A SeSyn grammar is transformational, as it transforms, via the CYCLE and the POSTCYCLE, SA tree structures into surface tree structures (SS), which serve as input to the morphology and phonology of the language L in question (Fig. 1-a). SeSyn thus makes explicit how propositional thoughts are expressed in any given L, having been generated by, or in, cognition and cast into the predicate-logical propositional form of an SA, via the LEXICON of L, combined with a speech-act operator. SAs are not universal but already geared to the L at hand through the choice of language-specific lexical predicates, which dictate the predicate-argument structure of the sentences of L.1

The algorithmic procedure from the SA input to the Phonetic Form output, is taken to be a module (Fodor 1983), i.e. a fully automated algorithmic procedure, inaccessible to introspection or interference and open to conscious control only at the input window, though with multiple feedback possibilities allowing for correction before and after final delivery (Seuren 2009a: 234)—not unlike a (very fast) production line in a factory.2 The grammar as such is thus fully automatic and meaning-independent, but its input is semantically determined, and integrated into the whole of the conscious and subconscious human mind.

SeSyn differs basically from a Chomskyan Generative Grammar (ChoGG), which generates sentences randomly, triggered by an arbitrary START symbol, and producing as yet meaningless symbol structures: ChoGG is a random sentence generator, whose intermediate output “Σ” is assigned a semantic representation yielding a logical form or LF (roughly SA), next to a phonetic interpretation producing a phonetic form, as shown in Fig. 1-b. A ChoGG grammar thus has a double output for every random input. Since humans are not random sentence producers, the system cannot represent a cognitive-realist grammar.

SeSyn began, during the mid-1960s, as an offshoot of ChoGG called Generative Seman-tics, a movement started by Jerrold Katz and Paul Postal in their (1964) and further developed mainly by Jim McCawley, Haj Ross and George Lakoff, with myself as a European founding member. The main argument was that a ChoGG-type grammar of the then current kind is nontrivially simplified and made empirically more adequate if it is assumed that the input consists of a well-defined SA-structure (the nature of which was developed by McCawley and myself), while the surface structure output is roughly as assumed in ChoGG. That argument still holds for all later manifestations of ChoGG.

The initiators of Generative-Semantics/SeSyn did not, at the time, realise that their theory actually originated with Wilhelm Wundt (1832–1920), who had proposed that natural-language sentences are derived from an underlying cognitively real propositional structure (Seuren 1998: 219–227). Nor did they realise that the Swiss linguist Albert Sechehaye (1870–1946), junior member of staff in Ferdinand de Saussure’s linguistics department in the University of Geneva, had taken up Wundt’s challenge (Sechehaye 1908, 1926) and had thus in effect laid the foundations of Generative Semantics and thus also of SeSyn (see Seuren 2018b for extensive discussion).

SeSyn immediately met with excessive resistance from Noam Chomsky, who, however, failed to provide any substantial argument, pointing merely at aberrations and excesses by some overenthusiastic but undisciplined adherents. Regardless of what motivated this hostility, the net result has been that ChoGG has, in actual fact, remained faithful to its neopositivist (Carnapian) origin as an algorithm triggered by pushing a start button and with an output in need of a semantic interpretation. Chomsky himself has always been ambiguous about whether ChoGG still follows the mid-20^th-century-type neopositivist model of a scientific theory, which excludes, as a matter of principle, any causal role of anything like ‘the mind’—the notorious horror mentis of 20^th-century science—or whether it is to be seen as a cognitive-realist theory, in the spirit of the Cognitive Revolution that took place at Harvard during the 1950s and to which Chomsky, then also at Harvard, had pledged allegiance (Chomsky 1968/1972). While the new Cognitivism promoted the mind as a prime causal factor in human behaviour, Chomsky, though considered a leading figure of the movement, never admitted that, in fact, it ran counter to the ChoGG grammar model he was in the process of developing and promoting, and whose theoretical foundations he was, apparently, unwilling to revise. What Chomsky attacked was behaviourism, not neopositivism, which makes him only a half-hearted participant in the Cognitive Revolution, despite his prominent position in it. ChoGG is thus still firmly rooted in early 20^th-century neopositivism and thus runs counter to the letter and the spirit of the post-1950 Cognitive Revolution, despite claims and appearances to the contrary.

In the wake of the Cognitive Revolution, a massive shift is now under way from neopositivism towards cognitive realism, drawing cognitive neuroscience into the picture. It rests on, and is motivated by, mere principles of good science: neopositivism fails for the cognitive side of the human sciences simply because it does not deliver the goods. A journalist recently asked me the blunt question: why is ChoGG wrong? My answer was simple: ChoGG has remained caught between, on the one hand, the cognitive absurdity of its formalism, blocking a realist interpretation, and, on the other, its perverse formal tortuousness if taken in a nonrealist sense (Seuren 2004: 70). Typologists, who, sanely, study language in its ecological environment, have developed an allergy to theoretical grammar, as that field is entirely dominated by ChoGG, which only provokes shock reactions. This has proved highly deleterious to linguistics as a whole, as it blocks a sound insight into the universals of language, which are to be found in the system rather than in the surface phenomena. Normal scientific method simply requires cognitive realism in the human sciences, and SeSyn fulfills that condition as far as possible.

No matter how one looks at it, when the foundations of a theory are unsafe, then sooner or later the edifice will collapse. This is now visibly happening to ChoGG, which signally fails to link up with adjacent disciplines such as psycholinguistics, neurolinguistics, sociolinguistics, historical linguistics, linguistic typology or any viable form of semantics, all of which are crying out for a serious theory of grammar but are not receiving anything worthy of that name.

Despite the many facts, no matter how important, revealed by ChoGG practitioners in many different languages, no recognition has ever come forward of the equally important facts revealed in terms of the SeSyn framework. In particular, Verb Clustering by means of the rule of PREDICATE RAISING, discussed below, has been systematically ignored by ChoGG practitioners, who themselves have no solution to offer: there is no dialogue between the parties—a very anomalous situation.

SeSyn, being cognitive-realist, fully fits into the pattern set by Cognitive Science. It actively strives for integration with cognitive (neuro-)science, even though the gap between mental structures and processes on the one hand and their neural correlates on the other has not (yet) been bridged. Like its syntax, the semantics of SeSyn is fully cognitive-realist, the notion of mental proposition being basic to all cognitive, logical and (con)textual processing. For that reason, SeSyn requires sweeping and highly consequential measures with regard to standard logic, which model-theoretic semantics has been unwilling to consider. Model-theoretic semantics is thus likewise irrelevant to the study of natural language. SeSyn manifests a novel, mentalist, anti-positivist approach to linguistics, cutting through the horror mentis that has plagued the human sciences for over a century.

The claim is thus that the cognitive-realist approach of SeSyn is empirically superior to any existing alternative. This assertion cannot be seriously underpinned in a single presentation but it can be demonstrated by a few examples. In the present context, the focus will be on two outstanding features of SeSyn: Auxiliation and Verb Clustering through the rule of PREDICATE RAISING (PR). Just to show how SeSyn works, the algorithmic derivations of an English and a French sentence are shown from their SA input to their output Shallow Structure (ShStr). The interest is that these cyclic processes are found in languages all over the world, allowance being made for language-specific differences.

This makes SeSyn especially relevant for linguistic typology, which ranks as the most pr0mising branch of linguistics but has so far lacked a satisfactory basis in the theory of grammar. The problem with linguistic typology is that it is happy with intuitive and impressionistic language descriptions, oblivious, it appears, of the fact that there is an underlying systematically functioning grammatical mechanism. As specifications of such a mechanism such descriptions are hopelessly inadequate. In this regard, SeSyn helps out: any typologist familiar with SeSyn will immediately recognise syntactic phenomena in the languages they study as instances of the rules and categories of SeSyn, just as any SeSyn theorist immediately sees what is going on in the syntax of a language unknown to them.

In each S-structure, the predicate is placed first, followed by the lexically defined number of argument terms (NP or S). English SA thus differs from SS, where the dominant order is: Subject-Verb-Object (SVO), that is, NP-VP. The transition to surface SVO-order is effected by the CYCLE.3 A predicate has minimally one and maximally three argument terms: subject, indirect object, direct object—in that order. If one, it is the subject; if two, the first is subject and the second is direct object. If three, the middle term is indirect object. Subject and direct object terms may consist of an embedded S. The embedding of Ss under a matrix predicate is known as COMPLEMENTATION.

Verb-initial, or VSO, languages are produced by leaving out one single rule feature, as will be shown in a moment. For Verb-final languages, such as Japanese or Turkish, the underlying order is, ceteris paribus, taken to be SOV. For the time being, two possible SA constituent orders are assumed: VSO and SOV. How to handle other possible orderings (including Ergative) in the languages of the world is a matter of current research.

The SA-structure in Fig. 2-a is divided into three parts: the Speech-Act operator, AUX and the LEXICAL MATRIX. The first is left undiscussed, given the format of this presentation. The LEXICAL MATRIX contains the main lexical predicate, which is, normally speaking, a Verb, Adjective or NP in SS, plus the argument terms. The subject and object terms may be an embedded complement-S. The Auxiliary Area (AUX), contains, in principle, all elements that somehow qualify the proposition expressed in the LEXICAL MATRIX, such as tense, modality, aspect. Overall, the MATRIX tends to incorporate elements of both AUX and embedded complement-S-structures—a tendency I call matrix greed. AUX-elements are usually, but not always, united with the MATRIX V-constituent, thus forming a complex V-cluster, as is demonstrated in the derivations presented below.

In English, as in most other European languages, AUX minimally contains two tenses, t1 and t2, in that order. Both allow for a double choice, the former between deictically defined ‘Present’ (PRES) and anaphorically defined ‘Past’ (PAST), the latter between ‘Simultanuous’ (SIM) and ‘Preceding’ (PREC) (both sim and PREC with regard to the time segment referred to by means of t1). For English, the following correspondences hold:

PRES+ SIM → Simple Present (eat) PRES + PREC → Present Perfect (have eaten)

PAST + SIM → Simple Past (ate) PAST + PREC → Pluperfect (had eaten)

In most languages, the precise semantics of the tenses is complex, idiosyncratic and strongly language-specific. The Futuricity operator is a modal predicate, both grammatically and semantically, not a tense, also in languages that express the future morphologically in the Verb form (for SeSyn this is just a matter of surface category specification in the LEXICON).

Modal predicates (will, may, can etc.) are optionally placed in SA between t1 and t2 and are thus part of AUX. This explains their ‘defective paradigm’: they lack infinitival and participial forms and occur only in the Simple Present and Simple Past, never in the perfective tenses, which require the perfective auxiliary have plus the Past Participle, but they can be followed by either a SIM or a PREC infinitive, as in may eat and may have eaten.

Historically, most AUX-predicates, notably modal and aspectual predicates, but also Tense predicates (as one learns, for example, from the history of Creole languages), started life as full MATRIX predicates embedding what is now the lexical MATRIX as a complement-S, but getting ‘bleached’ over time, which made them become part of AUX—a universally found tendency known as auxiliation (Kuteva 2004). Apparently, the Dutch Futuricity verb zullen (‘shall/will’) had not yet auxiliated when I acquired Dutch some eighty odd years ago, but it seems to have auxiliated in the meantime. For me and my contemporaries, a sentence like Hij had zullen vertrekken (‘*He had willed leave’), where zullen occurs in the Pluperfect, is fully grammatical. But nowadays, I find to my surprise, young speakers no longer accept that sentence, which means that zullen has auxiliated in the meantime. Its German equivalent werden auxiliated much earlier, as is shown by the fact that the equivalent German sentence *Er hätte abreisen werden is ungrammatical for all speakers. But other, similar, verbs have not, or not yet: Er hätte abreisen sollen (he should have left) is impeccable, like its Dutch equivalent Hij had moeten vertrekken. In Old-English, the modals still had participles and infinitives, but these disappeared in Middle-English (Fischer et al. 2004), clearly because they fell victim to auxiliation.

A sharp distinction is made between the status of lexical items in the lexicon on the one hand and their surface lexical category on the other. In the lexicon, all items are predicates: verbs, nouns, adjectives, adverbs, conjunctions, prepositions, quantifiers, negation, are all listed as ‘PREDICATE’’. The lexicon specifies for each predicate—apart from its semantic specification—its surface category, its argument structure, its phonological representation(s), and the cyclic rules induced by it (its rule features). This accounts for the sometimes surprising differences in word class of otherwise semantically equivalent lexical elements. In Finnish, for example, the negation is realised as a modal predicate followed by an infinitive, with a defective paradigm identical to that of the English modals.

The CYCLE carries the SA to the shallow structure (ShStr), which is is fed into the postcycle. The latter is only touched upon here as it is highly language-specific, while the CYCLE is surprisingly uniform across the languages of the world (ShStr is typically a halfway product, already recognisable as a yet unborn surface structure). Which rules are assigned to the CYCLE and which to the POSTCYCLE is a matter of experience and judicious choice.

So now the question is: what is the CYCLE?4 The CYCLE is the first processing box the SA is fed into. It contains a handful of universal rules either deleting (semantically recoverable) elements in a lower S, or lowering the PREDICATE of a given S into an embedded S, or raising an element of an embedded S into itself. Cyclic rules are induced by the predicate of any given S in SA. The CYCLE starts at the most deeply embedded S whose predicate induces a cyclic rule.5 The LEXICON specifies for each predicate which cyclic rules it induces (obligatorily or optionally) and which surface category is assigned to it as the CYCLE passes through it. The cyclic rules induced by a predicate as specified in the LEXICON are shown in the trees as rule features between angled brackets below the predicate at issue. When the predicate induces no cyclic rule, we say that the CYCLE passes vacuously through the S in question. When the CYCLE has done its work in a given S, it passes on to the immediately superior S, recursively, until it reaches the highest S, at which point the CYCLE comes to an end and delivers the shallow structure (ShStr) of the sentence being processed. This is then fed into the POSTCYCLE.

So let us now see how the CYCLE operates for the sentence The cat may have eaten the mouse (Figure 2). We start with the SA-structure (Fig. 2-a). The predicate of S3, the lowest S in Fig. 2-a, induces no cyclic rule (carries no rule feature), so that the CYCLE passes vacuously through S3, but the label “Predicate” over eat is replaced with the surface category label “V” (Verb). Then, at the S2-Cycle, the rule L (LOWERING) applies. L occurs in a number of varieties. In this case, L left-attaches the predicate in question to the V(erb) of the lower S, forming the cluster _V[_V[have] _V[eat]], as shown in Fig. 2-b (have, the surface form for prec, is specified in the LEXICON as a surface V). The original S2-node is eliminated given the general principle that an S dominating only another S is ‘pruned’. At the next cycle up, the S-modal cycle, the Predicate may, likewise a surface V, is lowered in exactly the same way as have, resulting in Fig. 2-c. At the S1-Cycle, two rules apply, SR (SUBJECT RAISING) and L, in that order. SR takes the subject-NP of the embedded S and places it in the position of that embedded S, which is moved one position to the right. In virtue of a general DEMOTING PRINCIPLE, the S of the now subjectless S is relabelled “/S” (traditionally known as Verb Phrase or VP), meaning that it needs an NP to be a full S. This gives Fig. 2-d. (Without the rule feature SR for t1, stage Fig. 2-d is cut out so that the language keeps VSO as surface order.) All that remains is the application of L at the S1-Cycle, yielding Fig. 2-e, the ShStr of the SA-input (PRES is labelled for surface category ‘Affix’). This last structure is now left to the mercy of the POSTCYCLE, the MORPHOLOGY and the PHONOLOGY of the grammar of English, each successive stage being more language-specific and closer to the corresponding SS than the previous one.

Corresponding phenomena occur in many other languages even though they differ considerably in other aspects of their grammar, especially when the causative predicate is relabelled ‘Affix’. The phenomenon known as ‘valency increase’ is, in most cases, simply due to PR with ‘cause’ relabelled as ‘Affix’: intransitive or transitive verb stems, augmented with a causative prefix, suffix or infix, or morphophonemically modified, acquire an extra argument in the dative case, as in the following Turkish example (Lewis 1975: 146–147):

(1) a. Mektub-u imzala-d-ım

letter-ACCUSATIVE sign-PAST-1sg

(I signed the letter)

b. Müdür-e mektub-u imzala-t-t-ım

director-DATIVE letter-ACCUSATIVE sign-CAUSATIVE-PAST-1sg

(I made the director sign the letter)

The dative müdüre signals that PR has been at play. If the embedded clause had been intransitive, as in I made the director leave, the word müdür (director) would be in the accusative case (müdürü), just as in French, Italian and the many other languages that show evidence of PR in their complementation system and have case marking. In general, the occurrence of a dative representing the semantic subject of a transitive clause embedded under a causative verb is a sure sign that the language in question has PR, rather than SR, in its complementation system. This also applies to many lexical predicates, taken by McCawley and others to incorporate a bit of syntactic structure, as in the lexical predicate show, prelexically analysed as ‘cause-to-see’: ‘Ann caused-to-see Didier the letter’ then becomes Ann showed Didier the letter. Apparently, English does have PR in its prelexical syntax, though it lacks PR in its open syntax, using SR instead, as already implied by McCawley.6 (Ironically, ChoGG adherents have now started to take to prelexical analyses as well.)

It is not so, however, that PR always induces a dative when the embedded complement-S is transitive. In German, for example, accusative case is assigned, as in (2a) (probably as a result of 16^th-century Latin influence in German schools). Overuse of PR may lead to a succession of more than three NP-terms in a row, separated from the corresponding series of PR-raised verbs, as in the Dutch subordinate clause (2b), where the accumulation of four NPs in a row strains the short-term storage capacity needed for processing the clause, even though it is produced according to the rules (luckily, Dutch makes PR optional in enough cases for the grammar to allow for more easily processable alternatives, such as (2c), which has three consecutive NPs) (for a detailed analysis and description, see Seuren 2018a/1996, Chapters 5 and 6):

(2) a. … daß Johann das Kind den Ball zu werfen gelehrt hat.

… that John the_acc child the_acc ball to throw taught has

(… that John has taught the child to throw the ball)

b. … dat Jan het kind de hond de bal wilde laten leren halen.

… that John the child the dog the ball wanted let teach fetch

… (that John wanted to let the child teach the dog to fetch the ball)

c. … dat Jan het kind de hond wilde laten leren de bal te halen.

… that John the child the dog wanted let teach the ball to fetch

(… that John wanted to let the child teach the dog to fetch the ball)

A further powerful argument supporting the PR-analysis is provided by the behaviour of anaphorical clitics in French and some other Romance languages. In these languages, unaccented pronouns and a few anaphorical adverbs occur as clitics attached to the Verb form in a strictly defined order (Seuren 2009b gives a detailed analysis of clitics in French and Italian). What is relevant here is that the placement of clitic clusters, at least in French and Italian, is sensitive to the /S (or VP) constituent they are part of. Consider the following two sentences:

(3) a. Anne la lui fera voir. (*Anne le fera la voir.)

(Ann will make him see it)

b. Anne veut la lui donner. (*Anne la lui veut donner.)

(Ann wants to give it to him)

Fig. 4-a shows what sentence (3a)—with Didier and la lettre replaced with the clitics lui and la, respectively—looks like after the postcyclic rules Ø-DELETION and AFFIX-HANDLING have applied to the ShStr shown in Fig. 3-f. Ø-DELETION takes out the Ø branch of the V-cluster of Fig. 3-f; AFFIX-HANDLING subsequently takes all affixes of the V-cluster, starting with the lowest one, and re-attaches them to the right of _V[faire], the new cluster being relabelled ‘Finite Verb’ (FV). Now assume a postcyclic rule Clitic Movement (CliticMov), left-attaching clitic argument terms within any given /S-constituent to the FV-constituent or, failing an FV, the V-constituent, of the same /S (the order in which this is done is of no concern here). Since there is only one /S in Fig. 4-a, the pronominal clitics lui (to him) and la (it) are left-attached to the FV-cluster, as shown in Fig. 4-b. This explains the position of la + lui in (3a) and the ungrammaticality of *Anne le fera la voir.