To Cognitive Robotics Home

To Stemmatic Syntax Home

Semio-linguistics and Stemmatic Syntax

Per Aage Brandt

 

 

 

1. A strong claim

The grammar of a natural language is its capacity to form sentences. These finite units of virtually infinite discourse are composite, i. e. syntactic, wholes, whose parts are determined in two ways: as ‘functions’ and as ‘words’ (lexical and morphological elements). There is no one-to-one correspondence between functions and words in a language; words inflect according to their assumption of multiple functions, and a function is compatible with multiple word flections.  An important portion of the morphological signs in a language is dedicated to the ‘insertion’ of words in functional slots. This double determination of the parts of the syntactic whole is irreducible, because sentences and words are distinct linguistic fundamentals. Both are semantic: words have ‘meanings’, but only when integrated in sentence ‘meanings’, they come to ‘mean’ what the speaker intends when he is saying what he ‘means’.  

 

    The practical account called a grammar of a language (a ‘school grammar’) intends to describe how words— lexical and morphological entities—enter a sentence in that language through this functional phenomenon called syntax. Linguists do not generally agree on any theory of syntax, despite the many attempts made throughout the twentieth century, but grammarians nevertheless often agree on the functional structure of a given sentence in the language they describe. Syntax therefore appears to be a natural property of language and a phenomenon that we can be immediately conscious of when sentences occur.

 

In linguistic perception—whether the speaker’s auto-perception or the hearer’s allo-perception—syntactic structure is directly apprehended. Therefore it is often described in terms of concatenated functions, like a sequence of notes in a melody. It is often rendered linearly, as if these functions simply followed each other and their linearity constituted the essence of their syntactic interrelations. We get descriptions such as: subject (noun phrase) – verb – object (noun phrase) – preposition – nominal (prepositional phrase); or NP V NP PP NP. The latter style of notation, used in cognitive linguists’ writings, e. g. as a shorthand for constructions like the famous ‘caused motion’ blend, is in fact a historical remnant of early generative grammar, which believed in linearity as a structural principle for pre-transformational ‘kernel sentences’. The reintroduction of such untheorized linearistic assumptions of generative or school grammar in current cognitive linguistics manifests its lack of courage in syntax as well as its predominant interest in semantics. But since linear sequences, ‘strings’, do not per se constitute structural coherence in language (which is not calculus, built on strings of symbols), we will still have to explain syntactic structure.

 

    It is clear that the theory of conceptual integration (blending) cannot replace a theory of syntax. The process of blending does not in any way account for the existence of ‘phrases’—verb phrases, nominal phrases, etc.—or ‘functions’—subject-of-finite-verb, object-of-verb, datives, genitives, etc. If some or even all constructions are blends of other constructions, then both the input constructions and the blended ones must be structured by the same conceptual principles underlying their perceptual linearity.

 

    As Descartes rightly and pertinently stated, the grammar of a human language is not determined by the specific meanings of its utterances—these enigmatic entities we call sentences. Grammatical structure only depends on the structure of the thoughts we express, and Descartes agrees with contemporary cognitive scientists in seeing these thoughts or conceptualizations as sufficiently independent of the real things they sometimes are about as to make it acceptable to generalize about human cognition as such. Rationalist philosophy is the historical origin of the cognitive sciences of our time. Human thought is free and creative, and so is language; both seem instead to be bound by their grounding embodiment. In all situations, humans will use their sensory-motor neural systems not only to bodily act and react, but also to think and feel, and to express inner on-goings. There are two main forms of embodiment. Firstly, when we categorize things in time and space, we use naming as a natural part of the implied mental act. Naming exploits our capacity to bodily produce acoustic form and to monitor and control it by auditory auto-perception. So phonetics is in fact a natural part of our categorization and ordinary use of categories. Secondly, when we unfold our stories about the categorized or categorizable entities we wish to create shared attention about, we dramatize by gestures the doings and beings of these entities. Narrative contexts for categories are naturally found in our experience of their instantiations. The dramatic gestures that serve our story telling are our primary syntactic expressions. Additionally, we are even able to phonetically call upon our categories while showing by gestures what they are doing, and what else we have to think and say about them; speech and gesture are now coordinated, so that we can bodily present both categories and stories simultaneously in communication. This non-trivial achievement is precisely what makes it possible to integrate words in syntactic phrases and clauses and to obtain sentences. The basic ‘logic’ of this syntactic integration is identical to that of our narrative gestures: it is our basic semantics of syntax—as semantically informed as the expressive register of gesture. It inherits from gesture its articulation into local meanings that modify each other, and its pulsation by short hierarchical clusters separated by closures (full stops, periods).

 

So when syntax gets ‘worded’, we both convoke categories and perform narratives; the semantics of things and the semantics of stories locally integrate. The complex bodily activity that obtains this achievement does it by integrating gesture and phonation. We know from stammers and foreign language learners that this operation is still a difficult and fragile neural process, and that it is more easily obtained when singing, i. e. when narrative gesture and evocative naming integrate in musical song lines. The earliest poetry probably coincides with the earliest forms of linguistic performance of our species; if I am right, singing preceded speaking in the evolution of language. Rhetorical uses of language—public addresses, news reading in the media, advertising, speech acts of some solemnity, etc.—still involves a form of chanting.

 

    The claim here is thus that sentence constructions follow a generic cognitive design that in broad outline determines their constituency and intelligibility. This design accounts for the so-called syntactic functions, and it mediates between the linear manifestation of sentences and their multi-dimensional content: it makes syntax an editing channel, an instance of mediation, between the sound and the meaning of language. Its structural architecture is a prefiguration of the semantics expressed, and also contains information specifying a finite set of possible linear manifestations. It explains the possibility of effortless, fast processing of language, even when contents are non-trivial and not all words known: the Chomskyan creativity, both in production and reception.

 

    This generic design is here referred to as the stemmatic design of sentence syntax. The most important property of this ‘stemmatics’ is that it is canonical. There is, according to the view presented, a finite series of possible constituents, or canonical complements, which appear in a canonical order (if they appear), and both this order and the complement types are determined by an overarching semantic plan, a conceptual scenario construction, which is postulated as universal.

 

    A claim this strong could seem to be an utterly preposterous or neck-breaking enterprise in contemporary linguistics. It would only be justifiable through extended comparative analysis, and the pilot studies carried out so far remain local. The grammatical analyses that have motivated the technical generalizations so far are based on straight-forward grammatical readings of texts and attempts at examining different constructions in languages known to the analysts and searching for occurrences that do not match the structural predictions of stemmatic theory. The methodology of this project is mainly that of a syntactic phenomenology—trained grammarians may be more sensitive to syntactic structure than average speakers, but should optimally arrive at the same results as these—and a simple comparative control of consistency—same constructions should be analyzed in the same way. Experimental grammatical teaching based on stemmatics has shown that stemmatic-syntactic training can enhance foreign language skills substantially. This effect might indicate that the stemmatic approach taps into a syntactic ‘nerve’ in our linguistic minds.

 

    There are evidently still many unsolved problems in semio-syntactic analysis (semantically informed functional modelling) along these stemmatic lines; and anyway, a Theory, in the classical sense of a complete doctrine, of Syntax is hardly a reasonable goal to strive for in the contemporary context of comprehensive human sciences covering research on all aspects of our behavior,  extending from poetics to biology… Nevertheless, a coherent view of syntax in a cognitive framework is still required, and it remains, as in the days of L. Tesnière or of the generativists, an important task to develop the study of the structural meaning of sentences.

 

2. Stemmatic structure.

All sentences are syntactically articulated wholes, and the parts into which they are articulated are all connected to this whole by some functional and thereby meaningful relation. The syntactic wholes have no unconnected parts. The task of a syntactic analysis is to study these connections. I call the network of connections constituting the syntax of a sentence its stemmatic structure. A stemma is, as the Greek term suggests, a 'binder' and a genealogical principle of linear 'descent'. It has a 'head' and a 'stem' to which 'complements' are bound in a regular way. The stemmatic head of a sentence is its finite verb; the head of a nominal phrase is its determiner morpheme (article, pronoun, or even a zero); other stemmatic phrases are headed by adverbs, adjectives, and substantives (in compounds).

It is an important characteristic of the stemmatic architecture of syntactic meaning that stemmata of all kinds share the same canonical set of possible complements. All stemmata can basically take the same possible complements, but the richest unfolding is found in VERB-headed stemmata.

    The elementary unit is the binding stemmatic node, which is currently written as follows (other representations are most welcome, and parentheses are sometimes preferable):

 

What happens semantically is that the heading concept is determined (modified) in some respect by the complement concept; if the binder is manifested by a morpheme, its form indicates a specified binding.

    The verbal standard stemma is an ordered rhapsody of 'school grammar' phrases. It has maximally eight complements, each of which has only one implementation, if any. Binders are symbolized by s, if there is no binding morpheme. A slightly simplified version of the general disposition is the following recursive format:

 

The complements are Ø (zero), verb-headed structures like the main stemma (matrix), other phrases (mainly nominal, or single words (often adverbs). All constructions can be described in terms of this extremely simple stemmatic account of aspects of occurrences—we might think of it as a ‘scenario scanner’. These complement nodes {1-8} are thus determined by standard semantic components of a situational understanding, modifiers that phenomenologically form a cascade of determinations: 8 is about 1-7, 7 is about 1-6, 6 is about 1-5, 5 is about 1-4, 4 is about 1-3, 3 is about 1-2, and 2 is about 1.

 

    Let us begin the exemplification with a complicated case. Structures may look like the following graph of a double relative embedding, found on a grammarians desk:

    The mouse that the cat that the dog chased caught is probably dead

(The linearization of this stemmatic structure raises the subjects C1 and the relative pronouns C3 to the left of their respective heads. The rest of the structure is linearized according to a basic principle: head–binder–complement).

 

The semantic charge of the eight nodes connects the stemma to three scenarios or spaces, showing 1) the mouse's present state, 2) what the cat did to the mouse, 3) what the dog did to the cat. The intriguing issue of understanding how semantic space embedding works might be elucidated by comparing the temporal ordering: {(3) -> (2) -> (1)} to the syntactic embedding; the lowest level of semantic imagery (the dog's scenario) takes the antecedent item (the cat) from the next higher level and fills it into its C3-slot; then the second level (the cat's scenario) follows the cat into this slot and takes the mouse from the highest level with it to fill its corresponding slot; and finally the mouse takes its state with it, so that time flows inwards in the resulting semantic embedding of the mouse scenario in the cat's scenario which is now in the dog's scenario. The past is then a surrounding circumstance of the present. Understanding the death of the mouse is reading its scenario from its surrounding scenarios. Syntactic embedding thus represents surrounding semantic frames of the matrix scenario (the mouse…). Syntactic and semantic embeddings are closely related, here in fact by being inverse.

 

    Stemmatic analysis applies directly to manifested structures and does not imply underlying 'deeper' structures. The integrative semantic operations are supposed to be performed directly by the formal organizations that sentences grammatically manifest. Structural variations yield semantic variations. So, in this case, we could have had, e.g.: The dog first chased a cat, and this cat then caught a mouse; this mouse is probably dead now. Here, coordination and the anaphoric determiners are active; temporal sequencing is explicit; and the meaning of the global utterance is quasi-identical to that of the stemma. But the story is no longer told backwards and can use forwards oriented (cataphoric) temporal adverbs that affect the narrative viewpoint, but not the narrated event series.

 

The stemmatic information given by head and complements C1-7 feeds a scenario modalized from the speech act stance (cf. the vocative and the imperatives), and the coordination of scenarios occurs by C8 (and certain punctuation or gesture markers).

 

    C4 and C5 often form a to – from oriented path schema (cf. the meaning of the last French sentence and its à – de) based on the inherent meanings of the directional node 4 and the projectional node 5.

 

    C5 also comprises the passive agent, the active instrumental, the comitative and the ‘stylistic’ indication of manner.

 

    C6 and C7 indicate framing circumstances: location in some space (C6) and realization at some time (C7), which is the same as epistemic ‘weight’.

 

    Relative clauses are considered epistemic determiners (C7) of their nominal antecedents (cf. Our Father, who…); some relative clauses are also explicative (parenthetical) and introduce additional meaning referring to the verbal matrix of the nominal antecedent. For example, it is relevant to mention the geographical location of Our Father, because his will is already done where he is, in heaven, and the request is now that it be done as efficiently or truly on earth as well.

 

    Genitives are inter-nominal indications of provenience (C5) or a host of similar meanings related to the projective node meaning (cf. notre pain de ce jour).    

 

    Stemmas and strings of the same sentences are openly interrelated—by a one-to-many projection in both directions. This openness explains of course the difficulty of designing mechanical grammatical parsers. Natural parsing in syntactic perception and monitoring is guided by the mental context of semantic networks that further specify the stemmatic, ‘semio-syntactic’ scenarios in discourse or speakers’ situation. But the possibility of semantically understanding  (writing and reading) literary texts—in poetry and fiction—which do not offer contexts of this sort, testify to the forceful restrictions that must be efficient in the process of linguistic reading, thus translating strings into stemmas and letting stemmas be semantic instructions for building imaginary wholes structured as narratives, descriptions, deliberations, or other coherent, articulated mental objects. The literary reading must in fact be literal, and complete an act of semantic construal, before it lets the reader perform an interpretive act of semantic generalization, and so do justice to the text’s possible artistic quality. This aesthetic evidence—of reading preceding interpretation—shows us that such a process of structural reduction is necessarily happening.

 

    If we conceive of stemmatic structure as an efferent, expressive, ‘funnel’ reducing and converting ideational semantics-of-mind into node-structured semantics-of-sentences and then sentence trees into strings, then the inverse, afferent process of expansion of such strings into full-fledged ideational contents, perhaps organized as webs of networks of mental spaces, would correspondingly start with a delinearization of the sentence-as-a-string, converting it into a sentence-as-a-stemma (or a limited set of possible stemmatic readings) and then converting the semantics of the stemmatic construction into a mentally visible whole. There are two equally difficult steps in the latter, afferent process: the projection string-to-stemma and the projection stemma-to-scene. My intuition is that the two projections are however equally realistic (such processes really exist); that they are phenomenologically distinct (we experience them sometimes as distinguishable); and that the first projection is technically both the simplest to describe (or elaborately simulate) and the most important to the problem of understanding how humans manage to communicate by exchanging sequences of sound.

 

 

References

 

Brandt, Per Aage, 1973, L'analyse phrastique. Introduction à la grammatique, Bruxelles: Ed. AIMAV

Chomsky, Noam, 1966, Cartesian Linguistics, New York. Harper & Row

Chomsky, Noam, 2000, The Architecture of Language, New Delhi: Oxford University press

Tesnière, Lucien, 1965, Éléments de syntaxe structurale, Paris: Ed. Klincksieck

 

*