Ryan Nefdt offers an original perspective on linguistics that will be of interest to philosophers of language and philosophers of science more broadly. Nefdt is influenced by Ladyman and Ross (2007; Ladyman 1998) in arguing for an ontic structural realist account of linguistics. The virtue of the view is that without making the success of our theories seem miraculous, it avoids commitment to the full ontology of our current theories in response to a pessimistic meta-induction. That is, it avoids the worry that since many, most, or even all the terms of previously accepted theories are seen as non-referring by present standards, future science will be equally dismissive of the objects we posit. And unlike these discarded terms, structures are thought by ontic structural realists to survive theory change. Nefdt’s book is novel in the ontology of language debate and novel in applying broader considerations in the philosophy of science to linguistics, and thus a significant contribution on both fronts.

First, concerning the metaphysics of languages, which is the focus of chapters 2 and 3, Nefdt offers an alternative to the medieval taxonomy of nominalism, conceptualism, or Platonism (also referred to as realism by proponents) that has shaped the debate. In the first half of the twentieth century, under the influences of positivism and behaviourism, languages were seen as concrete physical things, collections of written and spoken utterances. Theorists associated with this nominalist view include Bloomfield and Quine. By the late 1950s, the view was abandoned in favour of a conceptual understanding of linguistics advanced by Chomsky. Chomsky argued that the stimuli children experience during language acquisition underdetermine the grammar the child learns, leaving it mysterious how children settle on the correct grammar of their community so quickly, the so-called poverty of stimulus argument. The answer, says Chomsky, lies not in external, physical utterances, but rather in internal mental structures and representations children possess innately (internal, individual, intensional I-languages as opposed to external, public, extensional E-languages). Indeed, linguistic structure cannot be in the external physical tokens of language according to conceptualists, since external tokens of ambiguous sentences such as ‘Flying planes can be dangerous’ are physically identical. The different readings are disambiguated via distinct mental representations (think here of syntactic trees). Platonists such as Katz, Postal, and Soames accept the conceptualist rejection of nominalism but accuse conceptualists of confusing language itself with knowledge of language. Languages are external to the mind, but they are abstract objects, not physical, concrete tokens; linguistics is a formal science along with mathematics and logic, for the Platonist. Theorists impressed by the negative arguments against these three positions tend to be eliminativists about languages, whereas those impressed by the positive accounts in favour of each tend to be pluralists; either way, these positions determine the options: ‘We are aware that some philosophers and linguists think there are foundational positions distinct from nominalism, conceptualism and realism [Platonism]. Although we cannot deal with this issue here, every such putative alternative with which we are familiar reduces to one of the three standard ontological positions’ (Katz and Postal 1991, p. 515).

It is against this backdrop that Nefdt introduces his structural realist account. While sympathetic to the aim of pluralists to see languages as multi-faceted entities, Nefdt rejects the framework of reconciling physicalist, mentalistic, and abstract ontologies to do so, either as a loose union (Santana 2016) or an intersectional hybrid according to which the ontologies are compatible (Stainton 2014, p. 48): ‘no special status will be given to […] the traditional categorisations of linguistic ontology in terms of Platonism, nominalism, and mentalism’. Nefdt develops Dennett’s idea of a real pattern (chapter 4) as a better way to understand relations between formal and material modes of being.

Dennett (1991) uses the idea of a real pattern from information theory to explain intentionality. A real pattern in information theory contrasts with random noise; a signal contains a pattern if the signal can be described using less information than it takes to transmit the original signal. As a simple example, the description ‘a sequence one thousand digits long of alternating 1s and 0s, beginning with a 1’, is much shorter than the signal it describes, ‘101010…10’. Dennett’s originality was in extending the domain of application of this technical notion. To demonstrate the possibility of such an extension, Dennett analyses John Conway’s Game of Life in terms of real patterns. The Game of Life is a type of cellular automaton in which a small set of rules determine which pixels are activated at any given time. The underlying rules completely determine what happens in the Game of Life world. Nonetheless, robust and highly predictive patterns emerge. Entities such as gliders, blinkers, and blocks are fallible but useful for determining the behaviour of the Game of Life world, without appealing to the rules. By analogy, Dennett argues that we predict the behaviour of complex organisms without appealing to the underlying physics, which, just as the rules in the Game of Life world, completely determines what happens. Instead, adopting an intentional stance, patterns of rationality, in which creatures act to satisfy their desires based on their beliefs, allow for reliable though defeasible quick predictions. Nefdt follows Dennett’s lead in extending the notion of real patterns to the new domain of linguistics. For Nefdt, there are real patterns in linguistic datasets and grammars are linguists’ ways of representing them, that is, ‘grammars are compression algorithms’ (p. 70).

There are two aspects to Nefdt’s thesis. First, the compression thesis is that a grammar is a shorter way of representing the linguistic patterns. For example, a grammar for English is an algorithm that produces all and only well-formed English sentences, while being significantly smaller than a list of those sentences (an uncompressed representation of English). Grammars meeting this condition are weakly equivalent in being empirically adequate, but the features by which they generate the string set representing the sentences of the language may differ. Following Ladyman and Ross, Nefdt refers to the different features between grammars as indispensable: ‘Such features are likely to identify the structure of the real patterns’ (p. 76). The second aspect of Nefdt’s thesis is that indispensability is a strong guide to real linguistic structure and thereby we see the relation between formal and material modes from which can draw metaphysical inferences. Nefdt illustrates the point by contrasting a dependency graph, which highlights argument structure, with a phrase structure tree, showing hierarchical constituent relations, for the same sentence. Different representations highlight different aspects of real linguistic structure.

The indispensability claim is integral to Nefdt’s ontological position and therefore warrants more discussion than it receives. First, the conclusion seems stronger than the premises marshalled for it. In providing examples, Nefdt merely claims that in different grammars structure is ‘more perspicuous’ or captured ‘to greater effect’ (p. 76). This could be a weak methodological claim about cues for evidence of structure or an ontological claim about what some grammars represent, both of which fall far short of ‘vital and indispensable ontological information about the language qua real pattern’ (p. 76). Nefdt is clear that we only find real patterns in non-redundant grammars in the sense that they cannot be further compressed and still remain empirically adequate, but familiar underdetermination concerns at least invite the query as to why there cannot be a multitude of non-redundant, weakly equivalent grammars that nonetheless do not represent structures of, say, English. Perhaps the notion of non-redundancy can be cashed out to make this analytic, but then we are faced with an epistemic question as to when we have actually found such a grammar and a psychological question about native speakers’ representations. Perhaps other conditions are needed. I return to this point below.

To appreciate just what the patterns revealed through formal models qua grammars are patterns of, Nefdt adopts a systems-level perspective inspired by analogy with systems biology (chapter 5, especially p. 85). This goes beyond the competence of individuals to structural relations within the broader linguistic community: ‘formal structures represent languages which are patterns found in complex biological systems created by a feedback network of conventions, linguistic competence and environmental interactions. These are real patterns in the sense of Dennett and Ladyman and Ross’ (p. 89). The pluralist notion of hybrid objects is replaced by an emergent phenomenon that, Nefdt argues (pace Chomsky), neural networks can be trained to learn. Nefdt presses the point, applying this analysis to the ontology of words (chapter 6). On this view, words, like phrases and rules, are on a continuum and derive their ontological status from their roles in the linguistic structure. While he takes a new route to this position, critics will be quick to note that Nefdt is ultimately aligning himself with the connectionist side in the classical versus connectionist architecture debate and so leaves himself open to those criticisms (see MacDonald and MacDonald 1995).

With the ontological position laid out, Nefdt addresses the pessimistic meta-induction by tracing theory changes in syntax (chapter 7). He claims that structures have been preserved in transitions from proof-theoretic grammars to X-bar representations to the merge operation in minimalism, and he cites formal work by Stabler and Keenan (2003) based on isomorphisms of symmetry groups to identify structural invariants, at least in principle. We are not, however, presented explicit structures that are preserved in each formalism and, even if we were, Nefdt’s position is more nuanced. Structural invariance across theory change would amount to strong equivalence, but recall that for Nefdt, weak equivalence is a guide to real patterns in linguistic data. To square the invariance required to address the pessimistic meta-induction with the indispensability aspect of grammars as compression algorithms, Nefdt opts to supplement the invariance approach with a multiple models approach.

The idealization Nefdt has in mind is grammars as scientific models rather than theories. The key to the multiple models approach is that there is no expectation of a single best model. Each model focuses on some aspect of the target system. The position is pluralist in spirit though not ontological commitment, as structural realism can underpin an ontological unity at the meta-theoretical level. Models pick out indispensable real patterns and relate to each other via the underlying target structure, which remains invariant under theory change and thereby explains success and progress in linguistics.

Most work in philosophy of linguistics focuses on syntax and Nefdt is no exception here. However, the supplemented structural realist position he advances allows him to extend his view to include phonology and semantics (chapter 8). Following Jackendoff’s (2013) insight that these are autonomous generative components of language, Nefdt can include models of phonology and semantics along with syntactic models in the multiple models approach, doing more justice to the complexity of the underlying structure that is language. Nefdt adopts notions of gradients of abstraction (on par with Dennett’s stances) and levels of abstraction from Floridi (2008) to understand formally how the components may interface. Although the details are beyond the scope of this review, one example is that an interface between phonology and semantics that bypasses syntax can explain pidgins and creoles. Importantly for Nefdt’s overall position, the framework allows for different perspectives on the same system. And, more broadly, multiple disciplines of cognitive science, including linguistics, share a common argument pattern of using graph-theoretic models to discover invariant cognitive structures (chapter 9). Returning to the worry of underdetermination and how to specify non-redundancy, perhaps constraints on structure from other disciplines can narrow the set of weakly equivalent grammars to those whose features capture real patterns of a target language, consistent with psychological evidence about native speakers. In any case, the intersection between disciplines constitutes the core of the field and in structural terms linguistics sits in this intersection. Seen in this light, linguistics should re-take its central historical place in cognitive science, according to Nefdt.

Nefdt appropriately subtitles his book ‘A Journey into the Philosophy of Linguistics’. In addition to thoroughly traversing developments in syntax, Nefdt offers many side excursions not normally encountered in the study of linguistics, such as tours through graph theory, group theory, systems biology, and many more. His primary use of these many and varied disciplines is by way of analogy, to show how his points about structure work elsewhere and could apply to language. In many cases the analogies are more suggestive than definitive, but while the project is ambitious, its goal is modest: to make the case that ‘structural realism, buttressed with a scientific modelling perspective, is a viable foundation for the philosophy of linguistics’ (p. 145). On these terms, the book is a great success, introducing a fresh perspective on the study of language that future researchers will be compelled to consider.

Christopher Viger
University of Western Ontario
cviger@uwo.ca

References

Dennett, D. (1991). ‘Real Patterns’, Journal of Philosophy, 88, pp. 27­–51.

Floridi, L. (2008). ‘The Method of Levels of Abstraction’, Minds and Machines, 18, pp. 303–29.

Jackendoff, R. (2013). ‘Constructions in the Parallel Architecture’, in T. Hoffmann and G. Trousdale (eds), The Oxford Handbook of Construction Grammar, Oxford: Oxford University Press.

Katz, J. J. and Postal, P. M. (1991). ‘Realism vs. Conceptualism in Linguistics’, Linguistics and Philosophy, 14, pp. 515–54.

Ladyman, J. (1998). ‘What Is Structural Realism?’, Studies in the History and Philosophy of Science, 29, pp. 403–24.

Ladyman, J. and Ross, D. (2007). Everything Must Go: Naturalized Metaphysics, Oxford University Press.

MacDonald, C. and MacDonald, G. (1995). Connectionism: Debates on Psychology Explanation 2, Blackwell.

Santana, C. (2016). ‘What Is Language?’, Ergo, 3, pp. 501–23.

Stabler, E. and Keenan, E. (2003). ‘Structural Similarity within and among Languages’, Theoretical Computer Science, 293, pp. 345–63.

Stainton, R. (2014). ‘Philosophy of Linguistics’, Oxford Handbooks Online, Oxford University Press, available at doi.org/10.1093/oxfordhb/9780199935314.013.002.