Generative Grammar

The contribution of Generative Grammar to Latin and Ancient Greek linguistics


1. A brief history of Generative Grammar contributions to Latin and Greek linguistics

The starting point is Lakoff’s Abstract Syntax and Latin Complementation (1968) that sounds as a revolution among the linguists : all the attention is paid to syntax, the rules are borrowed from English analysis : it-deletion, abstracts verbs etc.

Reports on this book are not always good ; in France, Touratier is an exception ; he wrote :

  • « On doit reconnaître que ces hypothèses, même provisoires, ont l’immense avantage de proposer des règles explicites et formelles, et posent ainsi les bases d’une recherche approfondie et systématique qui rendrait compte non seulement de l’unité de la syntaxe d’une langue par exemple, mais aussi de tous les détails de ses structures syntaxiques les plus complexes. » Touratier (1969: 121)

H. Pinkster wrote in the introduction of the proceedings of the 1981 ICLL session (1983: xii) these rather critical words :

  • « Transformational Generative Grammar, in its development over the years, has been the most influential of all in Latin linguistics (which, however, is not to say that it has also contributed most to new insights into Latin) both in the field of Syntax and the field of morphophonology »

Besides the hegemonic tendency of Generative Grammar during this period, some theoretical difficulties arose and some key objections were made against the appliance of a generative theory to dead languages.

1.1. Difficulties and objections

a- The inadequacy of the theory for Latin and Greek linguistics :

The first difficulty comes from the object itself assigned to a Generative Grammar, we can summarize quoting Chomsky (1965 : 4):

  • « A grammar of a language purports to be a description of the ideal speaker-hearer’s intrinsic competence. »

The way of elaborating the grammar of a language is empirical : hypotheses are tested by an investigation of the competence. Therefore a Generative Grammar of dead languages seems to be impossible, for no speaker-hearer’s competence could be reachable.

Another theoretical difficulty is the scientific methodology which is applied. The empiricism that Chomsky assumed is based on verification of hypotheses that can be falsified. This scientific approach is not assumed by all the linguists and a recurring charge against this method is that the theory is mapped on to the data, instead of developing theory from the data.

b- The (obvious?) non-configurationality of Latin and Greek :

A set of rules is (or was) necessary to derivate the Surface Structure from the Deep Structure, among them Movement. This rule supposes we can distinguish two different syntactic positions. From a more general point of view, a geometric syntax is relevant only to configurational languages, i.e. to languages having a fairly rigid word order based on a specifically ordered D-Structure.

On the contrary, many other theories consider that Latin and Greek are non-configurational languages, because of their variable word-order and the dislocated form of DPs and VPs.

c- The negation of the relevance of quantitative considerations :

Chomsky (1957: 17) considers both semantics and statistics irrelevant to the theory of syntax:

  • « I think we are forced to conclude … that probabilistic models give no particular insight into some of the basic problems of syntactic structures. »

Many other theories consider on the contrary that statistical observations are essential to define the real use, for example Manning (2003 : 296) :

  • « generative grammar has produced many explanatory hypotheses of considerable depth, but is increasingly failing because its hypotheses are disconnected from verifiable linguistic data »

d- The use of corpus :

According to Chomsky (1965), the ideal speaker-hearer’s competence cannot be defined by investigating a set of performances, because several parasitic phenomena interfere during the production of these performances.

Our knowledge of Latin and Ancient Greek is entirely dependent on corpus. Latin and Greek could be considered as finite sets of data. So we have to explore these data and there is no other way to access Latin and Greek syntax.

1.2. Answers to these difficulties and objections


I borrow the first answer from Lakoff herself:

  • « in many cases, even nonnative speakers can judge the grammaticality of a sentence in a language they know well … therefore we do not need attestation to certify that a sentence is grammatical. » (Lakoff 1968: 2)

That is to say : classical grammars contain syntactic information we can rely on.

No data can be investigated without an explicit or implicit theory. The specificity of the empiricism is that, starting from a hypothesis, we conduct an experiment on data to verify it ; but at the same time we need to produce a contradictory hypothesis and verify that it is not valid, because it produces ungrammatical or unattested sentences.


It is demonstrated that Japanese or Latin, often considered as non-configurational, actually are configurational. There is no reason to exclude Latin or Greek from the Universal Grammar, UG, that the Generative Grammar intends to define. Nevertheless the geometric representation of the syntactic structures by the mean of trees remains difficult, as we can see in a very simple sentence (Touratier 1994: 702-704) :

and some linguists consider that, since the representation is unfit, all the syntactic structure is wrong. Now a more adequate representation of the same sentence would be :


Most linguists using the Generative framework lean on statistical evidences to identify the variations they have to explain. Grammaticality and acceptability are no longer presented in a dichotomous way : there are degrees of grammaticality and acceptability.


Corpus could be used as fields for investigation according to the empirical method of GG. An hypothesis can be tested on a corpus : this hypothesis makes certain predictions concerning data we can verify.

1.3. Some developments and new analysis

1.3.1. Chronology of the models

The first model, Syntactic Structures (1957), with its transformational rules didn’t produce many developments in the area of dead languages, apart from Lakoff’s book.

Aspects (1965) gave rise to new studies on crucial domains of the syntax. Fauconnier (1974) represents this new direction very well : he wrote 3 papers focussing on AcI construction in Latin, the reduction of relative clauses in Latin and grammatical agreement in Greek. New publications, such as Linguistic Inquiry, contain papers on Latin and Greek generative linguistics (Andrews (1971) on Case agreement in Greek).

Government and Binding (1981) and Principle and Parameters (1993), introducing a total change in the theory, that becomes modular, seem more appropriate to the characteristics of Latin and Greek. The theory contains a Case-theory and attention is paid to phenomena well known in Latin and Greek syntax, such as AcI, binding of anaphoras etc. During this period, from 1980 to 1990, we do observe an increasing interest of linguists for Latin and Greek, for example Milner on proleptic accusative in Greek (1980) and on reflexive pronoun in Latin (1978); at the same time linguists coming from classical philology became interested in generative theory and applied it to some classical problems of morphology or syntax, first of them in Europe G. Calboli who mentioned Chomsky’s new proposals in La linguistica Moderna e il Latino, I Casi (1972). Then I have to mention Oniga (1988) on compounds, Maurel (1989) and Biraud (1987) on NPs, Calboli and Goggin (1983) on AcI, Bertocchi on pronouns and Binding Theory and Maraldi on proleptic accusative… and Bortolussi (1987) on case assignment.

A new generation of linguists has been arising from 10 years, along with the development of the last framework, called Minimalism (1995). But the studies seem to be moreso continuations of Principle and Parameters than specific contributions of Minimalism. These studies focus on two aspects of Latin and Greek:

- variations, that could be explained by parameter changes : word order, in particular into NPs (DPs);

- diachronic changes, in particular how Latin structures changed to give birth to Romance languages.

1.3.2. Around phonology

The influence of Generative phonology on Latin and Greek linguistics is rather weak. I can mentioned one study, which is a comparison between the Prague’s structuralist model and the Chomsky and Halle’s one. From the birth of Generative phonology to the present time Paul Kiparsky is the only one who deals with classical languages and in particular with Sanskrit and Greek.

Other issues take place in non-linear models that appeared thereafter : squelettal, moraïc, autosegmental, for example, studies on compensatory lengthening in Greek (Wetzels 1986, Hayes 1989), simplification of geminate consonants in Latin (Pillinger 1983), geminate consonants in Sanskrit (Calabrese 2009), Latin accent (Wetzels 1986, Marotta 2000, 2006) and vocalic alternations and metrical theory (Oniga 1990, 2010).

Books on general phonology quote occasionally Latin or Greek, for example Michael J. Kenstowicz (1994 Phonology in generative grammar ) : dissimilation of –alis> -aris, p. 512 and Greek core syllables p. 264 sqq.

1.3.3. Around morphology

The first paper on dead languages morphology is from Kiparsky (1968). Although other papers appeared episodically in Linguistic Inquiry, for example Embick (2000) on the Latin Perfect, the main studies on Latin morphology are due to Oniga (1988, 1989, 1992, 2000 etc.), in particular his book on nominal composition. His synthetic presentation included in Il Latino (2007) shows the way to simplify the complexity of nominal and verbal paradigms, with few phonological and morphological rules : derivational or compositional rules, flexional rules and at last Phonological readjustment (Item and Process framework).

For example, the forms of the third conjugation are generated in the following way :

The rules are :

- drop thematic vowel before inflectional vowel

- convert short e to i before consonants (other than r)

These rules are very powerful to describe synchrony and moreover to teach Latin inflection, even if, from a diachronic point of view, the rules that have actually been applied may be different. But these rules can be found in many other languages and also apply in Latin evolution to Romance languages. See for example Calabrese’s study on case system “Some remarks on the Latin case system and its development in Romance ” (1998) in J. Lema and E. Trevino, eds., Theoretical Advances on Romance Languages, Amsterdam: John Benjamins, 1998, 71-126.

1.4. Syntax

1.4.1. Case assignment

The most expected contribution of the GB framework for Latin and Greek syntax concerns Case-assignment. In the previous decades many theories, (Kurylovicz, Happ, etc.) failed to explain how case assignment operates and GB theory appeared to be a promising approach. But in fact the new framework focussed more on conditions allowing case assignment in general than on individual morphological cases : structural case assignment (Nominative, Accusative and Genitive) is the only one really developped, whereas lexical and inherent cases depends on changing situations among the different languages.

Apart from my own study on accusative case in Latin, unfortunately unpublished, very few papers concern this topic. To summarize, the central hypothesis is that morphology offers a set of forms which are used at different levels ; so it is not surprising that the same case is employed in various ways, carrying potential ambiguity. For example, according to the modular approach of the theory, verbal object is not a very simple function ; we have to distinguish at least two levels: a semantic level in which is defined the thematic function (agent, theme and so on) and the structural level in which appear phrases (NP, PP etc.). Accusative case is assigned to phrases, without consideration of the thematic role, including situations in which an object has no thematic role, like internal objects (Bortolussi 1988).

In another paper (1998) I show that the minimal contrast between esse + dative and esse + genitive is not based on the opposite semantic values of dative and genitive, but on different structures in which cases are not assigned in the same way.

1.4.2. Empty categories

Very close to Case assignment is the next topic : empty categories. Government and Binding framework introduced empty categories, that are entities having no phonological representation, but having properties at the syntactic level and in the Logical Form. The goal is to explain how the interpretration is obtained or preserved, when a NP doesn’t occupy a syntactic position in which it is governed or when there is no NP at all occupying this position. We have to distinguish between 3 types of empty categories:

-t, the trace of a category which is moved
-pro, null free pronoun
-PRO, null bound pronoun

- Let us begin with t, trace :
A basic use of this empty category is found for passives. Concerning Latin and Greek, Calboli (1980: 193, 1983: 151) gives the following representation of NcI1):

  • (1) [(NPe) INFL V [S NP VP]] dicitur eo tempore matrem Pausaniae uixisse « one says that Pausanias’ mother lived at this time » / ἐλέγετο Συέννεσιν εἶναι ἐπὶ τῶν ἄκρων (Xen. An. 1,2,21) «one said that Syennesis was on the moutain top »
  • (2) [NPi INFL V [S ti VP]] Marcus dicitur bonus esse «on dit que Marcus est bon»/ πατρὸς… ὁ Κῦρος λέγεται γενέσθαι Καμβύσεω (Xen. Cyr. 1,2,1) « one says that Cyrus was Cambyses’ son »

The subject of the embedded clause is moved from its internal position to the subject position of the main verb. We need to specify in what conditions this rule can be applied, for this rule doesn’t act for every verb.

- pro is a null pronoun occurring as the subject of finite verbs, as in Italian and other pro-drop languages. Pro is a pronoun with gender, number and case features. For example Oniga (2007 : 193) gives the following representation of Lex XII Tab. 1,1 :

  • (3) Si proi in ius proj uocat, proj ito. Ni proj it, proi antestamino. Igitur proi eumj capito

As shown in the previous example, pro alternates with an overt pronoun, either in subject position, as for example in Italian or Spanish, or less frequently in Object position – which is impossible in romance languages.

- PRO appears for example with “control” verbs as subject of infinitive clause :

This theory provides us a new look at some phenomena of Latin and Greek syntax, in particular the AcI construction. The main problem everyone is confronted with is how accusative case is assigned to the subject. In the generative theory, concerning infinitive clauses that I have just introduced, a difficulty immediately arises : according to the Case-Filter the subject can’t receive a case (*N where N has no Case), if INFL doesn’t contain the [+agr] feature. Several explanations have been proposed :

- Subject-to-Object Raising rule is the oldest explanation in the generative framework (Pepicello 1977) : by a movement rule the subject of the subordinate clause is moved into the main clause, becoming the object of the main verb. This rule lacks motivation and doesn’t explain why an AcI constuction can be used also when the main verb is intransitive: see example (1) above ;

- Exceptional Case Marking, introduced by Chomsky (1981), and resumed in Maraldi (1980): the verb of the main clause, when could be able to assign case through the limit of the clausal domain, if CP doesn’t contain any conjunction. The difficulty in Latin and Greek is that AcI is not exceptional; thus this explanation seems to be unconvincing;

- Default Case : the idea is that in a case system one case plays the role of default case; this morphological case appears when no case at all could be assigned to the NP ; this is an alternative to PRO. This solution has been proposed by Goggin (1983) and is discussed in Calboli (1996). The fact that accusative case is, in Latin and in Greek, the default case has little evidence : Nominative seems to be a more evident candidate to this function, if we remember for example nominatiuus pendens in Latin ;

- Accusative assigned by Comp: Cecchetto & Oniga (2002) and Ferraresi & Golbach (2003) argue that Latin has an infinitive rich morphology, able to express tense and mood. The consequence would be, according to Cecchetto & Oniga, that infinitive clauses are introduced by a null and abstract complementizer, similar to for in English, that assigns accusative case to the subject of the embedded clause. A rather similar conclusion is drawn by Lucio Melazzo (2005). According to him, the infinitive clause occupies an argument nominal position (DP), the infinitive mood is connected with the C-domain and the null complementizer contains a feature licensing accusative assignment.

On Greek AcI Svedali’s works (2003, 2004, 2005, 2007, 2009) lead to conclusions partly close to the previous one. She argues that control infinitives and AcI have entirely distinct structures and that Ancient Greek is a mixed language, being agreement prominent in finite clauses, and being focus-prominent in non-finite clauses and that would explain case assignment. In her papers she uses Minimalism framework with very technical contents that it would be too long to present here.

Another problem, linked to the previous one, is to explain inside syntactic structures how reflexive pronoun se and ἕ / σφᾶς / ἑαυτὸν works. The first study is due to Bertocchi & Casadio (1980), with the minimal opposition represented by (6) and illustrated for example by (7) in Latin and Greek :

  • (6a) Ioannes sororem suam uidit
    (6b) Ioannes sororem eius uidit (1980 : 26)
  • (7a) (Alexander)i uix a sei manus abstinuit (Cic. Tusc. 4,79)
    (7b) ἁπλῶς τὴν ἑαυτοῦi γνώμην ἀπεφαίνετο Σωκράτηςi (Xen. Mem. 4,7,1)

The way of analysis is a structural one : the antecedent must c-command the reflexive pronoun or adjective, that is to say the antecedent is in an higher syntactic position, as in the following schema :

The previous analysis of PRO allows us to see that in the following examples the antecedent is in a syntactic position c-commanding the reflexive pronoun :

  • (8) (Caesar) Heluetiosi [PROi in fines suosi reuerti] iussit (Caes. Gall. 1,28,3)
    « He ordered the Helvetii, the Tulingi, and the Latobrigi, to return to their territories »
  • (9) οὐκ ἄξιον εἴη βασιλεῖi [PROi ἄφειναι τοὺς ἐφ’ἑαυτὸνi στρατευσαμένους ] (Xen. An. 2,3,25)
    « it was not seemly for the king to let those who had marched against him depart in peace »

This theory seems unable to explain two phenomena :

a- Emphasized possession

b- Indirect reflexives, which constitute overt violation of the Opacity condition (an anaphora must be bound inside the domain containing a subject)

It is often observed (Bertocchi & Casadio, Touratier) that in the first example Pompeio is topicalized and could be called a “logical” or a “semantic” subject. Since pragmatic categories, such as Topic and Focus, are considered in the last model of the theory to be syntactic categories belonging to the Left Periphery of the sentence, Pompeio occupies a syntactic position c-commanding the possessive, either sua domus is the subject of IP or in a focus position. However I am not sure that all the examples of emphasize possession could be explained in the same way.

The second problem is more difficult. Everybody argues that the syntactic approach is inadequate and that the subject of the main clause must be considered as to be the ‘logical’ subject licensing the reflexive pronoun. The conclusion of Bertocchi & Casadio’s paper is that “further investigation is required to clarify the semantic and the syntactic function, if there is one, of these Latin logical subjects.” (1980 : 39).

Zribi-Hertz (1996) faces this difficulty using “logophoricity”, a concept borrowed from Hagège and used before in the description of latin reflexive by Sznajder (1995) in another framework. A similar way of explanation is found in Calboli (1997), but without the concept itself. The more extensive study on this question is F. Nicol (1995) about Greek reflexives : αὐτός would be quite similar to english –self, that is to say a ‘réflexiviseur’ used to enhance the very reflexive pronouns ἕ and σφᾶς ; he tries to explain the distribution of simple reflexives and enhanced reflexives, saying that only enhanced reflexives are arguments of verbs. When an indirect reflexive pronoun appears outside a c-commanding structure (for example Dém. Olynth. 3,7), Nicol considers that the antecedent is a « sujet de conscience », a concept not so far from “émetteur de l’énoncé” in Fruyt (1987)

Conversely a non reflexive pronoun could be used , when the subject of the main clause doesn’t support the reality of the embedded statement :

  • (11) ἔλεξε Φιλοκρατὴς οὐτοσί ὅτι ἀδελφον αὐτοῦ ἀποκτείναμι (Ant. Chor. 21)
    « this Philocrates said I have murdered his brother »

1.4.3. Proleptic accusative

It may be surprising that generative theory has paid such great attention to such a little topic in Latin and Greek syntax : from Miller (1974) to Faure (2010), there are 7 studies on this question. Why is it so interesting ? Because this phenomena involves case assignment, thematic grid, subordination, movement, anaphora and pragmatic interpretation.

  • (12) metuo fratrem : ne intus sit (Ter. Eun. 610-611)

The first analysis is consistent with the traditional one, using Raising rule from the embedded clause to the main one. This analysis has been criticized for the first time by Milner (1980) about Greek and by Maraldi (1986) and Bortolussi (1987) about Latin. According to these authors, the structure is base-generated : the accusative is assigned by the main verb, although the proleptic NP is not an argument (cf. Bolkestein calling it pseudo-argument); and there is inside the subordinate clause an anaphora, usually a null pronoun pro, but sometimes a overt one.

The last studies on prolepsis focus on the pragmatic level. The Topical interpretation of the proleptic NP appears essential in Panhuis (1984) and Rosen’s (1992) studies. This dimension is incorporated in Fraser’s (2002) generative analysis, and at last in Faure (2010) and Bortolussi (forthcoming). The problem is now to identify the syntactic position occupied by the proleptic NP in the Left Periphery, either of the subordinate clause or of the main clause.

1.4.4. Other subordinate clauses

a- Apart from Lakoff’s initial book, very few linguists deal with subordinate clause, save AcI construction. I have to mention Faure (2010) work on interrogative clause.

b-Relative clauses are a traditional topic in generative syntax. Fauconnier (1974: 177 sqq) is the first to adapt Chomsky’s theory to the description of Latin relative clauses. A deletion-rule was the way to produce epithetic adjectives :

  • (13) Homo amat feminam quae est pulchra Homo amat feminam pulchram

At that time the question was to choose between two rules generating relative clause :


  • (14) the man [I saw WH- man] → the man [WH- man I saw]

b) ‘Raising’:

  • (15) ___ [WH- I saw the man] → the mani [WH- I saw ti]

Latin data show that Raising is an invalid explanation because examples exist such as (16):

  • (16) Habet bonorum exemplum, quo exemplo sibi / licere id facere quod ille fecerunt putat (Ter. Haut. 20-21)
    “He has the example of good Poets; after which example he thinks it is allowable for him to do what they have done”

with duplication of the antecedent, “Wiederholung des Beziehungswortes im Relativsatz” (H-Sz. 1965 : 563).

Another idiomatic relative construction is investigated by Maurel (1989) and Bortolussi (2005), ‘Relative Verschränkung’ in Hofmann-Szantyr (1965 : 568 sq) and Kühner-Stegmann (1955 : 309 sqq) :

  • (17) (aedes) quas quotienscumque conspicio fleo (Plaut. Capt. 97)
    *(une maison) que chaque fois que je (la) vois je pleure
    « a house that every time I see it cry »

Maurel’s hypothesis is that relative pronoun can raise in fronting position if and only if it ‘pied-pipes’ all the clause it is moved from. If pied-piping exists in Latin or Greek, we have a great argument in favor of movement rules and a proof that Latin is a configurational language.

1.4.5. NPs

Last but not least, we have to mention a great number of studies on NP structure. A crucial hypothesis concerning the clause structure is that all the categories have the same internal structure, that is to say a lexical Head, a Specifier on the left side and a Complement on the right side. Therefore we can compare:

Bertocchi & Maraldi (1990) showed that this analysis could be extended to Latin NPs : when a N has two arguments, for example an Agent and an Object, the Agent is expressed under Spec. and the Object is in the genitive case, in a complement position:

  • (18) Nostra defensio dignitatis tuae
    « our defense of your honour »
  • (19) laudis nostrae gratulatio tua (Cic. Att.1,17,6)
    « your congratulations for I praised you »

The following studies on this topic take place in a more extensive approach of NPs. We will look at them soon.

Retour à l'introduction ou Aller aux § 2-3

1) See also Maraldi (1980: 66) Marcusi dicitur [S ti bonus esse].