Book announced at https://linguistlist.org/issues/35-1498

Title: The Philosophy of Theoretical Linguistics
Subtitle: A Contemporary Outlook
Publication Year: 2024

Publisher: Cambridge University Press
http://www.cambridge.org/linguistics
Book URL: https://doi.org/10.1017/9781009082853

Author(s): Ryan M. Nefdt

Reviewer: Keith Begley

SUMMARY

The author, Nefdt (hereafter N), identifies as his target audience “advanced students of either philosophy or linguistics and experienced practitioners at the intersection between these fields” (p. x). N’s “goal is to provide not only a songbird’s-eye view of the interconnections between different subdisciplines and frameworks of linguistic theory but to showcase common problems and present novel analyses of the study of language that only a contemporary philosophical overview can offer” (p. ix). The book has xi + 231 pages, beginning with a short preface, and has an index. It is divided into eight main chapters, each addressing a different subfield or area, and a two-page concluding chapter. N says that “Each chapter can be considered a sustained, self-contained argument based on the subfield at hand” (p. ix). Each includes a helpful ‘Further Reading’ section with plentiful annotation.

In the introductory chapter, N asks what theoretical linguistics is. He tells us that it is identified by a guiding set of core questions:

“1a. What is Language?
b. What is a language?
2. How do we acquire languages?
3. How is linguistic communication possible?
4. How did language evolve?” (p. 3)

Answering these he takes to be neither necessary nor sufficient, although “To be a theoretical linguist, of whichever variety, you have to attempt to answer some, if not all, of these questions in a coherent manner” (p. 3). N rightly notes the logical priority of the initial questions over the others. Despite this, the core question approach is not developed much further.

The chapter also treats of grammaticality and formal approaches. I wonder about the wisdom of including only cursory descriptions of finite automata, etc., in view of the advanced target audience. At the same time, the question regarding what these “machines” are goes overlooked. After recalling early findings that natural languages outstrip regular languages and finite automata, N asks: “So why don’t we go for the most expressive grammars?” (p. 12). He gives the internally correct answer that it is because they would outstrip the grammar of natural language. However, any reasonable version of a computational theory of a physically dependent mind will recognise that it could not possibly be an unrestricted Turing machine (cf. p. 10), because there are no implementations of such machines, only of linear bounded automata or their equivalents. It is a truth rarely spoken, but one I would have expected to come out in this philosophical monograph.

N asks “if all the mathematical facts about formal languages were settled, what would be left of linguistics? To hard-line platonists, the answer might be nothing” (pp. 13–14). N neglects to say who these ‘hard-line platonists’ are, or why they would answer this way, but he refers to Katz and Postal in the previous paragraph. Katz at least does not hold this view (Begley 2023, pp. 20, 24, 26). More generally, it is a fundamental misunderstanding of Platonic enquiry to think that it must be in no way ‘empirical’ (Politis 2021, pp. 46–48). There are famous dialogues written about this, with footnotes. Here, as elsewhere, N refers mainly to Katz’ first book on realism in linguistics (1981), often without even mentioning his later books (especially 1990 & 1998). If the reader seeks a remedy for this absence, I shamelessly recommend my own recent treatment (Begley 2023).

In Chapter 2, entitled “What Is a Possible Human Language?”, N asks what he calls “a big-picture question that’s generally gone unexplored: what is a possible human or natural language?” (p. 23). N wishes to rule-out Language being an “a priori abstract object”, preferring to see it as “clearly a scientific abstraction from particular languages, just like […] other scientific objects are” (p. 30). However, without further argumentation, this is just to beg the question regarding those objects, as there is a broader and continuing debate in the metaphysics of science (e.g., Mumford & Tugby (eds) 2013; Berman 2020; Tugby 2022). N remarks that ontological debates are largely absent from science, a point that should be of no surprise to philosophers. He attempts an analogy: “No one in the cognitive neuroscience of memory asks how individual tokens or memories are related to some platonistic concept of Memory, divorced from its instantiations.” (p. 30: n . 12). As it happens, even Plato himself didn’t ask this, so it is uncertain what if anything it shows. Again, this is not how that kind of enquiry proceeds.

N quotes Devitt’s definition of language as being one that “lacks overt ontological commitments” (p. 31). The reader must forgive me for over time having grown highly suspicious of arguments from ontological detente. The definition includes the claim that language is a “system of representations or symbols that is constituted by a set of governing rules” (Devitt 2013, p. 95). It is not at all clear, to me at least, why we should take claims that something is representational or constituted by rules as being in any way ontologically neutral and uncommitting.

N takes “A Modal Approach to Linguistic Possibility”, by which he means a formal one. I found this to be a very surface-level introduction, with some inaccuracy. At one point, N even confuses possibility with contingency: “Possible claims (indicated by ‘◊’) are only true in some possible worlds and false (or have no truth value, depending on your view of vacuous truth) in others” (p. 41). N also equivocates on ‘possible’, using both a gradable notion (cf. “less possible”, “very possible”, p. 41) and a formal non-gradable one. He later cashes out the loose gradable notion for that of the remoteness of a possibility and pins the bill on Berto & Jago (2019). However, although a “more remote possibility” might be less probable, surely it isn’t “less possible” (p. 43). It is also somewhat of a mystery what dialectical purpose this served, if any.

N approaches the notion of a possible language through S5 modal logic. That’s the intuitive Leibnizian one with all the main axioms and whistles. However, no further effort is made to argue for S5 over any other modal logic. This is unusual for a purported work of philosophical logic. That is, a work that, so to speak, doesn’t just tell us about whistles, etc., but blows the right ones at the right times and advocates for each usage to be considered appropriate.

The aim of Chapter 3, entitled “Syntactic Metatheory”, is to “identify the core of syntactic explanation in theoretical linguistics, and to trace this conception across distinct approaches” (p. 51). N sets out a [BASIC] conception of syntactic theory comprising, roughly speaking, [RULE-BOUND], rules for the formation of compounds and constituents, [AoS], the independence and autonomy of syntax, and [RECURSION]. N then shows how each of these definitions must be modified or respected to accommodate modern minimalist and further approaches, including even radical construction grammar, which he argues can be reconciled with this picture. This is largely well executed. However, at one point a defined name [RECURSION]′ is mentioned prior to its definition on the following page (pp. 58–59).

Chapter 4 is entitled “The Science of Semantics”. N begins by distinguishing between metasemantics-1 and metasemantics-2, before proceeding only with the latter. Metasemantics-1 is “a grounding project” that provides the nonsemantic facts in virtue of which a semantic fact pertains (pp. 81–82). That is, a metaphysical project of more or less depth. Metasemantics-2 is “the metascientific study of the science of meaning” (p. 84), including:

“methodological issues such as whether semantic value is best modelled as truth conditions, dynamic context-change potentials, or numerical vectors representing collocational data, questions about the interface between syntactic structure and semantic interpretation, and the distinction between semantic and pragmatic information.” (p. 84)

N allows a role for metaphysics “only in so far as it investigates the repercussions of scientific theory” (p. 84). He argues that the resolution of metaphysical debates are “unattainable goals” (p. 85). N here casts aside most of the history of philosophy in the space of two or three pages. To my mind, such a position is far too deflationary, if not anti-philosophical. The procedure of distinguishing two kinds of metasemantics while ignoring their intersection and interaction is not adequate philosophically or otherwise. Metaphysical questions (never mind this or that metaphysical approach) cannot be wished away or deferred by fiat; they are compulsory. For example, nothing in N’s argument compels us to take “best”, in the question regarding how semantic value is to be best modelled, in a methodological sense that excludes metaphysical enquiry, nor should it.

N lists some work from the last 20 years that eschews truth conditions in explanations of semantic phenomena (p. 94), but of course the history is much longer than that. After all, Lewis’ oft-quoted remark, “Semantics with no treatment of truth conditions is not semantics” (Lewis 1970, p. 18; quoted at p. 82), was originally directed against an early version of Katz’ theory employing semantic markers (Katz & Postal 1964; see further references to Katz below, and Begley 2021, pp. 318–20, for some discussion).

N discusses formal semantics, dynamic semantics, and distributional semantics including vector-space representation. There is a curious shift here at one point from speaking about representations of meanings to speaking about meanings simpliciter, for example, “word meaning is represented as a vector” to “Meaning is extracted from patterns in corpora” (p. 97). Similarly, on the next page, very little interrogation is provided of the statement “Meanings are mathematical objects, in this case vectors” (p. 98). N says nothing here about what he thinks a mathematical object is, and previously ‘vectors’ were described as being representations. This is the kind of trouble we invite when unconcerned by metaphysical matters. The things we call ‘vectors’ in computers, etc., are data structures, perhaps acting as representations or surrogates for mathematical objects, which might also be called ‘vectors’.

N also discusses what he calls “rogue semantics”, which involves “violations of a particular interpretation of compositionality or the syntax-semantics interface” (p. 99). For instance, he mentions some views that go against the assumption that “words are semantically basic in some sense” (p. 99). However, this is not as new a trend as N makes out. Sub-word level decompositional semantics was famously discussed as early as Katz and Fodor (1963) and developed further especially by Katz (1972 & 1977), and prior to that in the form of componential analysis (Goodenough 1956; Leech 1969). There is of course an even longer philosophical history to the idea. I would have expected some of that history to be at least mentioned here. N recognises at least that “In another sense, lexical compositional semantics is very conformist. In fact, it’s compositionality all the way down!” (p. 102).

There is also a two-page discussion of “supersemantics”, that is, the extension of formal semantics to nonlinguistic phenomena such as “music semantics, dance semantics, primate grammar, gesture, and many others” (p. 104). N merely urges caution here and suggests that a “super-metasemantics” would be required to go along with it, but without saying anything at all about what that would entail.

Chapter 5 is entitled “Context and Pragmatics”. N compares the problem of the semantics-pragmatics distinction to demarcation in philosophy of science. N draws upon Lewis, Grice, and Stalnaker’s work to outline three “metapragmatic” possibilities, the indexicalist, cognitivist, and the socio-inferential views, respectively.

Chapter 6 is entitled “Signs, Sounds, Action!”. N complains that philosophers have paid phonology surprisingly little attention. In response, N offers “only a sketch of an action-theoretic account of phonology (and language) that takes both internal models and motor systems into consideration” (p. 155).

Chapter 7 is entitled “Computational Approaches to Language”. N argues that the priority of explanation over prediction in linguistics, and science more generally, is a post-positivist overcorrection to the positivists’ focus on prediction of behaviour. This is manifest, he believes, in the ‘Galilean’ paradigm of explanation and the competence-performance distinction.

“No matter how much performance data is amassed to contradict certain rules or principles posited by the theory, a linguist can merely evoke the distinction and relegate the recalcitrant data to performance theory (presumably beyond the remit of a tractable linguistic science)” (pp. 164–165).

N presents the early approach to quantum mechanics as a counterexample to the claim that explanation is prioritised over data in science generally (p. 166). However, the example is somewhat self-defeating as that is only one way of thinking about the priority in this case. One might nonetheless aim for the ideal of explanation and understanding, although one temporarily settles for shutting up and calculating if that is all one can do.

N also addresses the divide between discreet and continuous methods, which encompasses “the central question” of the chapter, namely, “what does [Deep Learning] have to offer theoretical linguistics?” (p. 171). Unfortunately, N’s introduction to artificial neural networks and machine learning contains infelicities, including in the sentence: “The general process uses large numbers of computational units called neural networks that take an input signal and feed it into a set of input nodes (or neurons) represented as numerical values, specifically, vectors” (p. 173). I take it this was meant to refer to computational units called ‘neurons’, for which the input is a vector (comprising a value for each node). N then quotes Linzen & Baroni (2021, p. 196): “Each of these units [neurons] calculates a weighted average of its inputs; this weighted average is then passed as an input to a simple nonlinear function, such as the sigmoid.” There are two issues with this. First, perhaps conflating it with a McColloch-Pitts neuron, N adds that there will be a threshold at which the node becomes active. However, sigmoids are precisely continuous functions without any threshold. Second, Linzen & Baroni have made an error; “weighted average” should say ‘weighted sum’. This is also evident from the continuation of the passage: “In other words, the function computed by each unit is σ(w1x1 +··· + wnxn)” (ibid., p. 196; in the original, each w and x is followed by a subscript). N says that a neural network “corrects itself based on the expected value of the output. This is what’s known as backpropagation” (p. 173). This is half the story. N doesn’t mention gradient descent. It is in fact possible (albeit less probable) to do gradient descent without backpropagation. Also, instead of ‘convolutional networks’, N refers to “convoluted networks” (though they may well appear to be).

In his discussion of epistemic opacity, N expresses what I take to be a ubiquitous misunderstanding of the black-box problem. N claims that the problem is that “accessing their inner workings is sometimes impossible” (p. 179). However, this is generally not a problem with respect to a trained model. A printout can be given of all the parameters, and thereby the matrix operations they constitute, for any output of a model on any input. The problem is that we often lack an explanation for the meaning of the output in terms of the domain from which the data arose. N then makes another overused appeal, comparing the lack of transparency in neural networks with that of our brains. However, the situations are actually distinct. I can succeed verbally in providing you with my reasons given in terms of the domain of interest, but you can (currently) only image my brain and surmise its activity. That is, you don’t get the printout of its parameters, so to speak.

In Chapter 8, entitled “Language and Evolution”, N aims to “turn the minimalist research agenda on its head with an alternative thesis: natural language is a complex system and its emergence is likely to have been prompted by multiple interacting factors” (p. 183). This he calls the “Maximalist Programme”. He discusses the minimalist hypothesis that the Merge operation arose from a single genetic macro mutation around 100,000 years ago, and considers various counterclaims suggesting a longer and more gradualist timeline that “opens us up to the standard resources of evolutionary biology” (p. 187). N argues for what he calls ‘systems biolinguistics’, which draws upon systems biology and complexity science, and takes a holistic and interdisciplinary approach (p. 193). He claims that “these (graph-theoretic) tools aren’t merely instruments but tell us something ontologically important” (p. 195). This is a strong claim, which doesn’t get fully substantiated here.

EVALUATION

I have already made known some of my evaluations and disagreements regarding specific points, and philosophical approaches, etc., though it is likely that different readers will find challenging elements in other areas of the book also. There is simply a lot covered in a short space, which sometimes leads to a problem with pacing. The audience is expected to be advanced or experienced, but sometimes the introductions are basic and assume no prior knowledge, sometimes very much is assumed (indeed, with lots of Further Reading suggested). So, the book somewhat falls between two stools, neither being an introductory textbook nor a cohesive advanced monograph. Another one hundred pages might have been needed to rectify this, or some more radical intervention, as there are presumably very few readers who are experts in everything discussed.

One might be hopeful upon reading the introduction that the privileging of an ordered list of questions would help to structure and unite the whole, but this doesn’t really come to pass. Some chapters have a clear line of argument pushed throughout, while others seem merely to expound a number of approaches and the line is less clear. While a few interconnections are made, the book lacks an architectonic structure. That having been said, each chapter considered on its own provides much food for thought. There is something for everyone concerned whether they be linguist or philosopher of whatever specialty or denomination, though they will need to be open to much more to make it through the whole book.

REFERENCES

Begley, K. 2021. ‘Atomism and Semantics in the Philosophy of Jerrold Katz’, In: U. Zilioli (ed.), Atomism in Philosophy: A History from Antiquity to the Present. Bloomsbury Academic, 312–330. https://www.doi.org/10.5040/9781350107526.0026

Begley, K. 2023. ‘Katz Got Your Tongue? The Metaphysics of Words’, Topical Collection: The Philosophy of Words, Synthese 202(4), 107: 1–29. https://doi.org/10.1007/s11229-023-04324-x

Berman, S. 2020. Platonism and the Objects of Science. Bloomsbury.

Berto, F., & Jago, M. 2019. Impossible Worlds. Oxford University Press.

Devitt, M. 2013. “The ‘linguistic conception’ of grammars”. Filozofia Nauki Rok XXI: 2(82).

Goodenough, W. H. 1956. ‘Componential analysis and the study of meaning’. Language, 32, 195–216. https://doi.org/10.2307/410665

Katz, J. J. 1972. Semantic theory. Harper & Row.

Katz, J. J. 1977. Propositional Structure and Illocutionary Force. The Harvester Press.

Katz, J. J. 1981. Language and other abstract objects. Rowman & Littlefield.

Katz, J. J. 1990. The metaphysics of meaning. MIT press.

Katz, J. J. 1998. Realistic rationalism. MIT Press.

Katz, J. J. & Fodor, J. A. 1963. ‘The structure of a semantic theory’, Language 39(2): 170–210.

Katz, J. J. & Postal, P. 1964. An Integrated Theory of Linguistic Descriptions, M.I.T. Press.

Leech, G. N. 1969. Towards a Semantic Description of English. Longmans, Green, and Co.

Lewis, D. 1970. ‘General semantics’. Synthese 22, 18–67. https://doi.org/10.1007/BF00413598

Linzen, T., & Baroni, M. 2021. ‘Syntactic structure from deep learning’. Annual Review of Linguistics 7: 195–212. https://doi.org/10.1146/annurev-linguistics-032020-051035

Mumford, S., & Tugby, M. (eds) 2013. Metaphysics and Science. Oxford University Press. https://doi.org/10.1093/acprof:oso/9780199674527.001.0001

Politis, V. 2021. Plato’s Essentialism: Reinterpreting the Theory of Forms. Cambridge University Press.

Tugby, M. 2022. Putting Properties First: A Platonic Metaphysics for Natural Modality. Oxford University Press.

Editor's note: A response to this review by the publication author has been posted and can be accessed here: https://linguistlist.org/issues/36-1239.

ABOUT THE REVIEWER

Keith Begley is an Honorary Associate of the Philosophy Department at Durham University. He does research in philosophy of computer science and artificial intelligence; philosophy of language and linguistics, especially the work of Jerrold J. Katz; history of philosophy, especially ancient (Heraclitus) and analytic philosophy (Wittgenstein); and computational philology, applying computer programs to parse ancient Greek texts. He holds an M.A. (Dubl.) and a Ph.D. in philosophy from Trinity College Dublin (TCD), The University of Dublin (2016), and a H.Dip. (2021) and an M.Sc. (2022) in computer science from University College Dublin (UCD), The National University of Ireland, Dublin. He has held positions including Teaching Fellow in Philosophy at Durham University (2023/24); Assistant Lecturer in Philosophy at Maynooth University (2022/23); Demonstrator in Computer Science at UCD (2021/22); and Adjunct Assistant Professor of Philosophy at TCD (2017–2020).