A central research question in the history of generative grammar starting from Chomsky (1956) until the eighties has been whether natural languages fall within the class of context-free languages or occupy a higher position in Chomsky’s hierarchy for formal grammars. As late as in 1982, Pullum and Gazdar (1982) claimed that the prevailing answer at the time, namely that context-free grammars are inadequate for the description of natural languages, did not have a solid base. However, immediately after that paper, three independent studies¹1

A fourth paper making the same point for Swedish followed after a few years (cf. Miller, 1991)

In formal linguistics, the interest for Chomsky’s hierarchy somewhat decreased after the eighties, partly because the main research question (does the grammar of natural language require context sensitivity?) had been answered, partly because efforts were concentrated on developing a theory which reduces the number of grammars that are accessible given fixed data, and this is at least in part independent of the Chomsky’s Hierarchy (cf. Lowe, 2021 for a recent discussion that relativizes the importance of Chomsky’s Hierarchy as a measure of complexity) . The radical simplification of the apparatus proposed in the Minimalist Program, including the development of Bare Phrase Theory (cf. Chomsky 1995 and much following work), contributed to this trend.

However, the interest has been revamped in more recent years, because grammatical abilities of non-human species have been investigated to evaluate the (alleged) uniqueness of human language (cf. ten Cate, 2017 and Berwick et al., 2011 for an assessment). More specifically, the position of non-human grammatical systems in Chomsky’s hierarchy has raised a significant controversy. It is debated whether some avian species can recognize serial nested dependencies characteristic of context-free grammars (cf. Gentner et al., 2006 and Ravignani et al., 2015 on this issue and Everaert & Huijbregts, 2013 for a discussion of similarities between birdsong and language). However, to the best of our knowledge no one ever attempted to show that non-human species can recognize sequences produced by mildly-context grammars. So, based on current knowledge, it is safe to say that mild context sensitivity is uniquely human.

In this context, identifying the locus of complex grammatical abilities is particularly important. Out of the four papers that take human language to be context-sensitive, three built their demonstration on clearly syntactic constructions, namely cross-serial dependencies in Dutch, Swiss German and Swedish. This is in line with the tradition that sought to establish that the internal complexity of the clause (as opposed to the internal complexity of the word) is the reason why natural languages are not context-free. These ‘syntactic’ attempts, building on the presence of cross-serial dependencies, followed other syntactic attempts, involving comparatives (cf. Chomsky, 1963) and sentences with respectively (cf. Bar-Hillel & Shamir, 1960).²2

Postal (1964) tries to build an argument based on the interaction of nominalization and incorporation in Mohawk but see Pullum and Gazdar (1982) for a rebuttal.

The Bambara paper is different, though. As already stated in its title, Culy (1985) posits that the non- context-free component lies in what he calls lexicon (although in more current terminology it would be called morphology). However, this specific claim is based on an observation that is very little developed (see below). Our goal in this note is to revisit the claim that what makes Bambara more complex than context-free is its morphology as opposed to its syntax. We believe that revisiting this question is interesting at the light of the current debate on where to draw a dividing line (if any) between syntax and morphology. Many current approaches reject the lexicalist view that there is a neat division of labor between morphology and syntax, namely morphology would come first and assemble morphemes into words, while syntax, which would come later, would assemble the output of morphology (words) into phrases and sentences. Non lexicalist approaches assume that such a line cannot be really drawn because there is only one computational component which is responsible for both word formation and clause formation (cf. Borer, 2013; Embick & Noyer, 2007; and Baunaz & Lander, 2018 for a presentation of different approaches that are however unified by the rebuttal of lexicalism). Under the assumption that there is a unique computational system, it is not expected for morphology to be context-free and for syntax not to be so (or the other way around). This expectation does not arise from lexicalist approaches, which remain neutral on this point.

A preliminary note: we assume for the sake of the argument a minimal notion of ‘word’, intended as a special kind of phrase that, unlike other phrases, respects a condition of lexical integrity (cf. Section 3 below for tests of lexical integrity and Fábregas, 2014 and Cecchetto & Donati, 2015, Chapter 1, for general discussion on the notion of word). Building on this, we develop our argument as follows. We introduce the structure that, according to Culy, makes Bambara not context-free. This structure results from the interaction of the two constructions that are described in Section 2. In Section 3 we ask whether these two constructions pass two lexical integrity tests, and we conclude that they do, converging with the initial characterization by Culy, who however did not systematically apply lexical integrity tests or other tests for ‘wordhood’. In Section 4 we discuss the Bambara findings at the light of the debate between lexicalist and non lexicalist approaches.

The context-sensitive configuration identified by Culy results from the interplay of two constructions. The first one, illustrated in (1) is what he calls ‘Noun o Noun’ construction, where o³3

The ‘’ O ‘’ in ‘N o N’ construction’ is a conjunction. It is written as ‘’ Ó ’’ with a high tone and it differs tonally from the determinant ‘’ Ò ‘’ with a low tone. Following Culy’s notation, accent or tone are not further indicated in this paper. A reviewer suggests that ‘N o N’ construction’ might be a case of syntactic reduplication in the sense of Travis (2010), who discusses cases like ‘student after student visited the professor on Monday’. We acknowledge that the syntactic or morphological nature of the construction is not easily determined but we are confident that the question is addressed by the tests that we apply in Section 3.

The second construction is an agentive construction which has the form ‘Noun(N) + Transitive Verb(TV) + la’ and translates as "one who TVs Ns".

This agentive construction is recursive, that is, the noun phrase in the agentive construction can result from a previous application of the same construction, as in (3).

Importantly, the noun phrase having the ‘Noun(N) + Transitive Verb (TV) + la’ form can be used in the ‘Noun o Noun’ construction.

Culy shows that the combination of the ‘Noun(N) + Transitive Verb (TV) + la’ construction and the ‘N o N’ construction in (4) causes ‘the vocabulary of Bambara’ to be context-sensitive. We refer to his paper for a formal demonstration, which we take for granted. Our goal in this paper is not to reconsider Culy’s demonstration, which, to the best of our knowledge, has never been questioned. Our goal is to double-check the claim that the structure that results from the interaction of the two constructions indeed belongs to what Culy calls vocabulary and can be called morphology in more current terms. Since his description of the relevant constructions is very short, we start by describing them in more details before switching to the main research question. The grammatical judgments reported in this paper are those of one of the authors, who is a native speaker of Bambara. When necessary, he consulted other Bambara speakers. As the basic facts involve sharp grammaticality judgments, we did not deem necessary an experimental investigation.

Having established that the linguistic constructions reported by Culy are attested and productive, we can switch to our main goal, namely establishing the locus of the non-context-free nature of Bambara. This is how Culy motivates that the complexity of Bambara lies in what he calls “vocabulary”:

As he does not further develop this point but simply refers to an introductory textbook description which in turn introduces a special rule for the ‘Noun o Noun construction’, the evidence is not compelling. Therefore, we decided to explore this issue more in depth. As we mentioned, the tests that most researchers adopt to tell apart words from phrases involve the Lexical Integrity Hypothesis. Two of these tests are applied to the ‘Noun o Noun construction’ and to the ‘Noun(N) + Transitive Verb (TV) + la’ construction in the next sections.

In this paper, we revisited the question of what is the locus of the structures that have been shown to make natural languages context-sensitive. While for Dutch, Swedish and Swiss German there is no doubt that the relevant structures are created in the syntactic component, the Bambara facts were less clear. After studying the properties of the relevant structures in more details and after applying two classical lexical integrity tests, we concluded that, assuming for the sake of the argument that a dividing line between syntax and morphology can be drawn, the initial characterization by Culy is indeed correct: Bambara is context-sensitive word-internally.

This result is relevant for the debate around lexicalism. That context-sensitivity is found in structures that lexicalists would characterize as morphological is fully consistent with the non-lexicalist view that there is just one computational system for structure building operations and this unique computational device create words as well as phrases and clauses. On the other hand, Bambara findings do not directly falsify lexicalist approaches, as lexicalists might argue that syntax and morphology, although being different modules, share some fundamental properties, including context-sensitivity.

Although overall non-lexicalists seem to be in a better position, a caveat should be advanced: in order for a non lexicalist approach to be fully convincing, it should be able to explain the facts that we reported in Section 3.1 and 3.2 as cases of violation of lexical integrity in a different way, as by definition, lexical integrity is not an analytical category available to non-lexicalists. Related to this, a reviewer asks whether the cases involving ungrammatical wh-extractions discussed in Section 3.1 can be explained as island effects. This explanation, although very attractive and fully in line with non lexicalist approaches, is not without problems. Take (25) and (26) above: the extraction site in both sentences is the complement position of the verb ‘meet’. This position normally does not introduce an island and furthermore (25) is grammatical while (26) is not. What changes in the two sentences is that in (26) there is wh sub-extraction out a category that a lexicalist would define a (complex) word. Of course it is possible to say that such a category is an island, but without independent evidence this is a stipulation. Furthermore, the island explanation does not immediately extend to the anaphora facts discussed in Section 3.2.

To summarize, we can confidently conclude that the context sensitivity of Bambara, differently from the context sensitivity of other languages, which is clearly syntactic, lies in the domain of what would be traditionally called ‘morphology’. While the fact that context sensitivity holds across ‘syntax’ and ‘morphology’ is very much in line with non lexicalist approaches, questions remain on how these approaches can explain the Bambara findings that prima facie call for the utilization of the lexical integrity hypothesis.