFL Theory, 'acknowledgement' is the expected response to a statement (Halliday & Matthiessen 2014: 137).

[5] To be clear, ChatGPT does not 'predict structures', because that is not how it operates; see [1] above.

[6] To be clear, ChatGPT is not 'negotiating affiliation', because ChatGPT is not an individuated meaner; see [1] above.

David Rose On What Makes Us Human And Fooled

David Rose wrote to Sysfling on 31 Mar 2023 at 9:23:

I was thinking of Shooshi’s what makes us human... dirty jokes ;-))

Re pickiness of humans... I have a sneaking suspicion GPT’s telling us something important about us, not just itself. How much are we fooled by our synoptic view of systems and texts? We can see emergent patterning when we stare at enough printed text, and then represent it as weighted options in systems. But how much do we actually know about how we process text above the lower ranks of expression? What actually is the relation between structural probabilities in text production and ‘meaning’.

Blogger Comments:

[1] To be clear, for Halliday, it is the stratified content plane of language that "makes us human". Halliday (2002 [1996]: 388):

The more complex type of semiotic system is that which evolves in the form of Edelman’s “higher order consciousness”. This higher order semiotic is what we call language. It has a grammar; and it appears to be unique to mature (i.e. post-infancy) human beings. In other words, it evolved as the “sapiens” in homo sapiens.

Halliday (2003 [1995]: 390, 430n):

In this paper I have tried to identify, and to illustrate, certain aspects of language which seem to me critical to a consideration of language and the human brain. In doing so I have assumed that language is what defines the brain of homo sapiens: what constitutes it as specifically human.

The emergence of grammar … is the critical factor in the development of higher-order consciousness; homo sapiens = homo grammaticus. See Halliday (1978a, 1979b); Painter (1984, 1989); Oldenburg (1986).

Halliday & Matthiessen (2014: 25):

This stratification of the content plane had immense significance in the evolution of the human species – it is not an exaggeration to say that it turned homo ... into homo sapiens (cf. Halliday, 1995b; Matthiessen, 2004a). It opened up the power of language and in so doing created the modern human brain. 

[2] To be clear, in the case of ChatGPT, "we" are not fooled by "our synoptic view of systems and texts", but by ascribing systems to an AI model of language that generates texts from (the lexical collocation probabilities of) instances, not systems.

[3] To be clear, this relation is given by the architecture of language proposed by SFL Theory: structures are specified systemically in the realisation statements attached to features whose probability of instantiation varies according to register. 'Meaning', in the narrower sense, is the stratum of semantics: its systems that are realised as structures, and instantiated as texts.

But importantly, ChatGPT does not use systems that specify structural probabilities to generate texts. Instead, it uses the lexical collocation probabilities garnered from a 'reservoir' of texts, each of which is the instance of the system of the meaner who produced it. (In lexicogrammar, collocation is the syntagmatic dimension of lexis, whereas structure is the syntagmatic dimension of grammar.)

David Rose On Why Language Needs A Stratified Content Plane

David Rose wrote to Sysfling on 28 Mar, 2023 at 9:11:

The way GPT ‘reads’ and ‘writes’ text as probabilistic strings of tokens highlights the question of why language needs a stratified content plane. It sounds something like dynamic processes of DS, as foreshadowed by Firth’s notion of ‘mutual expectancy’. The multiple ‘levels’ of its design could enable it to simultaneously predict and review appropriate structures at several scales and various functions, e.g. figures and figure sequences, lexical strings, reference chains, method of development, evaluation prosodies, exchange roles... [just *predicting structures* not making meanings as persona semiotica].

But lang requires these various scales of DS processes to be (re)organised in local processes of LG. Why?

Blogger Comments:

[1] To be clear, it is not that language "needs" a stratified content plane, but that, on the SFL model, a stratified content plane is what distinguishes language from other socio-semiotic systems.

[2] To be clear, here Rose is merely promoting Martin's discourse semantics, which confuses Halliday's textual grammar (lexical cohesion, reference etc.), Fries' textual grammar (method of development), Halliday's ideational semantics (figures and sequences), inter alia.

More importantly, the dynamics of one stratum, discourse semantics, are irrelevant to the stratification of content, because stratification is the relation between levels of abstraction, and so what stratification affords is the decoupling of congruent relations between strata to open up the enormous semogenic potential of grammatical metaphor. Moreover, as Halliday & Matthiessen (1999: 237) point out:

If the congruent form had been the only form of construal, we would probably not have needed to think of semantics and grammar as two separate strata: they would be merely two facets of the content plane, interpreted on the one hand as function and on the other as form.

[3] This is misleading. Martin's discourse semantics is not organised into 'multiple levels' and 'several various scales'.

[4] On the one hand, this misunderstands the relation between strata, and on the other hand, it is very misleading. Semantics and lexicogrammar constitute different levels of symbolic abstraction, so the lower level (Token) is not a local reorganisation of the higher level (Value): the lower level (lexicogrammar) is a realisation of the higher level (semantics). 

In all of Martin's work, which Rose continually promotes, there is a failure to understand strata as different levels of symbolic abstraction. For example, Martin (1992) misunderstands strata as modules (of the same level of abstraction), and confuses stratification with semogenesis ('all strata make meaning').

The reason why this is misleading is that it presents Halliday's lexicogrammar as a "reorganisation" of Martin's discourse semantics, whereas, in terms of theorising, Martin's discourse semantics is a reorganisation of Halliday's lexicogrammar (cohesion) and semantics (speech function). 

David Rose On The Reluctance To Divorce Language From Consciousness

David Rose wrote to Sysfling on 26 Mar 2023, 14:02:

A more general anxiety about relations between language and personhood exists in our own community. …

But SFL would not have progressed without separating out the systems from their uses and users.

One way our anxiety is expressed is a reluctance to divorce language from consciousness. Yet consciousness is a property of individual persons but language systems are a property of communities. They exist before, after and without the individual persons who use them.

By systems we mean both potential and actual – system and text. Instantiation is a relation between texts and systems, irrespective of persons.

Blogger Comments:

[1] To be clear, Halliday makes a useful distinction between 'person' as a social individual and 'meaner' (language user) as a socio-semiotic individual. Halliday & Matthiessen (1999: 610):

The human individual is at once a biological "individual", a social "individual" and a socio-semiotic "individual":

as a biological "individual", s/he is an organism, born into a biological population as a member of the human species.

as a social "individual", s/he is a person, bom into a social group as a member of society. "Person" is a complex construct; it can be characterised as a constellation of social roles or personae entering into social networks … .

as a socio-semiotic "individual", s/he is a meaner, born into a meaning group as a member of a speech community. "Meaner" is also a complex construct. 

[2] To be clear, this "progress" in SFL is Martin's confused model of individuation and affiliation (critiqued here). Martin et al (2013):

[3] To be clear, for neuroscientist Edelman (e.g 1992), it is language that distinguishes higher-order consciousness from the primary consciousness that humans share with many other species. For Halliday (Halliday & Matthiessen 1999), the theory of experience that has evolved in language equates the content plane of language with the content of consciousness. Ideationally, consciousness is the interior symbolic processing of sensing, and the exterior symbolic processing of saying, which create content through projection, and interpersonally, consciousness is the self enacted as meaner: as an interactant in exchanges.

[4] To be clear, on the SFL model, the collective nature of language entails that human consciousness is also collective. Halliday & Matthiessen (1999: 609):

Edelman's interpretation of higher-order consciousness referred to above suggests that this form of consciousness (unlike primary consciousness) is constituted in language. Language is a socio-semiotic system, so it follows that higher-order consciousness is constituted socio-semiotically; and since socio-semiotic systems are collective, it follows that higher-order consciousness must also be collective. Collective consciousness is an attribute of human social groups — the members of a given culture. But we need to distinguish between the consciousness of a social group and the consciousness of a species, whose collective construal of experience is codified in the structure of the brain. All human populations have the same brain, and to that extent all construe experience in the same way. But humans live in social groups, and their local environments vary one from the other; to that extent, different groups construe experience in different ways. The significance of this for us is that language is the resource for both: both what is common to the species as a whole, and what is specific to the given culture. In the way these two components are construed in the grammar, we cannot tell them apart. But it is the role of language in the construction of experience as meaning — as shared activity and collaboratively constructed resource — that gives substance to the concept of collective consciousness as an attribute of the human condition.

Moreover, here Rose even contradicts the models that he is promoting. In Martin's models, a language user (persona) is an individuation of a culture, and their meaning potential (repertoire) is an individuation of the meaning potential of the community (reservoir).

[5] Clearly, languages do not exist without the individual persons who use them, as demonstrated by the phenomenon of 'language death'.

[6] This misleading, because it is untrue. To be clear, 'system' refers to potential, and 'text' to the instance of that potential.

[7] This is not misleading, because it is true.