Note: This is a longer version of the review published here: https://www.cambridge.org/core/journals/philosophy-of-science/article/abs/review-of-paul-a-roths-the-philosophical-structure-of-historical-explanation/3F91E3462834E56669B4379A11A819D2

The philosophical structure of historical explanation.
Evanston: Northwestern University Press, 2020, 208 pp

Roth’s book is rich and intense. In a relatively short space, it seeks to locate historiographical explanations within our general metaphysical and epistemic predicaments. Understanding the argumentative nuances of the book requires knowledge of analytic philosophy of history, analytic philosophy in general, philosophy and history of science, and narrative theory. Following Roth’s argumentative moves from the beginning to the end, however, requires the acceptance of certain philosophical positions, arguments, and tenets that have been central in the analytic philosophy of history, as we will see. In this review, I go through these argumentative moves and discuss them critically. As Roth’s book provides an excellent overview of the potential richness and importance of philosophy of historiography, critically engaging with its contents and strategy enables us to see possible ways to cross-fertilize current philosophy of historiography and trends in other areas of philosophical research. Not only can philosophy of historiography, a self-consciously marginal field, benefit from insights elsewhere but also, as Roth’s book strongly establishes, other areas of philosophy must answer certain challenges that historiography as empirical research raises.

The book is structured as follows. Chapter 1 “Reviving Analytical Philosophy of History” identifies key issues of philosophy of historiography that merit and demand renewed philosophical considerations. Historiographical explanation in narrative form arises as the focal issue due to its intimate connections with epistemology and its peculiar features: nonaggregativity, nonstandardization, nondetachability (ASD) which imply, among other things, that historical knowledge cannot cumulate. Chapter 2 “Problems for Narrative Explanations: The Case of History” introduces three problems that ASD creates: Logical formlessness, evaluative intractability, and a metaphysical one, i.e. that narratives construct the past rather than tell how things were “in themselves”. Roth argues that once we abandon the metaphysical realism behind the metaphysical objection, it becomes possible to answer the other two objections. In the rest of the chapter and in Chapter 3 “The Past”, Roth argues against metaphysical realism and establishes an alternative position, irrealism, according to which there must be multiple pasts that we make up by imposing an order on what we take to have happened. In Chapter 4, Roth discusses essentially narrative explanations. “A narrative involves an “unfolding,” sequencing of a series of events that accounts for a development” (66) and a narrative provides an essentially narrative explanation if it cannot be paraphrased into some into a nonnarrative explanatory form. Essentially narrative explanations constitute the events they explain and therefore irrealism is their natural metaphysical partner.

In Chapter 5, “The Silence of the Norm” the book turns to the question of how historiographical explanations are related to other forms of scientific explanations. Roth argues that we have to understand the nature of historiographical explanation because historiographical insights inform our conceptions of science. Roth focuses on Kuhn and points out that, while Kuhn changed how we view science, we nevertheless do not understand how Kuhn’s historiographical approach made the change possible. In Chapter 6 “Kuhn’s Narrative Construction of Normal Science”, Roth argues that Kuhn’s explanation of the history of science is an essentially narrative one. Because only such historical explanation can tell, according to Roth, what counts as science, essentially narrative explanations are unavoidable: “Since narrative helps
constitute any explanation of what counts as normal science, narrative explanation cannot be divorced from what now counts as explanation in science.” (98). Chapter 7 “Methodological Naturalism
and Its Consequences” discusses how narrative explanations fit in a fully naturalized account of empirical knowledge. Roth argues that “assessing the goodness of such accounts will typically not be settled by appeal to agreed upon facts” (127) and therefore the multiplicity of accounts in historiography comes as no surprise. The chapter “Conclusion” argues that placing historiographical explanation in a naturalistic framework solves the philosophical problem of evaluative standards of historiography and philosophers are not in the position of providing criteria for rational evaluation. Historians construct historical events that serve as cognitive instruments and the “epistemic justification [for historical events] derives from their utility for purposes of explanation [–]” (145).

This step-by-step synopsis is useful because it enables us to observe two important issues. First, Roth’s book is not an attempt to describe in detail the conceptual structure of historical explanation in the same sense as theories of explanation in philosophy of science usually are (think of the interventionist theory, for example). Rather, it is an attempt to identify certain important features of historical explanation and to track down the epistemic and metaphysical consequences of those features. It maps historiography among sciences at a very abstract level without comparing in detail essentially narrative explanations with other theories of explanation. In fact, the book hardly mentions any other theories of explanation. Secondly, some of its conclusions – irrealism, DSA, the necessity of Kuhnian explanation in order to understand science – are rather strong in the sense that they do not belong to the mainstream of philosophical thought. This raises two questions. First, how strong are the arguments that lead to such strong conclusions? Is the prize of reviving philosophy of historiography too high or should we rather rethink the foundations? That Roth manages to bring a tradition of philosophy of historiography to its logical conclusions enables us to ask this question. As is well known, one person’s modus ponens is another’s modus tollens. Secondly, how can the strong conclusions be sustained in the face of the fact that other forms of scientific explanations do not appear to share the strongest consequences, given that historiography belongs, according to Roth, among empirical sciences?

Roth begins his discussion by taking a brief survey of philosophy of historiographical explanation from the late 1900th century to the present. He complains that the problem of historical explanation is one of philosophers own making because philosophers have been “implicitly or explicitly placing a demand on historical practice that emanates from philosophical preconceptions regarding the logical form of scientific explanation” (7). Equally misleading has been literary theorists’ emphasis on narratives as stylistic devices. Roth emphasizes that we have to study “how narratives address and impact epistemic concerns” (7). This brings us to Danto whose notion of narrative sentence plays a crucial role in Roth’s philosophical framework.

Danto (1962, 152) asks us to imagine an Ideal Chronicler. “He knows whatever happens the moment it happens, even in other minds. And he is to have the gift of instantaneous transcription: everything that happens across the whole forward rim of the Past is set down by him, as it happens, the way it happens. The resultant running account I shall term the Ideal Chronicle”. Roth argues that

“This thought experiment establishes that statements true of a particular time t cannot be comprehensively known at t, not even by someone capable of recording all that happens when it happens (the Ideal Chronicler). Danto’s now canonical example is this: “The Thirty Years War began in 1618.” This statement is true of what happens in 1618 but is not knowable in 1618, not even by an Ideal Chronicler. Danto calls these “narrative sentences,” and they demonstrate that there will be truths about any time t not knowable at t; truths about time t continue to accumulate after t.” (8).

Before going into the details of how Roth uses this insight to argue for irrealism and essentially narrative explanations, we should ask what the thought experiment really indicates. First, it is important to point out that the issue cannot be how we refer to some historical event or object. “The man who built the atomic bomb worked at a patent office” refers to Einstein but the description is not true of him. The way we refer to some event or object is independent of what is true of that event or object. Secondly, the Ideal Chronicler cannot write down anything that requires a crosstemporal relation (i.e. a relation that holds between events located at different times). Most notably, as Danto (1962, 159) points out, “”is a cause of” would not be a predicate accessible to the Ideal Chronicler”. This is important because there exists a natural distinction between properties and relations. Properties hold of the things while relations hold between things; they are not relations of anything. Given that a crosstemporal relation holds between two events, it does not follow that it is a relation of the earlier event. One could argue that the difference between us and the Ideal Chronicler is not that we know statements that are true of an event that the Ideal Chronicler does not know; the difference is that we know truths about crosstemporal relations that the Ideal Chronicler, by definition, is not allowed to capture. Truths about time t are not accumulating after t but truths about the past (part of which t is).

While the considerations above do not, of course, show that truths about the past are not accumulating as we move towards in time, it shows why there might be no mystery in the accumulation. If it is true about the past that a crosstemporal relation holds between C and E, this is true only if C and E both exist. That “the description of the past does not come closer and closer to an ideal chronicle but departs further and further from it as more descriptions become available which were not earlier available even in principle” (Mink 1987, 139; cited from Roth p. 9) is a consequence from the natural assumption that historiography captures crosstemporal relations (unlike the Ideal Chronicler).

Notice also that the Ideal Chronicler is not compatible with any theory of historical explanation. Given that explanation requires some sort of relation holds between the explanans and the explanandum, all possible theories of historical explanation require that we posit something that the Ideal Chronicle does not contain. In a sense, all explanations are retrospective: that C explains E requires that E is the case and an explanatory relation between C and E holds. These cannot be known at the time when C is the case.

The considerations above become important when we come to the crux of Roth’s argument: “This [that complete description of an event presupposes narrative organization] hints of a philosophical rationale for the metaphysical plurality and epistemic legitimacy of competing narratives [–]. In short, historians look to justify an explanation of an event under a particular description, a description tied to a retrospective and so narrative perspective”. (9). However, we have seen that retrospective perspective does not force a narrative account of explanation, as all explanations (of singular events) are retrospective in the sense implied by the Ideal Chronicle. Neither is it obvious that the accumulation of truths about the past forces us to accept the metaphysical plurality.

What is more, Roth argues that Danto’s insight reveals three key characteristics that distinguish between historical events and those that scientific theories target for the purposes of explanation: nonstandardization, nondetachability, and nonaggregativity. Nonstandardization concerns the fact that “historical events do not begin as constructs of some articulated theory of which they are a part” (9). According to Roth, this follows from the fact that narrative sentences reveal that “human relationship to history will always be dynamic and not static” and “histories, and so those who author them, constitute the events they explain under nonstandardized descriptions” (11). Nondetachability means that “events explained by histories exist qua events only as constructions of those histories” (14). The explanandum cannot be detached from the supportive narrative. Nonaggregativity means that there cannot be a “single account that links all possible events under one explanatory rubric” (67). Nonstandardization leads to nondetachablity because without standard descriptions, events exist only as a part of some narrative. Nonaggregativity follows because there are no narrative-independent events that the final explanatory account could tie together.

We can notice how strong conclusions begin to arise from Danto’s argument. At this point, the idea that historians construct different pasts by providing narrative explanations is more assumed than proved, and we will see soon what the further arguments for the claim are. However, already we have a case where cross-fertilization of philosophy of history and other areas of philosophy could be useful. The Ideal Chronicler thought experiment appears to be loaded with assumptions about truth, propositions, properties, relations, and reference which deserves clarification. It appears like a good strategy to spend energy on such clarification before drawing strong conclusions about the thought experiment – conclusions that might damage the overall defense of historiographical explanation, a topic discussed below. There is a decision to be made whether philosophy of history should follow its canonical arguments to their strongest conclusion or rethink the arguments by using the most powerful conceptual tools developed elsewhere. At the same time, the thought experiment also indicates that there are, at the core of historiography, issues that those conceptual tools need to resolve. The foundations of historiography should serve as a critical case for the presumed power of those tools and the philosophy of historiography does not deserve its marginalized position in this respect.

Roth continues the argument for the metaphysical plurality of the pasts by arguing that it is a mistake to think that “that there exists only a single past calling for explanation” (24). Roth states: “the sort of metaphysical picture I ultimately want to reject has it that events enter into processes by some natural historical dynamic inherent in the events and processes of which they are parts [–]. What needs to be rejected is the picture of a past that is simply there waiting for a historian to come along.” (25-26.) Roth’s argument is two-fold.

First, it is argued that “if the fixity of the past is a coherent notion, as it seems to be, then this implies that there could be an Ideal Chronicle” (28). However, there could not be an Ideal Chronicle because there are no ideal events to be recorded: “Without some description or other, there are no specific events; with an identifying description, we still do not know if the event is of the requisite ideal sort— that is, not primarily of our making” (29). Secondly, Roth relies on Goodmanian consideration of Ian Hacking who argues that nature does not dictate one organizing scheme and strengthens these considerations with Danto’s argument that truths about some time t are not fixed at any point of time (see above). These considerations lead to the rejection of the metaphysical reality of the past and also to the rejection of antirealism. “Since there will be narrative sentences true of any particular time t but not knowable at t, these could not be verified at just that moment. Thus, contra Dummett’s version of antirealism, there can be true but in principle unverifiable sentences at a given time even absent the assumption of realism” (48). Irrealism follows. There are multiple pasts that we make up by imposing order on what we take to have happened.

There are two interrelated problems in this argumentative strategy. First, it is not clear why metaphysical realism implies that an Ideal Chronicle is possible. As we have seen, the Ideal Chronicler is, despite the first impression, a rather limited cognitive agent. Given that metaphysical realism is committed to the idea that what is true goes beyond what can be known, why would metaphysical realism commit to the idea that a seriously limited cognitive agent could know everything? On the other hand, if the argument is that even a “super-chronicler” who can view all points of time at once is not possible because it should use some conceptual scheme to identify different events to write an Ideal Chronicle, then the impossibility-of-an-Ideal-Chronicle argument is redundant and all the argumentative force is on the consideration about how conceptual schemes made up the world.

Secondly, even though Roth claims “to give this metaphysical assumption of the objective past the most plausible form that I can, and then show that the assumption is untenable” (27), the most plausible form is one of Roth’s own making and, in the book, there is very little discussion about characterizations and arguments for metaphysical realism that the defenders of the position have given. Neither is there any discussion about the possible responses to the arguments in Chapter 3 concerning the metaphysical implications of conceptual schemes and theory-ladenness of empirical knowledge. This does not only harm the credibility of the argumentative steps in the book but also the possible cross-fertilization of philosophy of historiography and philosophy of science. For example, it would have been interesting to know how recent developments in perspectival realism, which is an explicit attempt to find a middle ground in between scientific realism and antirealism, relate to the conceptual space of irrealism. Even if Roth is correct and irrealism stands its ground, an opportunity to cash out the philosophical implications of the insights is lost. As we will see, this is a recurring theme in the book. Roth makes claims like “What is the case for historical knowing as a type of constituting extends to all forms of knowledge. What counts as evidence, and for what it counts, turns out to be a product of practices of inquiry as informed by the use of predicates (past or present)” (57) without filling in the details and tracking down their relation to other philosophical positions. Roth’s intriguing arguments would have deserved more detailed development.

The rejection of historical (metaphysical) realism is a crucial argumentative step as it provides “rationales for the defining features of historical narratives— nonstandardization, nondetachability, and nonaggregativity” (64). But what, exactly, are narrative explanations and how, exactly, do they carry ASD? “[A] narrative involves an “unfolding,” sequencing of a series of events that accounts for a development” (66). Narrative explanations explain narrative sentences. “A narrative explanation will be a presentation of a temporal series that answers why the explanandum turns out to be as it is. [–] [It provides] a sequencing of events that has the later event emerge as a consequence of the earlier” (68). To avoid confusion, it must be noticed that, for example, providing a narrative explanation for “The Thirty Years War began in 1618” does not explain, literally, “Why did the war began in 1618” but rather how the beginning is connected to later events so that those events together constitute the Thirty Years War, i.e. why some event turned out to be the beginning of the world. Narrative explanations necessarily require a retrospective stance: We can identify the event the Thirty Years War and thus its beginning only retrospectively. A narrative also “simultaneously constitutes and constructs both explanans and explanandum” (71). The sequencing of events that a narrative explanation provides also constitutes the event in question. Nondetachability follows because the explanandum is constituted by the explanans. Nonstandaridzation follows because there is “no settled theoretical “recipe” in historiography regarding how facts should or could be put together to make an event and which events they make” (67). Nonaggregativity follows because “different narratives identify different events and so different causal sequences. These cannot be expected to aggregate, to yield some integrated account about what happens and why” (67).

There are three main problems in this account. First, what makes a narrative explanation an explanation rather than a description of a set of events? Secondly, what does it mean, exactly, that narrative explanation involves a retrospective perspective? Thirdly, the arguments for and the characterization of ASD could be more precise.

The first problem concerns the fact that, usually, a mere sequencing of events does not explain anything. Roth is aware of this “what makes narratives explanatory[?] [–] narratives typically seem to be descriptive” but goes on to argue that the distinction between description and justification does not arise in the case of narratives. This seems to miss the point. Usually, a sequence of events is considered explanatory if it is causal. Roth seems to be aware of this but argues that “The causal sequence, in turn, can consist
only in this case of seeing facts as ordered and so related in a particular way” (73, emphasis original). Later it is added that “no functional distinction exists between describing that sequence and justifying causal links” (75). The problem is that if there is no stronger notion of causality than one that makes causality follow automatically from a description of a sequence of events, then causality does not add any explanatory import to the sequence. One cannot make causality carry the explanatory load without a notion of causality that restricts what sequences can be viewed as explanatory. The problem of causal explanation in historiography is deep and complex and we are left in darkness with respect to the relationship between narratives and causality – or any related explanatory relation, for that matter.

Secondly, we already saw above that all explanations are retrospective in the sense that we can only explain events that we know that has happened and we usually cite earlier events and conditions in the explanans. Normal causal explanations are an obvious example. The notion of retrospective stance required by narrative explanation must be a stronger notion in order to demarcate narrative explanation from other types of explanation. This is especially important because, in many arguments of the book, what can be known only retrospectively is used as a criterion to show that a particular explanation is a narrative one. For example, Roth argues (at the end of Chapter 4) that an account of explanation in evolutionary biology, developed by Beatty and Carrera, fits together with essentially narrative explanations because certain contingent events – events that could have been otherwise – that shaped the course of development can be identified only retrospectively. “[T]heir actual emphasis falls on the importance of retrospective knowledge, on knowing where the noteworthy forks exist and what did happen at those points”. However, the notion of retrospective implied by this case is not strong enough to prove that the explanation at hand is an essentially narrative one. Surely, if we know that C (rather than C*) had to be the case if E (rather than E*) is the case now, we can infer C from E. Moreover, if E was not bound to happen, C reveals it happened anyway. However, this “retrospective” structure is directly compatible with e.g. counterfactual theory of explanation (which says that C explains E if it is true that, had C* rather than C been the case, E* rather than E would have been the case). Essentially narrative explanations, by definition, would not allow such translation to another theory of explanation. In order to follow through the consequences of the notion of retrospective stance involved in narrative explanation, it is necessary (i) to precisely define the notion and demarcate it from the more benign notion, and (ii) to show how this precise notion identifies important explanations as essentially narrative ones that carry the epistemic and metaphysic consequences of such explanations.

Thirdly, there is a similar problem of ambiguity in ADS. For example, when Roth argues, in Chapter 7, that Kuhn provides an essentially narrative explanation, he points out that

“(T)he history that [Kuhn 1962] presents aggregates neither internally— thus incommensurability— nor with standard “textbook” histories of science. Indeed, Kuhn emphatically sets his account in opposition to those histories that maintain that accounts of scientific change do aggregate. That the account of what comes to be called at any given time a “science” fails to aggregate is at one with the claim that the various revolutions show successor theories to be incommensurable with those they displace/replace. The infamous consequence that results concerns how this makes it notoriously difficult to cash out any claim regarding scientific progress.” (101).

It is quite difficult to see how these types of nonaggregativity are related to the one associated with irrealism. First, the existence of historiographical revolutions hardly establishes that it is in principle impossible to knit together several narratives of the past. Secondly, the existence of historical ruptures and the lack of a notion of progress in science (or anywhere, for that matter) hardly shows that there cannot be a narrative that sequences events through such ruptures. Nonaggregativity was supposed to be a consequence of underlying metaphysics of historical explanation, not an empirical fact about historiography or historical continuity. Again, it is necessary (i) to precisely define the notions in ADS and demarcate them from the more benign notions, and (ii) to show how these precise notions identify important explanations as essentially narrative ones that carry the epistemic and metaphysic consequences of such explanations.

Still, the most interesting part of Roth’s book (Chapter 5) makes the important observation that, while historiography informs our understanding of science, we do not understand historiographical explanation well enough to tell how historiography does this. Roth goes to argue (in Chapter 6) that Kuhn provided an essentially narrative explanation of what counts as normal science. Roth adds that such explanation is necessary since if we wish to understand science naturalistically, we cannot rely on some a priori notion of science when defining what counts as science.

However, there are three interrelated worries that this argumentation raises. First, while it is true that historiography of science can be highly useful in order to understand science, it does not follow that we do not know what counts as science independently of historiographical work. Most people who do science or study it do not have a historiographical story to back their conception of science. Moreover, Roth himself argues that historiographical explanations are given retrospectively: “what proves adaptive as a “scientific” strategy comes to be known only retrospectively” (104). It seems that this implies that, in order to get historiography of science of the ground, we need to be able to identify current science independently of its history.

Secondly, even if historiographical explanation is required to identify science, it does not follow that essentially narrative explanations are the only – or the best – candidates to perform this function. I have personally defended a counterfactual account of explanation of the development of science. No matter how flawed that account is, it seems implausible that counterfactual explanations cannot, in principle, account for scientific development. Equally unsure is the necessity of the Kuhnian account of science. It might be that the Kuhnian account is either false, unclear, or unable to capture all the interesting aspects of scientific development. Moreover, we already saw that it is unclear whether ADS and retrospective stance unambiguously identify essentially narrative explanations and whether they unambiguously identify the Kuhnian account as essentially narrative. If narrative explanations are not necessary to identify science, one major reason to place them along the science falls, namely that “since narrative helps constitute any explanation of what counts as normal science, narrative explanation cannot be divorced from what now counts as explanation in science” (98).

Thirdly, there is a practical problem. Roth argues that historiography is a science if we understand science naturalistically. However, there is no Kuhnian narrative for how historiography has developed as a science. Given that a Kuhnian narrative is necessary to count something as science, then historiography does not count as a science.

In Chapter 7, Roth provides a defense of methodological naturalism and historiography’s place among empirical sciences. This chapter raises two general worries in the light of Roth’s argumentative strategy. First, given that essentially narrative historiographical explanations are required, according to Roth, to tell what counts as science and given that each explanation constructs its own past (irrealism), different historiographies of science construct different conceptions of science. When we attempt to understand science naturalistically, whose conception shall we rely on? Moreover, given that historiographical narratives do not aggregate, it seems that we cannot improve our understanding of science in a piecemeal manner by collecting historical insights concerning science. It seems that if we accept Roth’s picture about historiography and historiography of science, there is not much that philosophy of science can achieve in its attempts to understand science. Quite a heavy conclusion.

Secondly, the need for detailed studies of the epistemic limitations and workings of historiography does not stop even if we grant historiography a place among sciences. Even if naturalism “situates the study of humans, in all their aspects, as of a piece with those methods and theories used to investigate other objects in nature so conceived” (116) there might be important differences between science below this general level. This becomes obvious when we notice that not all research within a single field of science is equally good. However, Roth would probably argue that there is nothing philosophically interesting about these differences. For example, Roth argues that “it is hard to comprehend [–] which evaluative standards of historians supposedly cry out for a philosophical solution” (141) and the same probably goes for other questions concerning details of how science works.

However, this outright hostility towards philosophy does not seem quite justified and is in fact connected to the fact that Roth’s book does not discuss theories of explanation in detail. Roth writes:

“[E]ven with the waning of overt philosophical enthusiasm for some unitary model of scientific explanation, the problems attending historical explanation remain unchanged from their origins well over a century ago. In this respect, i.e., by virtue of implicitly or explicitly placing a demand on historical practice that emanates from philosophical preconceptions regarding the logical form of scientific explanation, I have termed the putative problem of historical explanation as one of our (i.e., philosophers’) own making” (7).
It seems quite unfair to dismiss the developments in the philosophy of scientific explanation with such sweeping claims. Currently, most philosophers working with the questions of scientific explanation work in as close connection with scientific practice as possible. This philosophy, as James Woodward put it, “recognizes that causal and explanatory claims sometimes are confused, unclear, and ambiguous and suggests how these limitations might be addressed”. Those, me included, who apply the conceptual tools developed by philosophers of science to history attempt to analyze the prospects and limits of the tools with respect to understanding history. This is far from imposing philosophical preconceptions – whatever that even means – on historical practice. It is hard to understand how one could deny, a priori, that such philosophical analysis might be important part of science, naturalistically understood.
.