Research paperOn the parallel evolution of syntax and lexicon: A Merge-only view
Introduction
Humans, and only humans, recursively combine words into sentences. Crucially, as modern theoretical linguistics has clarified and stressed many times (see Everaert, Huybregts, Chomsky, Berwick, & Bolhuis, 2015), human language syntax is based on hierarchical structure rather than linear order, so that it often happens that the same word order conveys different meanings depending on which structure underlies it.
(1a) has at least two readings, as illustrated by the different bracketings in (1b) and (1c). Structure dependence is one remarkable hallmark of human language, to be found nowhere else in animal communication.
Unfortunately, this fact is often misunderstood or ignored by nonlinguists. A recent example is the report by Suzuki, Wheatcroft, and Griesser (2016) that the Japanese great tit (Parus minor) has “compositional syntax.” It is an ingenious study of bird communication system but the experiment does not show at all what it purports to show. Here the two calls ABC and D are said to have distinct meanings, and the birds react “compositionally” when these are combined into ABC-D, but not when the order is artificially reversed (D-ABC). Obviously, this is a string-based signal without any hierarchical structure. Failure to react “compositionally” to D-ABC is evidence that the birds do not have the kind of compositional semantics we find in human language. The authors’ misrepresentation of their experimental data derives from their failure to understand human syntax and its semantic compositionality.1 Animal communication is definitely a valuable key to human language evolution, and it would be highly regrettable if its import was degraded by studies like this.
Words and sentences are two central units of analysis in linguistics, and the traditional approach sets a division of labor by two separate modules – the lexicon and syntax. Accordingly, in studies of language evolution, it has been a customary practice to discuss the evolution of the lexicon and that of syntax as distinct events. In the words of Berwick and Chomsky (2016, 66), the two tasks in explaining language evolutions are “to account for the ‘atoms of computation,’ the lexical items” and “to discover the computational properties of the language faculty.” A problem with this approach is that as opposed to syntax, up to now there has been no substantial proposal as to how human lexicon may have evolved. Berwick and Chomsky (2016, 84–85) stress that human words and concepts are radically different from what one may find in animal communication systems. This gap is too huge to be filled by gradual evolution from animal signals (primate or not), and the evolution of these unique properties of human lexicon needs to be explained, but then “how, no one has any idea” (Berwick & Chomsky, 2011, 40). There must have been a (r)evolutionary event to make this leap from animal signals to human words possible.
Largely by drawing on and extending current minimalist syntax of generative grammar (Chomsky, 1995, Chomsky, 2005, Chomsky, 2015, inter alia), I will suggest that the evolution of the lexicon was brought about by Merge, the fundamental combinatorial operation which gave rise to human syntax. The resulting picture will be a more parsimonious scenario of language evolution, where the lexicon and syntax did not evolve separately but both were driven by this single computational device. Effectively, this amounts to saying that “Merge is all you need” (Berwick, 2011, 491) not only for syntax but for the lexicon, too.2
This article is organized as follows. In Section 2, I first illustrate how certain verbs and their conceptual structure are built up by Merge, and suggest that lexicalism should be removed from studies of language evolution. In Section 3, I reexamine distinct notions of lexical items (LIs), and suggest that none of them are the target of language evolution. Instead, I suggest that some innate pre-linguistic symbols, which may not be even uniquely human, are enough for Merge to generate LIs. In Section 4, I challenge the view that Merge is a uniquely human function from its start, in favor of a more natural evolutionary scenario by preadaptation or exaptation. In Section 5, I introduce and refine the hypothesis of the motor control origin of Merge (Fujita, 2009, Fujita, 2014), according to which Merge evolved from sequential and hierarchical object manipulation. I suggest we can distinguish two ways of applying Merge recursively.
In Section 6, I address certain issues on labeling, another central topic for current minimalism. I introduce the notion of label superposition to capture the problem of label indeterminacy, and show that it is Merge that collapses such superposition, which further corroborates the Merge-only view of language evolution pursued here. I also discuss the possibility of directly merging conceptual units (or roots) to construct uniquely human rich conceptual structure. Labeling in action and music will also be briefly considered. In Section 7, as an illustration of how label superposition collapse actually works, I outline a new analysis of transitivity alternation. In Section 8, I introduce the string-context mutual segmentation hypothesis by Okanoya and Merker (2007) as a plausible scenario of how pre-linguistic symbols may have evolved, and propose to combine it with the motor control origin of Merge to construct a comprehensive hypothesis on human language evolution. Section 9 concludes the article.
Section snippets
Syntax within words
Traditionally, many theoretical linguists have adopted some sort of lexicalist position which claims, among others, (a) that lexicon generates words and syntax combines them into phrases and sentences, and more importantly (and I believe wrongly), (b) that lexical information determines syntactic structure. Thus, the reason for the ungrammaticality of (2b) as opposed to (2a), or that of (3b) as opposed to (3a), is just because the verbs arrive and arrest are lexically specified as unaccusative
On lexical items
Words are therefore not something pre-syntactically given in the lexicon, and to the extent that the lexicon refers to a list of words, it must be an output of syntax. Distributed Morphology (DM) is one influential approach along these lines today (see Embick and Noyer (2007) for a concise exposition), and it is worth noting that this framework provides a useful three-way distinction of the concept of a lexicon. First and foremost, the lexicon is a pool of conceptual atoms for syntax to combine
Against Merge as a human autapomorphy
The minimalist program of generative grammar (Chomsky, 1995 et seq.) adopts the view that Merge is the nuts and bolts of the human language faculty, and that our innate biological endowment for language (Universal Grammar) may be reduced to this single function. Accordingly, the central issue of language evolution can be equated with the evolution of Merge.
Merge takes two syntactic objects and combines them into one set (Merge (A, B) = {A, B}) and its recursive application (i.e., it can apply
How Merge may have evolved: action to syntax
Merge as a set formation operation has such a simple and general feature that it is fairly easy to find a similar function in other cognitive and physical domains of both humans and nonhumans. Among others, action and music have attracted much attention in the literature because of their possible link with human language syntax (for action and language, see Fitch & Martins, 2014, as well as Pulvermüller, 2014, and Boeckx & Fujita, 2014, for music and language, see Asano and Boeckx, 2015,
Label and Merge
As is clear in the above observation on compounding, the distinction between Pot-Merge and Sub-Merge is tightly connected to how the syntactic objects generated by Merge are labeled. Labeling is a hot topic in recent generative grammar (Chomsky, 2013, Chomsky, 2015) and no discussion on Merge is complete without reference to labeling.
Unlike what was once assumed in early minimalism (Chomsky, 1995), labeling is now detached from the definition of Merge. It is no longer a syntactic operation, but
Label superposition collapse and verb formation
I have argued in Section 2 that syntax is responsible for generating verbs with complex conceptual structure, using the split VP structure as an illustration. The import of this observation is that one can depart from the traditional lexicalism and reduce syntax and the lexicon both to the working of Merge, thereby rendering language evolution an even more accessible topic.
Assuming too much lexical information which needs to be defined pre-syntactically raises the issue of where such
On the origin of pre-linguistic symbols
Given only pre-linguistic symbols (conceptual atoms which are presumably not even unique to the humans), human language combines them by Merge into more and more complex linguistic forms (from LIs to words and sentences; see Fig. 1). The remaining question is, needless to say, where those pre-linguistic elementary symbols first came from.
One promising scenario is provided by Okanoya and Merker’s (2007) “string-context mutual segmentation hypothesis.” According to this hypothesis, our ancestors
Conclusion
Dependence on hierarchical syntactic structure is one remarkable feature of human language, and language evolution cannot be understood fully unless one also understands how this feature first appeared in the human lineage. Modern generative grammar has succeeded in deriving it from the working of Merge and thereby reducing language evolution largely to the emergence of this combinatorial operation.
However, Merge first requires basic units to which it can apply to construct syntactic structure,
Conflict of interest
The author has no financial or personal relationship that could cause a conflict of interest regarding this article.
Funding statement
Part of this work was supported by JSPS Grant-in-Aid for scientific research (C) #16K02765.
Acknowledgements
My deep gratitude goes to the two anonymous reviewers for their helpful comments on an earlier version of this article and to the guest editor Dieter Hillert for his kind invitation. I also thank the organizers of and the audience at the First International Symposium on the Physics of Language held at Sophia University, Tokyo, on March 4–5, 2016, where I first introduced the idea of label superposition. Usual disclaimers apply, of course.
References (55)
- et al.
Cortical memory mechanisms and language origins
Brain and Language
(2006) - et al.
The functional neuroanatomy of serial order in language
Journal of Neurolinguistics
(2014) Problems of projection
Lingua
(2013)- et al.
Sequential learning in non-human primates
Trends in Cognitive Sciences
(2001) - et al.
Structures, not strings: linguistics as part of the cognitive sciences
Trends in Cognitive Sciences
(2015) - et al.
Social cognition and the evolution of language: constructing cognitive phylogenies
Neuron
(2010) - et al.
The development of rulebound strategies for manipulating seriated cups: a parallel between action and grammar
Cognitive Psychology
(1972) A new notation system of object manipulation in the nesting-cup task for cimpanzees and humans
Cortex
(2007)- et al.
Engraved ochres from the middle stone age levels at Blombos Cave, South Africa
Journal of Human Evolution
(2009) On the similarity between syntax and actions
Trends in Cognitive Sciences
(2014)
Language evolution and an emergent property
Current Opinion in Neurobiology
The syntax of action
Trends in Cognitive Sciences
The evolution of language from social cognition
Current Opinion in Neurobiology
On the morphosyntax of (anti)causative verbs
Syntax in language and music: what is the right level of comparison?
Frontiers in Psychology
All you need is merge: biology, computation, and language from the bottom up
The biolinguistic program: the current state of its development
Why only us: Language and evolution
Elementary syntactic structures: Prospects of a feature-free syntax
Syntax, action, comparative cognitive science and Darwinian thinking
Frontiers in Psychology
The minimalist program
Minimalist inquiries
Three factors in language design
Linguistic Inquiry
Approaching UG from below
Problems of projection: extensions
Distributed morphology and the syntax-morphology interface
The manipulative complexity of lower Paleolithic stone toolmaking
PLoS One
Cited by (19)
Moving beyond domain-specific versus domain-general options in cognitive neuroscience
2022, CortexCitation Excerpt :For example, comprehending action verbs “kick”, “pick”, and “lick” activates motor regions associated with leg, arm, and face movements (Pulvermüller, 2005), indicating that those regions are reused in a newer context of language comprehension (Anderson, 2010). Moreover, as non-human primates display a certain degree of hierarchical combinatorial ability in the domain of action (Byrne & Russon, 1998; Hayashi, 2007; Matsuzawa, 1991), its reuse in syntax of language and music was suggested (Asano, 2021; Fujita, 2016; Greenfield, 1991; Hilton, Asano, & Boeckx, 2021; Pulvermüller, 2014). They all argue for continuity from action to language and music.
How did language evolve in the lineage of higher primates?
2021, LinguaCitation Excerpt :For example, if a speaker knows 25 words for each of the grammatical roles subject, verb and object, it is possible to produce more than 15,000 distinct sentences (Pagel, 2017). Furthermore, experimental studies on human cognition inform us that many mental and/or neural computations are not specific to language but shared with other cognitive activities or derived from more general cognitive computations (Fujita, 2017). It is, however, open to debate, which linguistic computations are the outcome of cultural accumulations and/or modifications and which are a direct product of the neural properties of the human brain.
Language and action in Broca's area: Computational differentiation and cortical segregation
2021, Brain and CognitionCitation Excerpt :In some versions of the Strong Minimalist Thesis (SMT), Labeling is separated from Merge, as a prerequisite for the conceptual-intentional system (see below; Berwick & Chomsky, 2017; Chomsky, 2013; Rizzi, 2016; but see Hornstein, 2009 for a different account). Phylogenetically, recent attempts propose that the basic syntactic objects themselves might be the result of some combinatorial operations merging atomic features together (Fujita, 2017). The Merge-based syntactic system is proposed to interface with two different systems, once all feature requirements are fulfilled: the sensory-motor and the conceptual-intentional systems (Berwick et al., 2013; Bolhuis, Tattersall, Chomsky, & Berwick, 2014; Fujita, 2014; Lasnik, 2002).
Communication: Animal Steps on the Road to Syntax?
2017, Current BiologyCitation Excerpt :First, documenting compositionality in more systems is required to establish how widespread this communicative capacity is, its general form and function, and the selection pressures underpinning its emergence. Second, although compositional, the structures so far identified in animals pale in comparison to the productive, hierarchical structures that characterise human linguistic systems [16]. Further studies are required to determine whether animals can communicate using compositions of more than two calls, and whether they can use the same calls flexibly in different configurations to further increase their repertoire.
Developmental changes in the interpretation of an ambiguous structure and an ambiguous prosodic cue in Japanese
2023, Issues in Japanese Psycholinguistics from Comparative Perspectives: Interaction Between Linguistic and Nonlinguistic FactorsReps and representations: a warm-up to a grammar of lifting
2023, Linguistics and Philosophy