This book "supplies a vocabulary of English words and idiomatic phrases 'arranged … according to the ideas which they express'. The thesaurus, continually expanded and updated, has always remained in print, but this reissued first edition shows the impressive breadth of Roget's own knowledge and interests."
Lycan, William G. (2001) Real Conditionals. Oxford University Press, 223pp, hardback ISBN 0-19-924207-0
Anne Reboul, Institute for Cognitive Sciences, CNRS, France
GENERAL PRESENTATION This very interesting book is a must for anyone, whether a linguist or a philosopher, interested in conditionals and conditional reasoning. However, it is a very compact book, suitable only for well-informed readers. It should on no account be taken for a text book or introductory work.
SYNOPSIS The book is made of eight fairly homogeneous chapters with an appendix on ''non-conditional conditionals''. Its aim is to give a truth-conditional account of conditionals avoiding the (mainly linguistic and syntactic) pitfalls in which most ''logical'' accounts of conditionals fall and keeping some of the virtues of the non-truth-conditional account known as the Ramsay Test.
It thus opens, unsurprisingly, with a chapter devoted to ''The Syntax of Conditional Sentences''. This begins with an attack against the general logical view of conditionals, according to which a conditional involves a ''syntactically unstructured binary sentence operator'' (1), i.e. ''if... then''. This is true whether ''if...then'' is taken as equivalent to the material implication or to the intensional or modal operator. This lead logicians to treat ''If A, then B'', ''B if A'' and ''A only if B'' as equivalent. This, Lycan claims, is wrong, for syntactic considerations. Though conditionals are generally defined as sentences with the connective ''if'', other adverbials such as ''in case'', ''in the event that'', etc. are semantically very similar and any theory of conditionals should take them into account. Other sentences, generally considered as logically equivalent to conditionals (e.g. some disjunctions) are syntactically quite different from them. Lycan proposes to consider as conditional any sentence in which ''if'' occurs and any sentence which is synonymous with, not merely logically equivalent with, such sentences. Finally, Lycan rejects the idea that conditionals are unstructured conjunctions, showing that conditional sentences exhibit syntactic properties radically different from conjunctions (e.g. neither conjunction reduction, gapping, across-the-board principle apply to conditionals). Neither are conditionals unstructured subordinating conjunction: they can be modified by ''even'' and ''only''. Indeed, the proximity of ''if'' with ''when'' and ''where'' seems to indicate that conditional clauses are adverbial, making ''tacit reference to events and circumstances'' (11). Finally, Lycan rejects, for syntactic reasons, the idea that ''unless'' is equivalent to ''if not''. Again, a Relative-Clause account should apply to ''unless'', as it applies to ''if'', though in the case of ''unless'', it undergoes a twist: the negation applies indeed, but not in the scope of ''unless''; rather it applies on the quantification over the events described in the clause.
The second chapter, ''Truth Conditions: The Event Theory'', is devoted to a description of Lycan's theory of conditionals. It is the longest chapter in the book and is indeed pivotal. The theory is a semantic theory in the sense that it intends to propose a systematic assignment of truth-conditions to sentences with ''if'', ''unless'', ''only if'' and ''even if''. This is supposed to account for the implications of such sentences, to explain the dependence of their truth-values on context and to agree with their syntactic properties. Lycan proposes the following paraphrases of such sentences: ''P if Q -- P in any event in which Q''; ''P only if Q -- P in no event other than one in which Q''; ''P even if Q -- P in any event including any in which Q''; ''P unless Q -- P in any event other than one in which Q''. All of these formulas involve universal quantification over a domain of events in which Q. Event is here to be taken as situation, in a sense similar to that of situation theory.
Formalizations of the paraphrases express the truth-conditions of the corresponding sentences. The universal quantifiers should be restricted to a reference class of ''real possibilities'', i.e. possibilities which ''the utterer must have (...) at least tacitly in mind as a live prospect'' (19). The reference class, however, should contain only ''relevant'' properties, which leads Lycan to a discussion of semifactuals and ''weak'' conditionals. A second problem is that the utterer might be wrong about possibilities. A solution to the first problem (the relevance of events) is to restrict the reference class to events where either P, non-P, Q or non-Q is true (Moderate Relevance Restriction) or to restrict it to events where either P or non-P is true (Strict Relevance Restriction). Both of these conditions probably apply to different classes of conditionals. The reference class is thus ''a hidden parameter that will vary with context'' (23). The second difficulty is more serious. A way out is to see that ''with the inclusion or non-inclusion of all actual relevant event in R [the reference class], stands or falls the validity of Modus Ponens'' (24). Lycan postpones the issue till chapter 3. Lycan's theory (henceafter the ''Event theory'') does not meet with the syntactic objections outlined in the first chapter against the unstructured-sentential-operator theory and has quite a few semantic benefits, among which avoidance of the paradoxes of material implication, accomodation (with some suplementation) of Stalnaker invalidities (antecedent- strengthening and transitivity), contraposition, semi-factuals and weak conditionals, explanation of the direction of conditionship, parametric differences between apparent equivalents, simplification of disjunctive antecedents and impossible antecedents.
The third chapter, ''Truth Conditions: Reality and Modus Ponens'', returns to the problem left aside in chapter 2. It critically examines the three major essays in the refoundation of conditional investigation, i.e. Adams (1965), Stalnaker (1968) and Lewis (1973). As Lycan points out, two different major paradigms emerged from these essays: the first one (from Adams) in terms of an epistemic assertibility semantics (non- truth-conditional); the second one (Stalnaker and Lewis) in terms of a possible worlds truth-conditional semantics. Both Adams' and Stalnaker's accounts stem from Ramsey's test. Ramsey's test consists in adding the antecedent hypothetically to one's present set of beliefs, revising this set where necessary, and checking whether the consequent is part of the revised set of belief. If it is, the conditional is assertible, if not, it is not. Adams translated it in terms of conditional probability relative to one's belief set, considering that indicative conditionals are restricted to epistemic assertibility and do not have truth-values. He said nothing of subjunctive conditionals. Stalnaker recasted Ramsey Test in terms of alternative possible worls with a selection function based on similarity. Conditionals would be evaluated by checking whether C holds in the world most similar to ours where A holds. Notably, Stalnaker defends this as a truth- conditional account of conditionals which he afterward adapted for subjunctive conditionals. Lewis rejected the notion of a uniquely nearest world with the accompanying acceptance of Conditional Excluded Middle. He replaced it with the notion of comparative similarity. Lewis' account was criticized for its reliance on an intuitive notion of overall similarity, through a number of counterexamples. He responded by discarding the everyday notion of similarity and advocating a brand of similarity specific to counterfactuals, though he was not specific about it. This is deeply unsatisfactory as noted by Lycan. The Ramsey Test is not immune from counterexamples either, though they mostly center on the relativity to epistemic situations it introduces. There are also counterexamples to the proximity between the Ramsey Test (taken to establish truth-value and not mere assertibility) and Similarity Theory, based on the fact that the two accounts do not always yield the same results. Notably, some exemples examined through the Ramsey Test will contradict Modus Ponens. This might be taken as an indication of the worthlessness of the Ramsey Test in establishing truth-value. This would lead to the choice of restricting the Reference class to ALL actual relevant events, whether or not they are envisaged or not (the Reality Requirement). However, Lycan points out that Modus Ponens does raise more problems than it solves, notably relative to Sobel sequences (i.e. ''if A then C'' may be false eventhough ''If A and B then C'' and ''A and B'' are true). This problem does not arise for Similarity accounts. However, it does for Ramsey Test accounts. In the Event Theory, which can be seen as a ''mixed view'', ''If A then C'' and ''If A and B then C'' can be true. Lycan, after discussing a few objections, turns to further counterexemples against Modus Ponens. One question which the Event Theory has to face is whether the reference class shifts when iteration or Sobel sequences are involved in the antecedent. Lycan's answer is clearly positive: this entails given up Modus Ponens as valid because of its FORM.
In the previous two chapters, Lycan had described the two great trends in approaching conditionals: the non-truth-valued (NTV) view and the truth-valued (TV) view. Though his approach tries to keep some of the advantages of the Ramsey Test, it is clearly TV. The fourth chapter, ''In defense of Truth Value'', offers some arguments for TV and counter-arguments for NTV. Lycan begins by listing arguments against NTV: its philosophical peculiarity; its linguistic bizareness; the fact that conditional speech acts suppose sincerity and truth; the many parallels between indicatives and subjunctives which make it hard to claim that, for instance, only indicatives would have truth-values; the possibility of embedding conditionals in longer sentences, where assertibility can play no role; their dependence on nomologicals, which goes against NTV; the problem of deductive validity for deduction involving conditionals; the problem of modals; and, finally, the problem with deflationism (which implies that conditionals should have truth-values). Lycan then shows that the Event Theory can answer all the problems which NTV answers without giving up truth-values. He now turns to what he calls the new horseshoe theory (NHT), which tries to make conditionals equivalent to material implication and to answers the numerous problems this raises. His criticisms of NHT are based on the claims that indicative conditionals are not restricted to the actual, that the material implication leads to a clearly invalid inference pattern (from ''not(if A then b)'' to ''A'', ''not-C''), and that it cannot explain the validity of some clearly valid entailments.
Lycan then turns to ''A beautiful but false theory of 'even if'''. He begins by claiming that ''even'' in ''even if'' means just... ''even''. He distinguishes three views about ''even'', given that ''even'' strongly implies a contradiction of contextual presumptions: the minimal view, according to which ''even'' contributes nothing to the semantics of the sentence; the conventional view, where the unexpectedness is supposed to be a conventional (though non-truth-conditional) implicature; the semantic view, in which ''even'' does have a truth-conditional contribution to make. The last one is the one favoured by Lycan, who defends the view that ''even'' expresses ''a comparison of expectedness with a contextually indicated reference-class'' (100). The specificity of ''even if'' is that the meaning of a sentence where it occurs does not seem conditional anymore. This leaves the role of the antecedent open. The situation is however slightly more complicated, given that the consequent may be entailed or not depending on focus phenomena. Thus some ''even if'' conditionals entail their consequent, while others do not. In fact, the Event Theory answers these questions, and also answers the pragmatic question of why the utterer asserts ''Q even if P'' where ''Q'' might seem sufficient. This leaves two questions unanswered: the place of focus in the theory and the defence of the semantic view of ''even''. This leads Lycan back to his begining assumptions and to show that they can account for focus: notably the idea that ''even'' adds to conditionals both the widening of the reference class and a universal quantification over all its members. There are, however two different classes, the comparison class of conditions attached to the ''even if'' conditionals and the reference class of real circumstances attached to the corresponding ''bare'' conditionals. Though the second is often a subclass of the first, it is not always the case. This leads to a reformulation of the original formulation for ''C even if A'', where A is explicitly included in the comparison class. This leaves Lycan with the defense of the semantic view of ''even if'', for which he gives three arguments: the rarity of semantically empty words in natural languages; evidence for the fact that ''even if'' involves universal quantification; the linguistic similarity between ''even'' and ''only'', which are ''syntactic soulmates'' (112) and ''logical contraries'' (113).
Lycan then turns to counterexamples to that theory which lead him to ''An Unbeautiful but less easily refutable theory of 'Even If'''.He begins with four apparent counterexamples to the theory, which drive him to definetively drop both the Reality Requirement and Modus Ponens. There is a more serious objection to the universal quantifier theory of ''even'' and that is that ''even'' seems to allow of exceptions (e.g. ''I'll eat anything on pizza, even squid or bull's testicles, but not a brick or a crowbar''). This is the problem of the contrast within the reference class. Weakening the quantifier to a less than universal reading (''many'' for instance) would not however be satisfying. A way of preserving the universal analysis is to consider the comparison class as encompassing everything ''within reason''. This however seems rather ad hoc. Another way of saving the universal hypothesis is to take ''even'' as meaning not ''every...including...'' but ''every... plus...''. This means that the Consequent-Entailment problem fades away: on this analysis, ''Q even if P'' does not entail ''Q'', though the assertion of ''Q even if P'' does most often amount to an assertion of ''Q'', notably when the reference class is a subclass of the comparison class.
Lycan then turns to ''The 'indicative'/'Subjunctive' distinction''. Despite the great similarities between indicative and subjunctive conditionals, there are some cases where the indicatives and the corresponding subjunctives differ in truth-values. Lycan begins by noting that the difference is not a matter of grammatical moods and that the terminology is, therefore, ill-chosen. He substitutes to it ''straight'' and ''boxarrow'' conditionals, noting that some so- called indicative conditionals are not conditionals at all and should be treated independently. After a reminder of some theories of conditionals (Adams, Lewis, Stalnaker, etc.), he turns to the distinction in the event theory. This turns out to be, quite simply,a difference in reference classes, indicated through ''lexical presumption''. This means that a straight conditional and its boxarrow counterpart share their logical forms but differ in the the values their parameters take, and thus in truth-values. More precisely, a conditional is straight when its utterer holds fixed a salient fact in his epistemic field, while it is boxarrow if the utterer neglects contextual facts and considers a wider range of possibilities. The Law of Conditional Excluded Middle (CEM) fails for boxarrows and, according to Lycan it also fails for straights. Finally, the straight/boxarrow distinction does not seem to apply to future conditionals.
Finally, Lycan turns to a well-known example, ''The Riverboat Puzzle'' in which (Simple version) a henchman signals to one of players what the cards of the other player are, then leaves. He utters two conditionals, one straight, one boxarrow. In the anomalous version, a second henchman, better informed (he has seen the hands of both players), utters conditionals with the same antecedents but the opposing consequents. Considering only the straight versions, it is claimed that they are both true and that it seems that the law of Conditional Noncontradiction does not hold. This has been used to argue for NTV. Lycan counterargues that in fact the first henchman's conditional is a backtracker, evaluated by holding fixed the present actual fact which is the most jarring relative to the counterfactual and adjusting other facts to make the counterfactual true. Backtrackers are not incompatible with ordinary conditionals and hence the law of conditional noncontradiction is not violated.
There are two appendixes to the book, the first one conjointly written by Lycan and Geis and the second one by Lycan alone, both on Nonconditional Conditionals (NCCs). In the first one, Geis and Lycan begin by justifying the claim that some apparently conditional sentences are not, in fact, conditional (e.g. Austin's example: ''There are biscuits on the sideboard if you want them'') by showing that they do not satisfy a list of syntactic, semantic or pragmatic features which are typical of conditionals. This raises two questions: what is the function of antecedents of non-conditional conditionals and how is it that non-conditional conditionals are interpreted as such, without any need for disambiguation from bona fide conditionals? A tentative answer to the first question is that consequents are indeed what is asserted in NCCs, while antecedents articulate some ''felicity condition'' of the consequent assertion, though sincerity conditions are excluded. This raises a new question. Why should this be expressed through a ''phony antecedent adverbial''? Another tentative answer to that new question is that the antecedents of NCCs, as the antecedents of bona fide conditionals, make explicit possibilities. Geis and Lycan then lists some less obvious types of NCCs which form a continuum but makes it even less easy to link NCCs with regular conditionals. In the second appendix, Lycan comes back to the problem of the semantics of NCCs, suggesting that NCCs could be accomodated in the Event Theory if the reference class is taken to include no non-Q event. This answers the third and fourth of the queries above, the second one being tackled through pragmatic considerations. Lycan's pragmatic answer is that NCC antecedents are metalinguistic and thus clearly not bearing a conditional relation to the consequent, which excludes NCCs being perceived as ambiguous.
CRITICAL EVALUATION Lycan began his career as a philosopher of language, has become best known in the past few years as a philosopher of mind and this book comes as a reminder that he still is a foremost philosopher of language. Lycan's book is extremely interesting and seems quite successful in combining the insights of both the Ramsey Test and Similarity Theory in a unified truth-conditional theory of conditionals. It also seems to take satisfactorily into account the syntactic features so often neglected by logicians. It does not eschew the difficulty which non-conditional conditionals have put in the path of semantic theories of conditionals. The book is thus well worth reading and indeed should, I think, be read by anyone interested in the subject.
However, I want to insist on the fact that the book is clearly not for beginners: extended knowledge of the field is needed. It should also be said that it is a comparatively short book (210 pages, including the appendixes) and that it has a high level of content, making for a rather dense, though quite clear, book. I think it might have been easier to read of it had been some fifty pages longer. However, it is a must for any one interested in conditionals and conditional reasoning.
Lewis, D. (1973), Counterfactuals, Cambridge, Mass., Harvard University Press.
Stalnaker, R. (1968), ''A theory of conditionals'', in N. Rescher (ed.), Studies in Logical Theory, Oxford, Blackwell.
ABOUT THE REVIEWER:
ABOUT THE REVIEWER Anne Reboul is a First Class Research Fellow at the French Center for Scientific Research (CNRS) in France. She has a Ph.D. in Linguistics (EHESS, Paris) and a Ph.D. in philosophy (University of Geneva, Switzerland). She has written some books, among which an Encyclopaedic Dictionary of Pragmatics and quite a few papers in French and English, on pragmatics and/or philosophic subjects.