Keywords

1 Introduction

On March 11, 2020, the World Health Organization (WHO) declared the COVID-19 epidemic rapidly spreading from China to most other countries in the world a “pandemic.” A month later, on April 16, that same organization published International Guidelines for Certification and Classification (Coding) of Covid-19 as Cause of Death, Based on ICD: International Statistical Classification of Diseases. At that time, four months into the deadly first wave of infections, comparability of health and mortality data across all the affected countries became a key concern, as different countries seemed to be reporting and discussing different things. As a body mandated to protect international public health via, among other measures, a uniform classification of diseases, the WHO formulated the following “definition for deaths due to COVID-19”:

A death due to COVID-19 is defined for surveillance purposes as a death resulting from a clinically compatible illness, in a probable or confirmed COVID-19 case, unless there is a clear alternative cause of death that cannot be related to COVID disease (e.g. trauma). There should be no period of complete recovery from COVID-19 between illness and death.

A death due to COVID-19 may not be attributed to another disease (e.g. cancer) and should be counted independently of preexisting conditions that are suspected of triggering a severe course of COVID-19. (“International Guidelines”, p. 3)

We will return to these guidelines for further analysis in Sect. 2.3, but what is immediately striking about them is that they mix substantive and linguistic concerns to a puzzling effect. On the one hand, WHO is discussing and organizing substantive medical knowledge over “cause[s] of death” in “probable or confirmed COVID-19 case[s]” where, one would assume, the weight of scientific evidence is decisive. On the other hand, the organization presents its main results as a “definition” and “classification”, which are two paradigmatic devices for metalinguistic and conceptual work. And this conceptual work is of paramount importance: “probable” COVID-19 cases are treated on a par with “confirmed” cases, and “independent” attribution of COVID-19 deaths is mandated even if other “preexisting conditions” such as cancer might have contributed to COVID-19 being severe enough to actually cause death. As is clear across the WHO’s document and in the broader debate over the issue, these are neither scientifically determined nor arbitrary conceptual choices. Instead, in the cases we discuss below, reasonable even if characteristically inconclusive arguments are given to justify any such choice.

In this contribution, we explore the plausibility and consequences of treating such arguments as metalinguistic arguments. While unquestionably related to the epidemiological and public health issues, these arguments are also arguments about how a term should be used. As such, they touch upon some of the foundational issues in meta-semantics, discussed in the recent literature on metalinguistic negotiations, conceptual ethics, and conceptual engineering (Burgess & Plunkett, 2013; Burgess et al., 2020; Cappelen, 2018; Plunkett, 2015; Plunkett & Sundell, 2013, 2021). Against this background, we analyze in particular how in the debate over what a COVID-19 death is, epistemic and practical reasons are intertwined in nuanced and complex ways to produce an interesting type of metalinguistic interventions.

We proceed as follows. In Sect. 2.2 we provide the theoretical basis for our analysis. We introduce the phenomenon of what we summarily call metalinguistic interventions, present their three key features particularly relevant to our case, and offer distinctions instrumental in grasping the rather non-standard type of metalinguistic interventions related to “COVID-19 death.” In Sect. 2.3, we analyze official statements (of WHO, national governments) and media reports to critically reconstruct the metalinguistic elements of the dispute in terms of prevailing forms of argumentation used. In Sect. 2.4, we discuss this analysis by developing two theoretically relevant points. First, the metalinguistic arguments revealed are inextricably linked to substantive, scientific issues and are partly determined by the imperfect character of our epistemic position on the subject. Second, they work in the service of broader practical arguments whereby scientific results are weighted against broader public policy values. We close by arguing that, in these ways, public metalinguistic arguments, while being a class of their own in need of precise analysis (see also Ludlow, 2014; Pruś, 2021; Schiappa, 2003), are of key importance to broader public debates and should be recognized as such.

2 Metalinguistic interventions

Metalinguistic uses of language have long been recognized as part and parcel of our communication. Perhaps most famously, Horn (1985) identified the mechanisms of “metalinguistic negation”, a form of negation that is not a logical operator on truth-conditional propositions, but rather an objection to previous uses of language perceived as erroneous or infelicitous on grounds ranging from prosodic to conceptual. A good example of conceptual metalinguistic negation marked one of the twists in the public discourse over the COVID-19 pandemic. On September 26, 2020, Richard Horton, the editor-in-chief of The Lancet, one of the medical journals publishing peer-reviewed research instrumental in the scientific understanding of COVID-19, published a commentary (Horton, 2020) entitled:

  1. (1)

    COVID-19 is not a pandemic.

This title, taken out of context, has risked becoming a viral sensation for the negationist argument,Footnote 1 thus turning Horton’s well-intentioned conceptual “precisation”Footnote 2 into a perilous slogan for a standpoint he vehemently opposes. But it takes only about 2 min to realize Horton’s argument was impeccably metalinguistic:

  1. (1a)

    COVID-19 is not a pandemic. It is a syndemic. […] The notion of a syndemic […] reveals biological and social interactions that are important for prognosis, treatment, and health policy. Limiting the harm caused by SARS-CoV-2 will demand far greater attention to NCDs [non-communicable diseases] and socioeconomic inequality than has hitherto been admitted. (Horton, 2020).Footnote 3

As is clear in (1a), Horton’s argument for conceptual shift from PANDEMIC to SYNDEMIC is justified on two grounds: scientific precision and public health response, with the latter taking the upper hand.Footnote 4 We will return to this interrelation of epistemic and practical arguments in our discussion below.

Such reasoned metalinguistic negations are, in our view, but one species of the argumentative and linguistic mechanisms that underlie public discussions where metalinguistic intervention (MI) plays a key role.Footnote 5 Attention to MI, encompassing various forms of reflection, discussion, and action on meanings, has been growing notably in recent analytic philosophy under various labels: ameliorative analysis (Haslanger, 2012), conceptual engineering (Cappelen, 2018), conceptual ethics (Burgess & Plunkett, 2013), meaning litigation (Ludlow, 2014), metalinguistic negotiations (Plunkett, 2015; Plunkett & Sundell, 2013, 2021), verbal disputes (Chalmers, 2011). While rooted in classic debates over the possibility of revisionary and pluralist approaches to meaning (Carnap, Quine, Davidson, Kripke, Putnam, Burge), this reinvigorated attention brings a new sense of relevance and urgency, as well as new methods, to the philosophical study of public uses of language. Lively theoretical disputes over the semantic/pragmatic nature of MIs, their metasemantic underpinnings, speakers’ control over meaning, social and political functions of MIs, their potential for amelioration or perversion of meaning, permeate this literature (Burgess et al., 2020; Marques & Wikforss, 2020). Still, the idea that MIs are often worthwhile and even central to public discussions is widely shared (see, however, Marques, 2017 and Stojanovic, 2012 for limitations).

An obvious objection to our approach would be to see the discussion over “pandemic” and “COVID-19 deaths” as basically a scientific dispute over facts. At the stage where the dispute takes place, we only have adequate epistemic access to a small fraction of the facts; we disagree about the rest because we infer different things about that rest based on the little knowledge we do share. For instance, in the case of COVID-19 deaths, the dispute revolves around different methodologies for calculating numbers of fatalities under fragmentary information, whereby full-proof medical evidence as to the causes of death of the thousands of suspected cases is missing. As a result, there is nothing metalinguistic patently involved just yet: after all, one of the defining characteristics of MIs is that disputants possess and mutually agree on all the relevant facts, and yet they disagree in virtue of the incompatible conceptual views they advocate on normative grounds (Ludlow, 2014; Plunkett, 2015; Plunkett & Sundell, 2013, 2021; Schiappa, 2003). They thus fix their beliefs, while trying to solve for the meaning. Accordingly, this objection would maintain that until any forthcoming empirical facts might be decisive in adjudicating the dispute, it is essentially a substantive, ground-level dispute.

This objection can be resisted on two grounds. First, it assumes that there is, eventually, the scientific truth of the matter on what a COVID-19 death is, and that the problem lies in the scarce resources and underdeveloped methods to arrive at that truth (e.g., precise and massive tests and autopsies). But this assumption can be legitimately challenged: multiple notions and conceptions of cause and, more specifically, cause of death have played a role in various scientific and medical contexts (Clarke & Russo, 2016; Lindahl, 1988, 2021; Reiss, 2016; Reiss & Ankeny, 2016). It is all but clear that any single one of these should or could be elected as the right or privileged one with which to form a univocal scientific concept of COVID-19 DEATH. Additionally, there is the issue of numerous particular cases of especially indeterminate nature, even within what seems to be a fixed framework.Footnote 6 Lindahl (2021) gives the example of situations of COVID-19 infection in patients with cancer, in which the two diseases “reciprocally interact, increasing the seriousness of the outcome” (2021, p. 72), thus rendering dubious the possibility of a clear choice of either morbidity as the underlying cause of death. Indeed, one can claim that “the cancer and the COVID-19 jointly initiated the train of morbid events leading directly to death” (Lindahl, 2021, p. 72, italics in the original), and given that only one can be reported on the death certificates, discretionary decisions need to be made by coroners. That’s where the guidelines such as the ones of WHO come to the rescue: complex situations of a rapidly spreading pandemic driven by a hitherto unknown virus are rife with uncertainty, indeterminacy, and certain arbitrariness of results that cannot be conclusively overcome by scientific means alone for the purposes of concerted public health response.

That brings us to the second argument against the objection. Even if the assumption of the scientific truth of the matter proved to be at least approximately adequate (perhaps with better diagnostic methods being developed and widely implemented), for our argument to take off the ground we do not need to resist this objection so far as the SCIENTIFIC CONCEPTFootnote 7 is concerned. Indeed, the objection can help us make clear that there are (at least) two concepts and two sets of issues converging and being conflated in these discussions. It does so by rendering it clear that the scientific concept and a set of related issues constitute just part of the concerns of health authorities when they discuss and issue operational definitions. On the other hand, at the same time, there is also a different concept—the INSTITUTIONAL CONCEPT—and a set of issues that are fully determined by institutional declarations. International and national health authorities, facing the need for urgent and decisive action under uncertainty, propose, discuss and establish uniform and operationally precise “definitions” and “classifications” which make it possible to overcome remaining uncertainties. It is the metalinguistic interventions on this second concept that we focus on here.

Our focus on institutional concepts as the domain of MIs over COVID-19 deaths is inspired by Searle’s social ontology (Searle, 1995, 2010).Footnote 8 Within this theoretical framework, by declaring a given epidemiological situation a “pandemic”, the WHO creates a new institutional reality in which various institutions and agents are endowed with new rights and obligations. For instance, we have the right to resort to the force majeure clause to cancel or alter our obligations and, simultaneously, we have the obligation to follow strict health-related regulations and limitations (e.g., travel bans). These conditions make up the declarative status of these acts. Declarations are precisely the speech acts that create new social realities by the very fact of being felicitously performed: a declaration of war by a legitimate head of state just starts the war, an official announcement of firing an employee by the employer just is firing him, etc. (Searle, 1975, 1995, 2010). All the same, there is a special sub-type of declarations that still create institutional facts but are grounded in some natural or social facts, namely, representative declarations (Searle, 1975, pp. 360–361): a judge declaring someone guilty just makes this person guilty, and yet also makes a factual statement to the effect that the accused actually did commit such-and-such criminal acts. Similarly, the WHO declares a “pandemic” because, to the best of WHO’s knowledge, there actually is a pandemic.Footnote 9 There are, then, belief-relevant sincerity conditions related to such acts that do not exist in pure declarations, e.g., in the act of declaring a war or opening an academic conference. As a result, one can be right or wrong in such declarations, and one can lie in them too.Footnote 10 Interestingly, in the case of correct declarations, the objective natural or social facts are coextensive with the declared institutional facts. However, in the case of incorrect or even manipulative declarations we have two parallel facts running their own course. For instance, for all the legal intents and purposes, we might act, and even be obliged to act, under the conditions of pandemic as an institutional fact, while the pandemic as an epidemiological fact is actually not happening (and vice versa, as witnessed by the situations where authorities declare an end to lockdown restrictions without obvious changes in epidemiological facts). There exist erroneous verdicts.

Note that this is precisely Horton’s argument: the WHO declared the wrong kind of health emergency. Instead of ‘pandemic’, we should officially talk about ‘syndemic’, a concept that not only better captures the evolving epidemiological facts, but also points to more adequate ways of addressing the short- and long-term effects of COVID-19. SYNDEMIC is thus epistemically more precise and prescriptively more fruitful, thus meeting two classic criteria for conceptual work (Brun, 2016; Carnap, 1950; Dutilh Novaes, 2020; Plunkett, 2015).

Further, and most importantly to our discussion: in the case of pandemic, the WHO used their recognized prerogative to apply the standing declaration to an individual case at hand. Standing declarations are constitutive rules determining what would be an acceptable applied declaration (Searle, 2010, p. 13). In our case, it is within WHO’s powers to declare a pandemic antecedently defined as “the worldwide spread of a new disease”—and they did just that on March 11, 2020.Footnote 11 However, one can also discuss and institute a standing declaration in the first place, thus fixing the general rule X counts as Y in C. This type of declaration takes the form of an institutional definition, or a part of it: e.g., Dying with recognizable COVID-19 symptoms (dry cough, fever) but without any further evidence counts as dying of COVID-19 in the context of Belgian elderly care homes residents. Institutional definitions, while linguistic, thus require an extra-linguistic institution, against Searle’s arguments to the contrary (1975, p. 360; 2010, Chaps. 4–5). Any such definition, when duly approved and recognized, becomes a standing declaration which, whenever implemented, creates an institutional fact, a recognized status that comes with certain rights and obligations, as described above.

While Searle’s original intention was to theorize how institutional reality is constructed and maintained, we re-use his distinctions in order to precisely delineate the domain where metalinguistic arguments over “pandemic” and “Covid-19 death”—and multiple other similar cases—can and do happen. Given the relevance of epidemiological and medical facts, arguers are bound to discuss and pronounce representative declarations. Further, since the focus of these arguments is not merely on how to apply a given concept under specific circumstances but rather how to “define” or “classify” that concept in the first place, the MIs concern standing declarations. Such standing representative declarations are the object of our study here.

In the literature, the complex interrelations between metalinguistic and substantive issues are well recognized (Chalmers, 2011; Plunkett, 2015). In principle, there might exist criteria for distinguishing between the two; in typical cases, we enter the realm of MIs when speakers continue to disagree while, factually speaking, all is said, done, and mutually agreed on (including facts over the other speakers’ meanings).Footnote 12 However, most curious in the philosophical debates are various hybrid and messy cases. Indeed, the standing representative declarations we analyze in this chapter are clear and interesting instantiations of such mixed phenomena. Here, the declarative, definitional element accounts for the metalinguistic or conceptual aspect, while the representative element accounts for the substantive aspect. The dispute is thus indexed and accountable to some external reality—just as much as a judge’s decision to declare someone guilty is—but, once a declaration is issued, it does become an institutional fact itself.

Beyond this fundamental aspect of conceptual work over institutional facts, we point to three key elements of the philosophical dispute over MIs, particularly relevant to an analysis such as ours.

First, as elaborated in their unique ways by Haslanger (2012), Plunkett & Sundell (2013, 2021; Plunkett, 2015), and others (e.g., Ludlow, 2014; Schiappa, 2003) MIs—or at least those most persistently argued about—are driven by normative, rather than descriptive, concerns. Plunkett & Sundell (2013, 2021) distinguish between descriptive metalinguistic disputes over how a term is used (e.g., “For us in Europe ‘football’ means a different game than for you guys in the USA”) and normative metalinguistic disputes over how a term should be used (e.g., “Waterboarding is torture”; “Horses are athletes”). In contrast to the descriptive cases, the issue cannot be conclusively settled by appeal to current usage or some linguistic authority (e.g., by restriction to the current regulations within the legal domain)—which makes them disputes of particular philosophical interest. Plunkett & Sundell call them, somewhat misleadingly, metalinguistic ‘negotiations’ (see below). Normativity itself is, however, another forbiddingly complex notion.Footnote 13 For the sake of illustrative simplicity, we can divide normative grounds of MIs into three large classes, recognized since antiquity: the true, the good, and the beautiful. The former two are especially relevant to our discussion here. In the first place, one engages in MIs for the sake of epistemic enhancement. In doing so, one can appeal to Carnapian values of specifically scientific exactness, simplicity,Footnote 14 and fruitfulness in the pursuit of methodic inquiry (Brun, 2016; Carnap, 1950; Dutilh Novaes, 2020) or to a broader metaphysical value of “carving nature at its joints” (Campbell et al., 2011; Scharp, 2020). Such appeals can support and explain prototypical examples of conceptual refinement such as those concerning FISH and ATOM (Carnap, 1950; Dutilh Novaes, 2020; Rast, 2020). In the second place, MIs work in the service of ethical concerns, that can be quite general and abstract or more applied, focused on concrete cases (Burgess et al., 2020). Ideals of fairness, equality, or dignity are thus often invoked in attempts to intervene metalinguistically on a concept such as FREEDOM or on a concept such as MARRIAGE or TORTURE. Importantly, in either case, a broadly pragmatic approach can be defended, tying the grounds and forms of MIs to the goals at hand, e.g., those of scientific inquiry or of public policy. Plunkett & Sundell (2021) stress the primacy of such overarching practical goals when they insist that “arguing about whether waterboarding is torture is a way of arguing about whether we should waterboard, or about how we should treat people that do it, or some other normative issue” (p. 162, emphasis in the original).Footnote 15

As we shall see, these concerns are indeed crucial in the public debate over COVID at large, and COVID deaths in particular.

Second, MIs can be performed via disputes over terms and concepts explicitly mentioned as arguable, or via disputes over terms and concepts implicitly used as arguable (Burgess & Plunkett, 2013; Plunkett & Sundell, 2013, 2021; Rast, 2020). While the latter seem finer and more elusive thanks to their intricate pragmatic mechanisms, the former are more directly amenable to the analysis of the arguments driving the dispute. In this case, however, we would rather not call them metalinguistic ‘negotiation’ but ‘argumentation’, given the centrality of “rational conflict” to the concept (Plunkett, 2015): rational conflict, or a disagreement instigated by rational concerns, when managed on rational grounds via linguistic exchange, just is argumentation on the most standard meaning of the term (see Dutilh Novaes, 2021; van Eemeren & Grootendorst, 2004). Accordingly, one of the tenets of argumentation theory is that it is public argumentation, and not private reason, that promotes rationality, precisely due its explicitness. For Johnson (2000), argumentation is not only rational, but manifestly rational, so that arguers can mutually see, test, and acknowledge the rationale behind inferential steps taken.Footnote 16 By contrast, ‘negotiation’ denotes a linguistic activity of arriving at a reciprocally agreeable private compromise, that can be rational, but can also be purely transactional (Godden & Casey, 2020).

In this way we second Ludlow’s idea that the driving force behind MIs is to come up “with progressively more serviceable modulations via a normatively constrained process of argumentation” (Ludlow, 2014, p. 111). These processes can be based on analogical arguments or arguments from authority (Ludlow, 2014) or on various other forms of definitional and semantic arguments (for a recent overview, see Pruś, 2021). However, as we show below, in line with point one discussed above, practical reasoning seems to be a central type of argumentation grounding MIs.

Finally, one of the key concerns in conceptual ethics or engineering is this: can we really control the change of our concepts? Ludlow (2014) and linguists working within lexical pragmatics (Allott & Textor, 2012; Hall, 2017; Wilson, 2003) argue that in communicative contexts, speakers can tweak meanings via pragmatic or semantic modulations. For Ludlow, this idea comes with radical contextualism whereby interlocutors, as it were, create their “micro-languages” from scratch in any given conversational context, and thus are free to adjust their meaning at will. Diametrically opposed to this position, and rooted in a particularly unrelenting understanding of semantic externalism, we find Cappelen’s lack of control argument: given that meanings (intensions and extensions) supervene on long-term patterns of usage within a broad linguistic community, local and individual attempts at meaning change can only have a minute and unpredictable impact, if any at all. Cappelen admits, however, that attempts at MIs continue, driven by normative concerns: even if a lasting, widespread semantic change is arguably beyond speakers’ control, we still engage in MIs if only because “in general, we don’t make normative judgments […] only when we have worked out a strategy for how to change the world” (Cappelen, 2018, p. 75). Our normative reasons, discussed above in points 1 and 2, thus prevail over practical limitations: we pursue, however unwittingly, the “right” meanings of our words even if we cannot fully understand, let alone control, processes of meaning change.

In between these two extremes, various options for effective intervention on our concepts and meanings are conceivable and have been explored in the literature, from forms of metalinguistic activism (Sterken, 2020), to engagement in “collective long-range” meaning change efforts (Koch, 2021), or even the engineering away, from our very concepts of CONCEPT and MEANING, of whatever features stand in the way of agents’ control over their representative devices (Riggs, 2019). While we are not in a position to further explore here, let alone resolve, this debate, we note a special context where control over meanings is well possible, and even expected. This is exactly the area of social ontology, discussed above. It is within the deontic powers of certain certified bodies—international organizations, constitutional assemblies, parliaments, municipal and faculty councils, but also courts of various instances, notably supreme courts—to declare on certain conceptual choices via their legal authority to do so, thus pronouncing binding semantic resolutions. Vivid examples of this—anything from what is a PERSON to SUSTAINABLE FASHION to SANDWICH—are discussed by Ludlow (2014) and within argumentation theory (Greco & De Cock, 2021; Schiappa, 1993, 2003). In such instances, Searle’s formula for constitutive rules operative in declarative speech acts—X counts as Y in context C—replicates itself thus creating social reality, with its network of intentional states and background capacities (Searle, 1995, 2010).

In this way, we thus carved out our approach to MIs: we specifically focus on MIs (1) grounded in various forms of normative argumentation, (2) explicitly debatable in the public sphere and (3) aimed at meaning change in the domain of institutional facts. These three characteristics jointly converge on an approach to MIs particularly fruitful in our inquiry over what counts as a COVID-19 death.

3 Arguing Over What a COVID-19 Death Is

In this section, we argue that public understanding of the COVID-19 pandemic, and a successful response to it, depend in part on an answer to a seemingly simple question: What do or should we mean by a “COVID-19 death”? This concern is reflected in the metalinguistic arguments of health authorities and public media that we analyze here.

3.1 The Early Confusion

Consider the discussion over case mortality rates of COVID-19 that, in the early stages of the pandemic in Europe (February–March 2020), varied from 1% (Germany) to 10% (Italy, Spain, Belgium). Explanations abounded on how to account for this difference.Footnote 17 Obviously, “facts on the ground” were brought up: demographics such as average population age, health, and density; overall quality of healthcare with a focus on available ICU beds and ventilators; government response, including the timing and severity of the lockdown measures; availability of the personal protective equipment (masks, gloves); even air quality. Further, testing methodology was recognized as playing a key role: tests could be limited to patients with severe symptoms and their direct contacts, resulting in higher mortality rates reported, or included a broader, asymptomatic population, producing lower rates. Quite recognizably, such background facts and methods are two standard grounds for substantive disputes over this and similar cases.

However, from the onset of the pandemic, a third line of explanation has been present, one that focuses on the “differences caused by clinical definitions of what counts as a Covid-19 death” (“BBC report”).Footnote 18 Such differences can be seen as particularly artificial when urgent and concerted action demand adequate worldwide comparison and coordination in the counting of cases. As we have already mentioned above, the question of “what counts as a Covid-19 death” does not admit of an obvious, single answer. Given the virus has been particularly lethal among older patients with other underlying illnesses (so called “comorbidities”), how were doctors advised to discern whether a patient died “as a result” of COVID-19, or rather a bacterial pneumonia, terminal cancer, or heart attack? While during the early stages of the pandemic most countries instituted a simple principle—any death of a patient tested positive “counts as” a death “caused by” COVID-19—actual clinical practice across and within different European countries varied, spurring a dispute among health professionals, policymakers, and the general population.Footnote 19

Here, we defend the position that this problem—as well as any of the attempted or possible solutions—is a metalinguistic one.Footnote 20 Some institutions we analyze below explicitly mention this as being a matter of definitions and classifications (WHO, ONS in the UK). However, even more importantly, a confirmation that the relevant lack of coordination in accounting for COVID-19 deaths is, at least in part, semantic in nature stems from the fact that it can straightforwardly give rise to verbal and metalinguistic disputes. It is quite natural, in this context, for someone to abstain from answering an object-level question like “Is this a COVID-19 death?”, or “Did x die of COVID-19?”, and to reply, instead, at the meta-level, with something like “It depends on what you mean by ‘COVID-19 death’.”Footnote 21

Indeed, in the spring of 2020, nascent metalinguistic arguments began to emerge. The predominant line defended the broad definition as an adequate indicator of the dangers of the pandemic. Others called for a more precise, narrower approach needed for better clinical practice and public response: COVID deaths need to be actual COVID deaths, not just deaths of people who happened to have a positive result, but in fact died from other illnesses, or simply old age.Footnote 22 In an apt rejoinder, the liberal side responded that, given the early scarcity of tests, counting only the positively tested cases amounted to a gross underestimation of the actual scope of the pandemic.Footnote 23 Compared to other pressing epidemiological concerns, these might sound as futile verbal disputes. Still, these semantic arguments illustrate the first efforts to understand and fix what ‘COVID-19 death’ means and to properly gauge the impact of the pandemic across the world’s population. An argument from analogy was also put forth (see “BBC Report”): In the case of the 2009 swine flu pandemic, depending on the way health professionals “assigned causation”, the death rate varied from dangerous 5.1% (early reports) to mere 0.02% (current corrected rate, based on a careful revision of medical data, including definitions and assignments registered in death certificates). Should the disputes over COVID-19 reveal a similar effect, then arguments over meaning would be very much worth having. Yet, as the pandemic raged in the spring of 2020, no consensual and conclusive reasons managed to decisively tilt these meaning disputes toward one solution or another. At this stage, international and national institutions stepped in.

3.2 Solution 1: WHO’s Broad Concept

In April 2020, the WHO intervened, producing “International guidelines for certification and classification of COVID-19 as cause of death based on ICD: International Statistical Classification of Diseases.”Footnote 24 Referring to “probable or confirmed COVID-19 cases” WHO’s “definition for deaths due to COVID-19” stipulated that:

  1. (2)

    A death due to COVID-19 is defined for surveillance purposes as a death resulting from a clinically compatible illness, in a probable or confirmed COVID-19 case, unless there is a clear alternative cause of death that cannot be related to COVID disease (e.g. trauma). There should be no period of complete recovery from COVID-19 between illness and death.

    A death due to COVID-19 may not be attributed to another disease (e.g. cancer) and should be counted independently of preexisting conditions that are suspected of triggering a severe course of COVID-19. (p. 3)

The definition is lax in its epistemic demands and broad in its reach. Quite surprisingly, it counts merely “probable” cases on a par with “confirmed” ones and determines that deaths due to COVID-19 “should be counted independently of preexisting conditions”, even those “preexisting conditions that are suspected of triggering a severe course of COVID-19” (for further discussion, see Amoretti & Lalumera, 2021 and Lindahl, 2021).

In view of our foregoing discussion, important questions arise: Can the WHO determine what a ‘COVID-19 death’ means? And, in this particular case: Did the WHO at least produce a sound argument for what a ‘COVID-19 death’ means or should mean?

Resorting to the distinctions introduced earlier, the WHO argues over an institutional concept of COVID-19 DEATH, precisely because of the obstacles to the deployment of a scientific concept, both principled (the nature of cause of death) and practical (insufficient capacity to test and perform autopsies). Not only is the very possibility of electing a single, natural, scientific concept of COVID-19 DEATH doubtful for the reasons discussed above, such a concept would not, in any case, be immediately adequate in the context of turmoil, fragmentary information, and pressure for quick measures and pronouncements. Under these circumstances, the WHO is able, indeed obligated, to intervene and fix the institutional (and operational) meaning for ‘COVID-19 death’. It undoubtedly has the effective power to implement worldwide changes in how the term is applied in official documents and statements. Accordingly, the document starts with the broad standing declaration of what should count as COVID-19 death (2), and then moves on to specific instructions on how to apply this declaration in concrete cases (2a). It thus first offers an argument to definition and, once this is settled, an argument from definition (see Pruś, 2021; Rigotti & Greco, 2019). Importantly, the conceptual argument to definition is grounded in normative concerns “of importance for public health” that are relevant “for surveillance purposes” and “the most useful cause of death statistics possible.” The concern for producing data “comparable to data from other countries” further reinforces this argument.

In this way, as also noted by Amoretti & Lalumera (2021) and Lindahl (2021), values other than medical or scientific accuracy govern this intervention. The WHO is explicit about the heterogeneity of considerations shaping their definitions and instructions:

  1. (2a)

    With reference to Sect. 4.2.3 of volume 2 of ICD-10, the purpose of mortality classification (coding) is to produce the most useful cause of death statistics possible. Thus, whether a sequence is listed as ‘rejected’ or ‘accepted’ may reflect interests of importance for public health rather than what is acceptable from a purely medical point of view. Therefore, always apply these instructions, whether they can be considered medically correct or not. Individual countries should not correct what is assumed to be an error, since changes at the national level will lead to data that are less comparable to data from other countries, and thus less useful for analysis. (pp. 8–9)

Key scientific values such as precision and self-correction are thus overridden by a straightforward practical argument: in the current situation marked by scientific uncertainty and lack of consistency, and given our institutional mandate of protecting public health in an internationally coordinated manner, the best definition-qua-rule we can institute is: any death resulting from a clinically compatible illness, in a probable or confirmed COVID-19 case, counts as COVID-19 death (unless there is a clear alternative cause of death that cannot be related to COVID disease) in the context of the current pandemic.

3.3 Solution 2: Belgium’s Broad Concept

A much-debated version of the WHO’s definition of what counts as a COVID-19 death was introduced in Belgium. The controversy revolved around how Belgium decided to fix the meaning of ‘COVID-19 deaths’ by including in it ‘probable deaths’ and counting such cases in the official statistics of COVID-19 deaths.Footnote 25

  1. (3)

    “As in other European countries, there wasn’t enough test capacity in the beginning to extensively test patients in nursing homes,” said Joris Moonen, a spokesperson for the agency that oversees nursing homes in the Dutch-speaking region Flanders. “We choose to register every death who had potentially died from COVID-19 to detect in which nursing homes the virus had hit. We were aware this would lead to an overestimation but found the signaling more important.”Footnote 26

Placing the value of public health “signaling” over the possible epistemic “overestimation” mirrors WHO’s arguments. Critics called it simply “stupid”, on both epistemic and practical grounds. First, as reported, “of Belgium’s registered deaths, 44% died in hospital (and were tested). The majority 54% died in a nursing home—and only in 7.8% of those cases was COVID-19 confirmed as the cause.”Footnote 27 That leaves almost 50% of official numbers in the medical dark. Second, such approach possibly had adverse practical consequences: “Neighboring countries may be less likely to reopen their borders for Belgian companies or tourists once European governments start to loosen their confinement measures.”Footnote 28 Indeed, in the early weeks of the pandemic (March–April 2020), Belgium had the highest per capita death rates in Europe and even in the world. This indicates another set of pressing practical arguments relevant in debating the institutional concept of COVID-19 DEATH, namely, public image factors (e.g., not appearing a failed state) and commercial interests (of national businesses or tourists).Footnote 29

However, defenders of the government’s policy produced counter-counter-argumentsFootnote 30:

  1. (3a)

    “It’s important that people are aware of the deceases outside the hospitals,” Van Gucht [who chairs the government’s scientific committee for coronavirus] said. “A broad way of counting enables us to monitor and quickly intervene where needed. Numbers are very important to create a sense of urgency—for example for the nursing homes. Belgium shouldn’t be ashamed about that.”Footnote 31

Again, the argument of efficient public health response takes precedence here over slow-paced medical accuracy. Importantly, while Belgian officials explicitly discuss “a way of counting”, it is worth noting that in this context the statistical sense of “counting” is derivative of the definitional sense of “counting” as in the formula X “counts as” Y in context C. This, as we have argued, accounts for the metalinguistic aspect of the dispute over institutional facts.

3.4 Solution 3: UK’s Narrow Concept: ONS Versus GOV.UK

Belgium’s chief scientist’s argument that “it’s important that people are aware of the deceases outside the hospitals” is not a standalone reason, but rather a direct objection to the decisions taken in other European countries, notably the United Kingdom, the recently estranged ex-member of the European Union. In the UK, the government instituted a principle that only deaths (1) with a confirmed positive COVID-19 test and (2) those occurring in hospitals count as COVID-19 deaths to be reported in official statistics. This practice directly contradicted WHO’s instructions and practices of countries such as Belgium. Unsurprisingly, this triggered a public debate, outside and inside of the UK. A useful summary of this early debate can be found in the official blog of ONS, the Office for National StatisticsFootnote 32:

  1. (4)

    ONS figures by actual date of death (death occurrence) tend to be higher than the GOV.UK figures for the same day. This is because:

    • We include all deaths where COVID-19 was mentioned on the death certificate, even if only suspected: the GOV.UK figures are only those deaths where the patient had a positive test result

    • We include deaths that happened anywhere in England and Wales, for example some might be in care homes: the GOV.UK figures are only those that happened in hospital.

  • So who is right about the number of deaths?

  • The issue is not really about right or wrong, but about each source of data having its own strengths and weaknesses.

  • The figures published on GOV.UK are valuable because they are available very quickly, and give an indication of what is happening day by day. Their definition is also clear, so the limitations of the data can be understood. But they won’t necessarily include all deaths involving COVID-19, such as those not in a hospital.

  • Numbers produced by ONS are much slower to prepare, because they have to be certified by a doctor, registered and processed. But once ready, they are the most accurate and complete information.

  • Using the complete death certificate allows us to analyse a lot of information, such as what other health conditions contributed to the death.Footnote 33

This post nicely captures the institutional dilemmas to be resolved. UK Government has a “definition” of COVID-19 death that is clear, fast, and frugal. But ONS deems it too far removed from “the most accurate and complete information”, something ONS is after in their approach. The government was responsive to such arguments, and in August 2020 changed its definition by removing the condition of hospital death, thus defining COVID-19 deaths as:

  1. (4a)

    deaths in people with COVID-19 that occur within 28 days of a first positive laboratory-confirmed test.Footnote 34

Hence the condition of a “positive laboratory-confirmed test” remains necessary. In their justification of this decision, the government argued the following:

  1. (4b)

    ONS reports deaths where a doctor suspects COVID-19 as a cause – these data include a clinical assessment as recommended by WHO but are subject to variation in clinical judgement as to the cause of death.

In other words, the institutional extension of ‘COVID-19 deaths’ should be a subset of the scientific extension: discretionary powers of individual doctors, which inevitably include subjective suspicion and varied judgment, should not yield to “laboratory-confirmed” truth of the matter.Footnote 35 Both the WHO and the ONS are thus mistaken in their approach—and so is the Belgian government. Whichever way the argument goes, however, the British case demonstrates the possible transience of conceptual interventions, an issue central to Ludlow’s (2014) framework. Certain conceptual solutions might be adequate in a certain context, while certain specific conditions hold, and inadequate when something changes. This consideration brings us to the last option for conceptualizing COVID-19 deaths in the context of, by mid-2021, a prolonged, unrelenting pandemic.

3.5 Solution 4: Excess Deaths

In February 2021, half a year on since its August 2020 update, GOV.UK, while maintaining its official reporting policies and distinguishing itself from the ONS, considered yet another approach:

  1. (5)

    But there is a third measure, which arguably provides the most comprehensive overview of the impact of the pandemic: excess deaths.

    These are the number of deaths over and above what would be expected, based on trends in previous years. Because they capture deaths from all causes – not just COVID-19 – they give us an idea of both the direct and indirect impact of the pandemic.Footnote 36

So defined, the concept of EXCESS DEATHS has been gaining prominence in the discussions as the pandemic progressed and its impacts have become ever more apparent. Apart from direct COVID-19 deaths (notwithstanding all the methodological challenges on how to account for them, especially in the case of comorbidities such as cancer, hypertension, or diabetes), there is a large category of indirect deaths that includes: (a) people who died of other conditions that appeared or aggravated during the pandemic but were not properly treated because of lack of access to healthcare, whether actual (discontinued treatments, cancelled operations, no hospital beds available) or perceived (fear of going to hospitals and contracting the virus) and (b) people who suffered depression and other mental health issues, possibly leading to suicides. (All the same, due to reduced mobility and limited transmission of other viruses, there was also a marked decrease in mortality due to, e.g., traffic accidents or seasonal influenza.)

Among the institutions that proposed to refocus attention on the concept of EXCESS DEATHS are the Institute for Health Metrics and Evaluation (IHME), an independent population health research center at the University of Washington and the Center for Global Development, a think tank in Washington, D.C., that prepared a report on excess deaths in India, one of the countries hardest hit by the pandemic and also widely suspected of inefficient reporting of COVID-related data.Footnote 37 These institutions brought up two key concerns: (1) comparison of excess deaths to the estimated total (direct) COVID deaths and, in turn, those to the data on COVID deaths as officially reported by various governments; (2) the import of the concept of excess deaths itself.

As for (1), the IHME reports the following:

  1. (6)

    Deaths that are directly due to COVID-19 are likely underreported in many locations, particularly in settings where COVID-19 testing is in short supply. Most excess mortality is likely misclassified COVID-19 deaths. An analysis by the Netherlands statistical agency suggested that all excess deaths in the Netherlands were directly due to COVID-19. In fact, their analysis actually suggested that direct COVID-19 deaths may be higher than estimated excess deaths because deaths due to some other causes have declined during the pandemic.Footnote 38

Moreover, drawing from different data sources, IHME evaluated the “ratios of total COVID-19 deaths to reported COVID-19 deaths”: in their global tally, Belgium is among the countries with the lowest distortion of officially “reported” deaths to actual “total” deaths.Footnote 39 This indicates the Belgian broad concept might, in the end, have the sought-after empirical adequacy.Footnote 40

Regarding (2): These complex comparisons and estimates make clear that EXCESS DEATHS are not being proposed as a more precise, simpler, or more fruitful version of the concept of COVID-19 DEATHS. Instead, they are being proposed as a concept that can, as it were, cut the knot and supersede the concept of COVID-19 DEATHS altogether. The argument for this conceptual replacement, rather than for continuous refinement of the notion of COVID-19 DEATHS, runs as follows: What counts in the bigger scheme of things—global public health, global economy, etc.—is the overall impact of the pandemic on the world’s population. And the concept of excess deaths allows to gauge this impact in a more robust, adequate, and methodologically neat way. Excess deaths thus mark a conceptual shift similar, indeed directly related to, the shift from PANDEMIC to SYNDEMIC, discussed above in Sect. 2.2 (examples 1 and 1a).

4 Discussion

Our analysis lets us develop two points. First, metalinguistic arguments over “COVID-19 deaths” are inextricably linked to substantive, scientific issues and, as we hypothesize, are partly determined by the imperfect character of our epistemic position on the subject. Second, they work in the service of broader practical arguments whereby scientific results are weighted against broader public policy values (e.g., a broader definition might justify more decisive containment measures).

4.1 Between Scientific and Institutional Concepts

What is at stake in the broader debate we analyzed is a natural expectation of a simple correct answer to the question “Which (and how many) deaths are due to COVID-19?” Nonetheless, as we have shown, there has been no single, privileged, natural concept of COVID-19 DEATH and no simple answer to this question. Principled concerns, most centrally related to the notion of “cause of death” in complex medical situations, make a clear classification of cases problematic, even assuming ideal access to the relevant information. Worse still, we are far from ideal access to the relevant information, as practical concerns of limited capacity for widespread testing and thorough autopsies have shaped the pandemic since its onset. Despite such concerns, in light of urgent need for public health intervention, some concrete response is needed. As we argued, uncertainties marring the scientific concept of COVID-19 DEATH recommend metalinguistic intervention on the institutional concept, designed to provide a fitting response to the circumstances. Such institutional interventions have two important features, already adumbrated by Searle.Footnote 41 First, reasoned control over meanings is well possible: being authoritative declarations, official interventions on the meaning of ‘COVID-19 death’ belong to the recognized deontic powers of the institutions mandated, among other things, to pronounce on the meaning of disputed terms. Second, such interventions are not entirely divorced from the attempts to get at the truth of the matter. They are, after all, representative declarations, expected to track, as much as possible, the features of the natural concept in question.

Taken together, these two features have some notable consequences. Metalinguistic interventions, as we understand them, are never closed or definitive. While they are meant to resolve some initial indeterminacy, they can typically not avoid all relevant sources of vagueness and indeterminacy. Indeed, if enough problematic cases accumulate after a first intervention, further action may be justified—as evidenced in the British case, where the definition of the COVID-19 deaths has been altered as new data became available. Such dynamicity of conceptual work allows to further understand the fertile tensions and interactions between “the facts on the ground” (revealed, e.g., via more accurate tests and autopsies) and the metalinguistic work performed by the institutions.

4.2 Metalinguistic Interventions as Practical Arguments

All the interventions we analyzed, starting from the WHO’s definition reflecting primarily “interests of importance for public health”, also reveal the heterogeneity of factors shaping MIs. As we have amply illustrated, the aim of following, however approximately, the truth of the matter is only one of the interesting themes and determinants of MIs. Other considerations alien to the question of descriptive accuracy clearly contribute to the forging of institutional concepts. These are primarily practical concerns of public health policies: in case of any epistemic doubt, apply classification most conducive to battling the disease from the public health perspective (e.g., precautionary principle, the lesser risk, etc.).

As such practical normative grounds take precedence, it is worth reconstructing many of the reasons behind the metalinguistic declarations as instances of practical argumentation (Lewiński, 2017, 2018, 2021). Practical argumentation starts from an action-question: What shall we do under current (unwelcome) circumstances to reach the desired goals? These goals embody our main values. In our case, these are chiefly related to international public health—i.e., the prevention of deaths and disease, and control of the pandemic—and explicitly formulated in terms of availability of fast, frugal, and easily comparable data instrumental in efficient coordination among countries. Yet, over and above such health concerns, confidence of citizens in the institutions of the state, preservation of a good international image of a country, outlooks of economic recovery, etc., are also carefully balanced in addressing the practical question of which measures should be taken to best attain these heterogeneous goals. In the case of metalinguistic arguments, the measures to be taken, that is, the conclusion of a practical argument, is precisely the definitional declaration issued in the form: (all things considered, given our goals and values, under current circumstances and best knowledge we have,) we should count X as Y (see esp. Sec. 2.2).

As we have discussed earlier, many forms of metalinguistic arguments have been identified in the literature: arguments from analogy and from authority (Ludlow, 2014); dissociative arguments which split the current concept into two new concepts via a subscript gambit, e.g., COVID-19 DEATHSCIENTIFIC and COVID-19 DEATHINSTITUTIONAL (Chalmers, 2011; Pruś, 2021; Schiappa, 2003); as well as the whole wealth of definitional and semantic arguments, such as arguments from verbal classification (Pruś, 2021). On our analysis it seems, however, that the class of metalinguistic arguments is just coextensive with the class of arguments at large, in the sense of recognized forms of informal arguments. Concepts can be carved out and defended by analogy, authority, dissociation, example, causal relations, etc. In the context of our analysis, practical arguments to a specific definition have been particularly prominent.

All these forms of argumentation are surely worth investigating in terms of the role they play in metalinguistic interventions. Indeed, attention to argumentation lets us better see such interventions, which might otherwise remain inconspicuous even as they shape our collective lives. It also lets us better evaluate them: public arguments in support of such metalinguistic interventions should be explicitly made and open to scrutiny as publicly accountable forms of normative argumentation. With our analysis, we hope to have contributed to such scrutiny, however modestly.