University of Sydney and Laboratoire d’histoire des théories linguistiques, Université Paris-Diderot
Few questions in linguistics can be as hoary, fundamental or, perhaps, as unsatisfactorily handled, as that of the discipline’s empirical status – a question typically presented as one of linguistics’ ‘scientificity’. Among the many issues needing attention from anyone who wants to make a serious epistemological effort to clarify the character of linguistic theory, one in particular has received strikingly little discussion: the presence (or nonpresence) in linguistics of measurement.
Measurement couldn’t be more central to canonical sciences: theories are characteristically formulated in mathematical terms, and contain hypotheses about quantified data, a situation which naturally presupposes the measurement of the base phenomena (see Kuhn 1961 for a fascinating explosion of ‘myths’ about measurement in physics). In linguistics – or, at least, in the core theoretical domains of phonology, morphology, syntax and semantics – measurement comparable to that observed in the empirical sciences plays no obvious role. That, presumably, is the reason for the ambient silence about the topic. But could there be some more subtle respect in which theoretical linguistics does, after all, involve something analogous to measurement?
This question isn’t without interest, since it forms part of the comparatively neglected methodological side of the question of the scientificity of linguistics – taking ‘science’, of course, in its typical English sense, and not in the broader sense captured by German Wissenschaft. As everyone now knows, there’s no straightforward criterion of the ‘scientific’: just what qualifies something as a science, in fact, is – fortunately – a subject of debate (see the useful Stanford Encylopedia of Philosophy entry). One thing, though, should be clear: it’s not enough for a discipline to count as a science that it simply have an empirical object. This, however, is typically the implicit grounds of linguists’ protestations about the scientificity of their discipline: languages are empirical objects, and the linguist studies them in the same way that other scientists study other empirical objects.
But, in that formula, it’s the second half – in the same way as other scientists – that really counts. Scientificity, whatever that is, resides as much in the methodological question of how the object is studied, as in the ontological question of what kind of object the science has. Plenty of non-sciences have empirical objects: history is the obvious example. Even literary criticism studies empirical objects – works of literature. As well as having an empirical object, the canonical sciences also have to be able to model their objects determinately (modelling being the subject of this conference next year). And modelling is something that best proceeds mathematically.
In a very interesting paper from 2007 – one of the only ones to address the question directly – Kent Johnson claims that theoretical linguistics also involves ‘measurements’. Linguistic measurements, however, ‘typically are not quantitative, but instead concern such things as the grammaticality or acceptability of a sentence, its sound and meaning, etc’ (p. 384). For Johnson, you are measuring linguistic properties of a sentence when, for example, you categorize its subject NP as an Agent.
Johnson claims (p. 367), in fact, that there is a ‘remarkably tight point-by-point agreement between the relevant aspects of linguistic methods and the underlying logic of the other sciences,’ and he sets about trying to show it in his article. If he is successful, this would be a remarkable result. The scientificity of post-Saussurean linguistics has been a topic of continuing and unresolved methodological and philosophical dispute for many years. Given the enduring controversy over the question, Johnson’s claim to discern a point-by-point agreement in the logic of linguistic and scientific explanation would stand as a major achievement in itself.
In particular, however, any attempt to argue that judgements of grammaticality or of semantic features like Agency are measurements has to reckon with two major disanalogies with the quantifiable measurements of the empirical physical sciences. The disanalogies concern (a) the determinacy of the measurements in linguistics vs canonical sciences and (b) the measurements’ openness to mathematical manipulation. Let’s consider each.
(a) Determinacy. In empirical science, the utility of measurements of quantified variables derives from their determinacy and hence replicability. The measurements yielded by observation and experiment in, for instance, physics, need to be determinate (within, of course, some agreed margin of error and agreed degree of rounding) in order for them to be replicated by other scientists: you need to know what the experimental predictions of a theory are, and what range of values constitutes confirmation. This is also what makes technological applications of theories possible. As pointed out by Wigner (1960: 5), machines like radars and nuclear reactors are ones where ‘the physicist creates a situation in which all the relevant coordinates are known so that the behavior of the machine can be predicted’. For this to be possible, the relevant variables have to be measured determinately.
In science (barring certain well known phenomena, irrelevant here) a given act of measurement typically yields a definite reading (of either a precise number to some degree of accuracy, or a range), and this reading should not vary with the instrument making the measurement. Furthermore, it is typically not unclear what reading a measuring instrument gives for measurements within its standard range of operation. In linguistics, however, judgements of grammaticality and meaning notoriously aren’t determinate in this way, but vary both within and between subjects. Semanticists and syntacticians will probably never agree on the correct characterization of meaning; certainly, the history of the debate on theta-roles, on lexical aspect and on any analysis in lexical semantics suggests exactly that.
Nor would this seem to be a feature of the youth of these subfields. It doesn’t seem plausible to suggest that the kind of disagreement among serious investigators that presently characterizes these areas (see for instance Sasse 2002, still surely relevant) is just a product of theoretical immaturity. No one should hold their breath in the hope that some invisible disciplinary hand will gradually shape a consensus in semantics about the correct representation of ‘facts’ about meaning in a way that will be able to resist critiques in the way that the best established theories of physics can. The study of meaning just isn’t like that, and as soon as semanticists come to terms with the fundamentally hermeneutic character of their discipline, the better (see Rastier 2009).
(b) Mathematical manipulability. In science, measurability (quantifiability) is what opens phenomena to the possibilities of mathematical manipulation. It’s a surprising fact about the physical world that it’s able to be represented mathematically, and that the results of formal mathematical operations can be used to predict empirical conditions. As a self-contained deductive system, mathematical reasoning allows formal, rule-governed operations on mathematical symbols; this enables the scientist, in the words of Dirac (2003 : 102), ‘to infer results about experiments which have not been performed’. Mathematical representation allows the discovery of latent structures and dependencies which couldn’t otherwise be brought to light, and enables predictions to be made on their basis which would have been unavailable otherwise. The value of measurability in science, then, lies in the fact that measurable physical quantities can be manipulated by the formal procedures of mathematics. Without mathematical manipulability, measurability is of no inherent scientific interest. The star system used to assess hotels or to rate films operates with ‘measurements’ in a naive sense. But these aren’t open to any interesting level of mathematical manipulation, and it would be a mistake to claim that these rating systems are measurements in any scientifically interesting sense. Likewise, the ‘measurements’ of meaning made in semantics are precisely unable to be manipulated mathematically. There are no known mathematical operations which can be used to model thematic roles, Jackendovian or Wierzbickian semantic primitives, and so on.
So, if linguistics doesn’t involve measurement, it’d seem to lack a key feature of the very empirical sciences it’s so keen to lay epistemological claim to. I’ve tried to explore some of the consequences of the non-scientificity of semantics elsewhere (the last chapter of my 2005). In particular, it’s important to stress that non-scientific – whatever that might mean – certainly doesn’t mean irrational or without foundation. Appreciating that fact alone would go a long way to raising the epistemological self-consciousness, and to banishing the scientistic false-consciouness, that are arguably widespread in the discipline.
There’s a tendency for researchers to dismiss longstanding debates on the grounds that they are stale and out of date. On the question of the ‘scientificity’ of linguistics – or, rather, its position in between the natural and the human sciences – that would be a mistake. If a question is longstanding, that means it’s unresolved. Assuming that it’s not fundamentally malposed, that suggests that more work needs to be done. Clarifying the epistemological status of linguistic theories, and plotting linguistics’ consequent coordinates in the geography of knowledge, raises many unresolved questions, but is still the most necessary task in the philosophy of the discipline.
Dirac, P. 2003 . The relation between mathematics and physics. Resonance, August 2003, 102–110. Originally in Proceedings of the Royal Society (Edinburgh) 59 (1938-39), 122–129.
Johnson, K. 2007. The Legacy of Methodological Dualism. Mind & Language 22: 366–401.
Kuhn, T. 1961. The Function of Measurement in Modern Physical Science. Isis 52: 161-193
Rastier, F. 2009. Sémantique interprétative. 3ed. Paris: PUF.
Riemer, N. 2005. The semantics of polysemy. Reading meaning in English and Warlpiri. Berlin and New York : Mouton.
Sasse, H.-J., 2002. Recent activity in the theory of aspect: accomplishments, achievements, or just non-progressive state? Linguistic Typology 6: 199–271.
Wigner, E.P. 1960. The Unreasonable Effectiveness of Mathematics in the Natural Sciences. Communications On Pure And Applied Mathematics 13: 1-14
How to cite this post
Riemer, Nick. 2013. ‘Measurement and unobservables in linguistic theory’. History and Philosophy of the Language Sciences. https://hiphilangsci.net/2013/07/17/do-linguists-measure-anything