Rational Rules Towards A Theory of Moral Learning Shaun Nichols Full Chapter PDF

Rational Rules: Towards a Theory of
Moral Learning Shaun Nichols

Visit to download the full and correct content document:
https://ebookmass.com/product/rational-rules-towards-a-theory-of-moral-learning-sha
un-nichols/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
The Evolution Of Moral Progress: A Biocultural Theory

Allen Buchanan
https://ebookmass.com/product/the-evolution-of-moral-progress-a-
biocultural-theory-allen-buchanan/
From Deep Learning to Rational Machines [converted]

Cameron J. Buckner
https://ebookmass.com/product/from-deep-learning-to-rational-
machines-converted-cameron-j-buckner/
From Deep Learning to Rational Machines 1st Edition

Cameron J. Buckner
https://ebookmass.com/product/from-deep-learning-to-rational-
machines-1st-edition-cameron-j-buckner/
Einstein's Unfinished Dream: Practical Progress Towards

a Theory of Everything 1st Edition Don Lincoln
https://ebookmass.com/product/einsteins-unfinished-dream-
practical-progress-towards-a-theory-of-everything-1st-edition-
don-lincoln/
Phenomenology Shaun Gallagher
https://ebookmass.com/product/phenomenology-shaun-gallagher/
Classics of Moral and Political Theory 5th Edition,

(Ebook PDF)
https://ebookmass.com/product/classics-of-moral-and-political-
theory-5th-edition-ebook-pdf/
Moral Error Theory 1st Edition Wouter Floris Kalf
https://ebookmass.com/product/moral-error-theory-1st-edition-
wouter-floris-kalf/
A Translation Theory of Knowledge Transfer: Learning

Across Organizational Borders Kjell Arne Røvik
https://ebookmass.com/product/a-translation-theory-of-knowledge-
transfer-learning-across-organizational-borders-kjell-arne-rovik/
The Uses of Delusion: Why It's Not Always Rational to

be Rational Stuart Vyse
https://ebookmass.com/product/the-uses-of-delusion-why-its-not-
always-rational-to-be-rational-stuart-vyse/
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 15/12/2020, SPi
Rational Rules
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Nichols, Shaun. Rational Rules : Towards a Theory of Moral Learning, Oxford University Press USA - OSO,
Rational Rules
Towards a Theory of Moral Learning
SHAUN NICHOLS
1
3
Great Clarendon Street, Oxford, OX2 6DP,
United Kingdom
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide. Oxford is a registered trade mark of
Oxford University Press in the UK and in certain other countries
© Shaun Nichols 2021
The moral rights of the author have been asserted
First Edition published in 2021
Impression: 1
All rights reserved. No part of this publication may be reproduced, stored in
a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by licence or under terms agreed with the appropriate reprographics
rights organization. Enquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above
You must not circulate this work in any other form
and you must impose this same condition on any acquirer
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data

Data available
Library of Congress Control Number: 2020941874
ISBN 978–0–19–886915–3
DOI: 10.1093/oso/9780198869153.001.0001
Printed and bound by
CPI Group (UK) Ltd, Croydon, CR0 4YY
Links to third party websites are provided by Oxford in good faith and
for information only. Oxford disclaims any responsibility for the materials
contained in any third party website referenced in this work.
For Sarah and Julia

Contents
Preface ix
Acknowledgments xiii
List of Figures xv
I. RATIONALITY AND RULES

1. Rationality and Morality: Setting the Stage 3
2. The Wrong and the Bad: On the Nature of Moral
Representations 25
II. STATISTICAL LEARNING OF

NORM SYSTEMS
3. Scope 49
4. Priors 82
5. Closure 95
6. Status 109
III. PHILOSOPHICAL IMPLICATIONS

7. Moral Empiricism 139
8. Rational Rules and Normative Propriety 164
9. Rationalism, Universalism, and Relativism 192
10. Is It Rational to Be Moral? 211
References 227
Index 245
Preface
My first book in moral psychology, Sentimental Rules, emphasized the role

of emotions in moral judgment. But I never thought emotions exhausted
moral judgment. There are numerous features of moral judgment that are
hard to explain just by appealing to emotions. Why do we tend to think that
it’s wrong to produce a bad consequence, but not wrong (or not as wrong) to
tolerate such a consequence happening? How do we come to think that
some evaluative claims are universally true but others only relatively true?
What kinds of rules can be learned? How do we determine whether some
novel act is permitted or prohibited?
These are questions that arise for moral psychologists and experimental
philosophers. Most work in these areas aims to uncover the processes and
representations that guide judgments. This is the agenda in discussions
about whether people are incompatibilists about free will (e.g., Murray &
Nahmias 2014), whether moral judgment is driven by distorting emotions
(e.g., Greene 2008), and whether judgments about knowledge are sensitive to
irrelevant details (e.g., Swain et al. 2008). Much less attention has been paid
to historical questions about how we ended up with the representations
implicated in philosophically relevant thought. There are different kinds of
answers to these historical questions. One might offer distal answers that
appeal to the more remote history of the concept. For instance, an evolu-
tionary psychologist might argue that some of our concepts are there
because they are adaptations. Or a cultural theorist might argue that some
of our concepts are there because they played an important role in facilitat-
ing social cohesion. On the more proximal end of things, we can attempt to
determine how the concepts might have been acquired by a learner. Those
proximal issues regarding acquisition will be the focus in this book.¹ I will
argue that we can explain many of the features of moral systems in terms of
¹ Of course proximal and distal issues are not unrelated. For an evolutionary psychologist,
the proposal that a concept is an adaptation will typically be accompanied by the expectation
that the characteristic (proximal) development of the concept is not explicable in terms of
domain-general learning mechanisms (see, e.g., Tooby & Cosmides 1992).
x 
rational learning from the evidence. To locate this in contemporary moral

psychology, a bit of background is in order.
In naturalistic moral psychology, sentimentalism is the dominant view
(e.g., Blair 1995; Greene 2008; Haidt 2001; Nichols 2004c; Prinz 2007), and
there is considerable evidence that emotions have numerous influences on
our moral psychology. Emotions seem to impact our judgments about moral
dilemmas (e.g., Bartels & Pizarro 2011; Koenigs et al. 2007). Emotions seem
to influence the resilience of certain moral rules (Nichols 2004c). Emotions
seem to motivate prosocial behavior (Batson 1991). Emotions seem to
motivate punishment for cheaters (Fehr & Gächter 2002). Sentimentalists
have drawn on these results to argue for philosophical conclusions. To take
what is perhaps the most prominent example, the impact of emotion on
certain kinds of moral judgments has been used to challenge the rationality
of those judgments (e.g., Greene 2008; Singer 2005).
I have counted myself among the sentimentalists, but I’ve also argued that
emotional reactions don’t provide a complete explanation of moral judg-
ment. In particular, I’ve argued that rules play an essential role in our moral
psychology (Nichols 2004c). However, I had no account of how we come to
learn these rules. Many moral rules seem to trade on subtle distinctions. For
instance, from a young age, children treat harmful actions as worse than
equally harmful omissions. Children also judge that it’s wrong to harm one
person to save five others from harm. Children are never explicitly taught
the distinctions to which these judgments conform.
The prevailing explanation for how we come to have such subtle distinc-
tions is nativist. Contemporary moral nativists hold that the best explan-
ation for the uniformity and complexity of moral systems is that moral
judgments derive from an innate moral acquisition device (e.g., Harman
1999; Mikhail 2011). Such views hold that the moral systems we have are
partly constrained by human nature. Just as linguistic nativism proposes
constraints on possible human languages, moral nativism implies that there
are constraints on possible human moralities (Dwyer at al. 2010). Although
nativist accounts have been widely criticized (e.g., Nichols 2005; Prinz 2008;
Sterelny 2010), there has been no systematic alternative explanation for how
children acquire such apparently complex moral systems.
My collaborators and I have been developing such an alternative explan-
ation for the acquisition of moral systems. The inspiration comes from an
unlikely source—statistical learning. Recent cognitive science has seen the
ascendance of accounts which draw on statistical learning to explain how we
end up with the representations we have (Perfors et al. 2011; Xu et al. 2012).
 xi
I’ve come to think that statistical learning provides a promising avenue for
answering central questions about how we come to have the moral repre-
sentations we do.
I will argue that a rational learning approach can explain several aspects
of moral systems, including (i) how people learn to draw the act/allow
distinction given limited evidence, (ii) how people come to have a bias in
favor of act-based rules, and (iii) how people use consensus information as
evidence on whether a moral claim is universally true.
The picture that emerges reveals a starkly different side of moral systems
than traditional sentimentalism. The learning processes invoked are, by
standard accounts, rational. This insulates moral judgment from important
charges of irrationality. For instance, if our deontic judgments depend on
rules, and these rules are acquired via rational inference, then we cannot
fault the process by which the judgment is made. This doesn’t insulate the
judgments from every critique. For instance, the rules themselves might be
defective. But that challenge requires a deeper inquiry into the epistemic
credentials of the rules.
The resulting account also contrasts sharply with nativism. The learning
processes that I will draw on are not specific to the moral domain. Indeed,
statistical learning affords the moral psychologist a diverse empiricist tool-
kit. Moreover, the rational learning account suggests that humans are
flexible moral learners, with no innate constraints on the kinds of rules

that humans can learn.
The view that I defend is obviously rationalist in important ways. But that
doesn’t entail a rejection of the significance of emotions for moral judgment.
Indeed, I continue to think that much of the sentimentalist picture is correct.
Emotions play a critical role in amplifying the rules of morality. This
plausibly holds for online decision-making—rules that resonate with strong
emotions will end up having a greater influence in our decision-making. The
emotional amplification of rules also likely explains the cultural resilience of
certain moral rules. To ignore these influences of emotions is to ignore
fundamental aspects of human morality. A persistent commitment of sen-
timentalists down the ages is that without the emotions, we would have
radically different normative systems than we do. I certainly don’t mean to
retreat from that sentimentalist commitment. However, the fact that emo-
tions are critical to our moral systems doesn’t mean that the role of ration-
ality is negligible. On the contrary, I’ll argue, rational learning provides a
much better explanation than emotions for how we acquire normative
systems in all their complexity. The ultimate view, I think, must be some
xii 
form of rational sentimentalism, where rational learning and emotions both

contribute in key ways to our moral judgments. But in this volume, I want to
emphasize the rational side of things. Although moral judgment and deci-
sion might be distorted in many ways, there’s reason to be optimistic that the
fundamental capacity for acquiring moral rules is rational and flexible. The
way we learn rules is plausibly responsive to the evidence in appropriate
ways, and, at least at some developmental stages, supple enough to adjust to
new rules in the face of new evidence.
Acknowledgments
First, I’d like to thank my collaborators on the empirical studies that form
the center of this book: Alisabeth Ayars, Hoi-Yee Chan, Jerry Gaus, Shikhar
Kumar, Theresa Lopez, and Tyler Millhouse. I owe special debts to Theresa,
Alisabeth, and Jerry. Theresa’s dissertation (Lopez 2013) is the first work that
maintained that Bayesian approaches to cognition might provide an alterna-
tive to Chomskyan accounts of moral cognition. If it hadn’t been for Theresa’s
insightful dissertation, I never would have started a project on moral learning.
Alisabeth worked extensively on the project when she was a graduate student
in psychology at Arizona. She had several key experimental ideas; she was also
incisive on the theoretical issues (as evidenced in Ayars 2016). This project
would have been much worse without her contributions. Finally, Jerry was an
ideal collaborator on the empirical work that we did together. More generally,
Jerry has been an intellectually invigorating colleague and friend. It was my
good fortune to be in the same department with him.
Many friends and colleagues have influenced my thinking on these
matters through conversations, discussions in Q&A, and comments on

some of the chapters. In particular, I’d like to thank Mark Alfano, Ritwik
Agrawal, Cristina Bicchieri, Thomas Blanchard, Selmer Bringsjord, Mike
Bruno, Stew Cohen, Juan Comesaña, Fiery Cushman, Justin D’Arms, Colin
Dawson, Caleb Dewey, John Doris, LouAnn Gerken, Josh Greene, Steven
Gross, Heidi Harley, Toby Handfield, Dan Jacobson, Jeanette Kennett, Max
Kleiman-Weiner, Josh Knobe, Max Kramer, Tamar Kushnir, Sydney Levine,
Jonathan Livengood, Don Loeb, Tania Lombrozo, Edouard Machery,
Bertram Malle, Ron Mallon, Eric Mandelbaum, John Mikhail, Adam
Morris, Ryan Muldoon, Scott Partington, Ángel Pinillos, Dave Pizarro,
Jesse Prinz, Hannes Rakoczy, Peter Railton, Sarah Raskoff, Chris
Robertson, Connie Rosati, David Rose, Adina Roskies, Richard Samuels,
Hagop Sarkissian, Sukhvinder Shahi, Dave Shoemaker, David Sobel, Tamler
Sommers, Kim Sterelny, Justin Sytsma, Josh Tenenbaum, John Thrasher,
Hannah Tierney, Mark Timmons, Bas Van Der Vossen, Steve Wall, Jen
Wright, Jonathan Weinberg, David Wong, and Fei Xu.
All of the empirical studies for this project were funded in part by the
U.S. Office of Naval Research under award number 11492159. I’m grateful
xiv 
to Paul Bello, who was the ONR program officer, for supporting the work, as
well as for numerous helpful discussions about it.
Chapter 6 draws substantially on material from Ayars & Nichols (2020),
Rational learners and metaethics, Mind & Language, 35(1), 67–89. I thank
the journal for permission to reprint that material here.
I spent academic year 2017–18 on fellowship at the Center for Human
Values at Princeton. I’m grateful to the Center and to the University of
Arizona for affording me the opportunity to focus on writing the book. In
addition to freeing up time to write, I got excellent feedback from many
people at the Center, including Stephanie Beardman, Mitch Berman, Liz
Harman, Dylan Murray, Drew Schroeder, Amy Sepinwall, Peter Singer,
Michael Smith, Monique Wonderly, and especially Mark van Roojen. Mark
read and commented on much of the book while I was there, and he’s been a
tireless and wonderful correspondent about these issues ever since.
Walter Sinnott-Armstrong arranged to have his research group, Madlab,
read the first draft of the manuscript. This was incredibly helpful. I’m
grateful to all the lab members for taking the time to read and think about
the manuscript. I’d like to single out several people in the group whose
comments led to changes in manuscript: Aaron Ancell, Jana Schaich Borg,
Clara Colombatto, Paul Henne, J. J. Moncus, Sam Murray, Thomas
Nadelhoffer, Gus Skorburg, Rita Svetlova, and Konstantinos Tziafetas.
Mike Tomasello also participated, which was a delight. And of course I’m
especially indebted to Walter, both for organizing the event and for being
characteristically constructive and indefatigable. Dan Kelly also read and
gave terrific comments on the entire manuscript at a later stage. His careful
attention led to numerous improvements in book.
I had the benefit of three excellent referees for OUP, one of whom was
Hanno Sauer (the other two remained anonymous). Thanks to all of them, and
to Peter Momtchiloff for his characteristically excellent stewardship at OUP.
Victor Kumar first encouraged me to write this book. The book turned
out to be a lot more work than I expected, but I still thank Vic for prompting
me to write it, and for excellent comments along the way. Michael Gill and
I have been discussing issues at the intersection of moral philosophy and
cognitive science for twenty years, and his influence and encouragement has
been central to this work. Finally, I’m lucky to have been able to talk with
Rachana Kamtekar about every sticky philosophical problem in the book,
and everything else besides.
List of Figures
1.1 Stimuli for probabilistic inference task 18

3.1 Potential scopes of rules 53
3.2 Hierarchical representation of hypothesis space, based on similarity
judgments 54
3.3 Representing the sizes of hypotheses 57
3.4 Extension of words represented as subset structure 59
3.5 Sample violations of novel rules 62
3.6 Hypothesis space for Principle of Double Effect 67
3.7 Results on intentional/foreseen study 69
3.8 Alternative hypothesis space for Principle of Double Effect 70
3.9 Set of potential patients for a rule 76
3.10 Schematic depiction of display for parochial norms study 78
3.11 Schematic depiction of display for parochial norms study,
20 percent condition 79
4.1 Complete list of violations for overhypothesis study 90

5.1 The rectangle game 99
6.1 Universalist and relativist models for different patterns of consensus 116
6.2 Correlation between perceived consensus and judgments
of universalism 119
6.3 Results on universalism/relativism for abstract cases, by domain 124
6.4 Different relativist models for split consensus 131
7.1 Empiricist model of learning 144
7.2 Hypothesis space for scope of rules 160
9.1 The Stag Hunt 200
9.2 Choosing sides 200
PART I
RATIONALITY AND RULES
1
Rationality and Morality
Setting the Stage
“Moral distinctions are not derived from reason.” Thus does Hume begin his
discussion of morality in the Treatise. Rather, Hume says, moral distinctions
come from the sentiments. Contemporary work in moral psychology has
largely followed Hume in promoting emotions rather than reason as the
basis for moral judgment (e.g., Blair 1995; Greene 2008; Haidt 2001; Nichols
2004c; Prinz 2007; cf. May 2018; Sauer 2017). While I think moral judgment
is tied to emotions in multiple ways, in this book I want to explore the
rational side of moral judgment. I’ll argue that rational processes play a
critical and underappreciated role in how we come to make the moral
judgments we do. In this chapter, I’ll describe the basic phenomena that
I want to illuminate with a rational learning account, and I will explicate the
primary notion of rationality that will be in play.
1. The Phenomena
Don’t lie. Don’t steal. Keep your promises. These injunctions are familiar
and central features of human moral life. They form part of the core
phenomena to be explained by an adequate psychological account of
moral judgment. Why do we make the judgment that it’s wrong to lie or
steal?
In addition to these specific judgments, an adequate moral psychology
must also explain important distinctions that seem to be registered in lay
moral judgment. For example, people tend to think that producing a bad
consequence is worse than allowing the consequence to occur. Much of the
work attempting to tease out an implicit understanding of these distinctions
is done using trolley cases (Foot 1967; Greene et al. 2001; Harman 1999;
Mikhail 2011; Thomson 1976, 1985). For instance, people tend to say that in
the following case, what the agent does is not permissible.
4  
Footbridge: Is it permissible for Frank to push a man off a footbridge and

in front of a moving boxcar in order to cause the man to fall and be hit by the
boxcar, thereby slowing it and saving five people ahead on the tracks?
By contrast, people tend to say that what the agent does (or rather fails to do)
is permissible:
Footbridge-Allow: Is it permissible for Jeff not to pull a lever that would

prevent a man from dropping off a footbridge and in front of a moving
boxcar in order to allow the man to fall and be hit by the boxcar, thereby
slowing it and saving five people ahead on the tracks?
People also tend to say that what the agent does in the following case is
permissible:
Bystander: Is it permissible for Dennis to pull a lever that redirects a

moving boxcar onto a side track in order to save five people ahead on the
main track if, as a side-effect, pulling the lever drops a man off a footbridge
and in front of the boxcar on the side track, where he will be hit? (Cushman
et al. 2006: 1083–4)
These cases have been taken to suggest that people are sensitive to surpris-
ingly subtle distinctions in their normative evaluations.
If people really are sensitive to these distinctions in their moral judg-
ments, these are relatively high-level psychological phenomena. At an even
higher level, we find that people seem to have systematic judgments about
the nature of morality itself. For instance, people tend to think that moral
claims have a different status than conventional claims. This has been
explored extensively with questions like the following:
Authority dependence: If the teacher didn’t have a rule against hitting,

would it be okay to hit other students?
For actions like hitting, people, including pre-school children, tend to say
that it’s wrong to hit even if the teacher doesn’t have a rule. But for actions
like talking during story-time, people are more likely to say that if the
teacher doesn’t have a rule on the matter, it’s okay to talk during story
time (e.g., Nucci 2001; Turiel 1983). More recently, people have explored the
  :    5
extent to which people think moral claims are universally true, using
questions like the following:
Disagreement: If John and Mark make different judgments about whether

it’s okay to rob a bank, does one of them have to be wrong?
For actions like bank robbery and assault, people tend to say that if two
people make different judgments, one of them has to be wrong, but they do
not tend to say this when it comes to aesthetic claims or matters of taste
(Goodwin & Darley 2008; Nichols 2004a; Wright et al. 2013).
These are the phenomena that I want to investigate. Note that much of
our moral lives is not included here. I won’t try to explain our aversion to
suffering in others, our propensity to guilt and shame, or our use of empathy
and perspective taking in moral assessment. Nor will I try to characterize the
ethical abilities enjoyed by non-human animals. The moral capacities that
I’m targeting are, as far as we can tell, uniquely human. How do we arrive at
these sophisticated judgments, distinctions, and meta-evaluations?
2. A Challenge to Common-Sense Morality

Before setting out my positive story, I want to address briefly the prevailing
skepticism about moral judgment. Moral psychologists often cast lay mor-
ality as critically flawed. There is evidence that moral judgment is com-
promised by incidental emotions, misleading heuristics, and confabulation.
Philosophers have used such evidence to develop debunking arguments
according to which key areas of common-sense ethical judgment are epi-
stemically rotten—they are based on epistemically defective processes (see
Sauer 2017 for discussion of debunking arguments). Debunking arguments
have been developed for both common-sense normative ethics and
common-sense metaethics.
2.1 Debunking Normative Ethics
Perhaps the most familiar debunking accounts draw on dual process theories,
according to which there are two broad classes of psychological processes. System
1 processes tend to be fast, effortless, domain specific, inflexible, insensitive to
new information, and generally ill-suited to effective long-term cost–benefit
6  
reasoning. System 2 processes are flexible, domain general, sensitive to new

information, and better suited to long-term cost–benefit analysis, but they are
also slow and effortful.
On Greene’s dual process account of moral judgment, when we are
presented with the option of pushing one innocent person off of a
Footbridge to save five other innocent people, there is competition between
a System 1 emotional process (screaming “don’t!”) and a System 2 process
that calculates the best consequence (saying “5 > 1, dummy”). The proposal
is that cases like Footbridge trigger System 1 emotions that subvert System 2
utilitarian cost–benefit analysis.
A closely related dual process model comes from Jonathan Haidt (2001).
On Haidt’s social intuitionist account, our moral reactions tend to be driven
by System 1 affectively valenced intuitions. System 2 plays a subsidiary
role—it primarily generates post hoc justifications for our affective intu-
itions (2001: 815). One of the key studies that motivates Haidt’s view
suggests that people will hold on to their moral views even when they are
unable to provide a justification for them. For instance, participants were
presented with a vignette in which siblings Julie and Mark have a consensual
and satisfying sexual encounter, using multiple forms of birth control:
Julie and Mark: Julie and Mark are brother and sister. They are traveling
together in France on summer vacation from college. One night they are
staying alone in a cabin near the beach. They decide that it would be
interesting and fun if they tried making love. At the very least it would be
a new experience for each of them. Julie was already taking birth control
pills, but Mark uses a condom too, just to be safe. They both enjoy making
love, but they decide not to do it again. They keep that night as a special
secret, which makes them feel even closer to each other.
What do you think about that? Was it OK for them to make love?
When presented with this vignette, most participants said that it was not
okay for Julie and Mark to make love. When asked to defend their answers,
participants often appealed to the risks of the encounter, but the experi-
menter effectively rebutted the justifications (e.g., by noting the use of
contraceptives). Nonetheless, the participants continued to think that the
act was wrong, even when they couldn’t provide any undefeated justifica-
tions. A typical response was: “I don’t know, I can’t explain it, I just know it’s
wrong” (Haidt 2001: 814). Haidt interprets this pattern as a manifestation of
  :    7
two processes: the moral condemnation is driven by an affective intuition

(rather than reasoning) and the proffered justification comes from post hoc
rationalizing—confabulation.
As Greene and Haidt characterize System 1 processes, the judgments that
issue from those processes are unlikely to be responsive to evidence. Greene
argues that if System 1 is indeed what leads people to judge that it’s wrong to
push in cases like Footbridge, this provides the foundation for an argument
that challenges the rational propriety of non-utilitarian judgment. Greene
suggests that deontological judgments, like “it’s wrong to push the guy in
front of the train,” are defective because they are insensitive to rational
considerations, in sharp contrast with consequentialist evaluations:
[T]he consequentialist weighing of harms and benefits is a weighing

process and not an ‘alarm’ process. The sorts of emotions hypothesized
to be involved here say, ‘Such-and-such matters this much. Factor it in.’
In contrast, the emotions hypothesized to drive deontological judgment are
. . . alarm signals that issue simple commands: ‘Don’t do it!’ or ‘Must do it!’
While such commands can be overridden, they are designed to dominate
the decision rather than merely influence it. (Greene 2008: 64–5)
Greene maintains that since our deontological judgments derive from emo-
tional reactions that are not responsive to rational considerations, we should

ignore them in normative theorizing (2008; see also Singer 2005: 347).¹
Although there is a diverse array of evidence supporting the view that
emotions play a role in judgments about Footbridge (e.g., Amit & Greene
2012; Bartels & Pizarro 2011; Koenigs et al. 2007), emotions cannot provide
a complete explanation for the basic phenomenon of non-utilitarian moral
judgment. Many dilemmas that people rate as generating very little emo-
tional arousal—e.g., those involving lying, stealing, and cheating—elicit
non-utilitarian responses (see, e.g., Dean 2010). Consider, for instance,
cases of promise breaking. People don’t get emotionally worked up by
vignettes that involve promise breaking, but they still make non-utilitarian
judgments about promise breaking. For instance, in one study, participants
were asked whether it was okay for one person to break a promise in order to
prevent two other people from breaking promises; in this case people
maintained that it was wrong for the first person to break a promise even
¹ For direct responses to this argument, see Berker (2009) and Timmons (2008).
8  
though it would minimize promise breaking overall (Lopez et al. 2009: 310).
So emotion doesn’t seem to be required to make non-utilitarian judgments.
Indeed, the asymmetry between Footbridge and Bystander is found even
when the potential human victims are replaced by teacups (Nichols &
Mallon 2006).²
The fact that people make non-utilitarian judgments in the absence of
significant affect indicates that there must be some further explanation for
these responses. This undercuts debunking arguments that depend on the
view that non-utilitarian judgments are primarily produced by arational
emotional reactions. The fact that we find non-utilitarian judgments without
concomitant affect also exposes the need for a different explanation for the
pattern of non-utilitarian judgment that people exhibit.
2.2 Debunking Metaethics
As noted above, people tend to think that at least some moral claims are
universally true, and they treat aesthetic claims as only true relative to an
individual or group (Goodwin & Darley 2008, 2012). Why is this? Why do
people believe of some moral claims that they are universally true?
Philosophers have offered several explanations for the belief in universalism,
and the best-known proposals serve as debunking explanations. In his

influential treatment, Mackie (1977) proposes a number of non-rational
explanations for the belief in universalism. One idea is that motivational
factors, like the desire to punish or compete, play a distorting role in
generating metaethical judgments (see, e.g., Mackie 1977: 43; see also
Fisher et al. 2017; Rose & Nichols 2019). Another of Mackie’s suggestions
is that the belief in universalism derives from the tendency to project our
moral attitudes onto the world. Relatedly, our emotional reactions toward
ethical violations may persuade us that moral wrongs are universally wrong.
The most direct attack on the propriety of metaethical judgments comes
from a study by Daryl Cameron and colleagues (2013). They presented
subjects with brief descriptions of practices in other cultures (e.g.,
“Marriages are arranged by the children’s parents”). In some cases, these
descriptions were presented on a background displaying a disgusting picture
(unrelated to the content of the description); in other cases, the background
² In addition, recent work indicates that Bystander is just as emotionally arousing as

Footbridge (Horne & Powell 2016).
  :    9
was emotionally neutral. Cameron and colleagues found that when the
description was accompanied by a disgusting picture, participants were
more likely to give universalist responses.³ Such an influence is plausibly
epistemically defective. Cameron and colleagues make this clear by drawing
on the distinction between incidental and integral effects of emotions:
Integral emotions may contain information that should appropriately

influence moral judgments: guilt may signal that you have behaved badly
towards others, and anger may signal that others have behaved badly towards
you (Frank 1988). In contrast, incidental emotions are conceptually
unrelated to subsequent judgments, and so are ethically irrelevant
(Doris & Stich 2005). Whereas incidental emotions may influence moral
judgments, they are not appropriately cited as evidence in the justification
of these judgments (719).
If you are more universalist about arranged marriages because you are
seeing a revolting picture of worms, then you’re being swayed by an epi-
stemically defective process.
I focus on the study by Cameron and colleagues because it has a clean
experimental design, and it provides some of the most direct evidence for
the role of an epistemically defective affective process in judgments of
universalism. However, there is a pressing limitation of the study.

Although the results indicate that there is some influence of epistemically
defective processes, the extent of influence is, for debunking purposes,
trivial. The mean difference in universalist judgments produced by inducing
disgust was only 0.1 on a 5-point scale.⁴ Thus, the strongest debunking
conclusion this study can fund is: “To some slight extent, people are not
justified in their belief that a claim is universally true.” Clearly, we cannot
take these results to show that people’s belief in moral universalism is largely
based on a defective process. The results simply don’t explain much of why
people think moral claims are universally true. As a result, they don’t do
much by way of debunking the belief.
³ Cameron and colleagues used a slightly different universalism measure than the standard
disagreement measure (Section 1). They asked participants to evaluate whether an activity
practiced in other cultures (e.g., “Marriages are arranged by the children’s parents”) is wrong
regardless of the culture in which it is practiced.
⁴ More generally, it turns out that the impact of occurrent emotion on moral judgment is
quite weak (e.g., Landy & Goodwin 2015; May 2014, 2018).
10  
I’ve argued that some of the most prominent debunking arguments are
inadequate. Obviously this is a limited selection of the debunking arguments
that have been made. There is a broader lesson here, though. The most
prominent kinds of arguments that purport to debunk lay ethical judgments
appeal to the distorting effects of occurrent emotions. But many of the ethical
judgments that we want to understand do not seem to be explained by
occurrent emotional processes (see also Landy & Goodwin 2015; May 2014,
2018). So I think there is good reason to be skeptical of the attacks on lay moral
judgment. However, skepticism about these accounts hardly constitutes a
positive defense. Even if the extant debunking arguments fail, that doesn’t
mean lay moral judgment is in good repair. The main work of this book is to
promote a detailed positive defense of the rationality of lay moral judgment.
3. Rationality
3.1 The Many Rationalisms
“Rationalism” is used in several strikingly different ways in philosophy. For

much of this book, the notion of rationality in play will be an evidentialist one
on which a person’s belief is rational or justified just in case it is supported by
their evidence. I will argue that several key distinctions in common-sense

morality are acquired through a process of rational inference based on the
evidence that the learner receives. I will set out the evidentialist notion of
rationality in a bit more detail below, but first, I want to chart several different
notions of rationalism which contrast with evidentialism in important ways.
In metaethics, “rationalism” is often used to refer to a view about the
relation between moral requirements and reasons for action. This action-
focused version of rationalism (sometimes called “moral/reasons existence
internalism”) holds that it is a necessary truth that if it is morally right for a
person to Φ then there is a reason for that person to Φ (Smith 1994; van
Roojen 2015).⁵ This view of rationality, unlike a pure evidentialist view, ties
⁵ Michael Smith distinguishes two versions of this rationalist thesis. The conceptual ration-
alist thesis holds that “our concept of a moral requirement is the concept of a reason for action; a
requirement of rationality or reason.” The substantive rationalist thesis holds that this concep-
tual claim bears out in the world. That is, “there are requirements of rationality or reason
corresponding to the various moral requirements” (1994: 64–5).
  :    11
rationality directly to action. Evidentialist approaches to rationality can, of

course, have rationality inform action, since one’s actions will and should be
guided by one’s rationally acquired beliefs. But evidential rationality only
applies to beliefs, not directly to desires or actions.
In cognitive science, “rationalism” is often used to pick out nativist
views in contrast to empiricist views (e.g., Chomsky 2009). These views
are emphatically not rationalist in the evidentialist sense (see, e.g.,
Mikhail 2011: 32). A leading idea of Chomskyan rationalism is precisely
that there are capacities the acquisition of which cannot be explained in
terms of inference over the evidence. For instance, Chomskyans main-
tain that children’s acquisition of grammar cannot be explained in terms
of children drawing apt inferences from the available linguistic data.
As we will see in Chapter 7, moral Chomskyans hold that there is a
moral grammar the acquisition of which can’t be explained in terms of
evidential reasoning. This will be at odds with much of the story that
I develop.
A third notion of rationalism, which prevailed in moral philosophy in the
Early Modern period, is an a priori notion of rationalism. Mathematics
served as the leading example here. The early moral rationalists maintained
that mathematical truths are a priori, and that many of these truths are self-
evident and accessible to us without relying on any kind of experiential
evidence. Similarly, the early moral rationalists (e.g., Clarke, Locke, Balguy)
held there are a priori moral principles that are self-evident, and these self-
evident principles provide the basis for deductive inferences to further moral
claims (see Gill 2007 for discussion). In Chapter 9, I will take up the relation
between this traditional moral rationalism and the more modest moral
rationalism that I’ll promote.
The evidentialist notion of rationality contrasts with all of the above, but
it is the dominant framework in analytic epistemology. According to the
kind of evidentialism that I’ll be using, S’s belief that P is rational or justified
to the extent that S’s belief that P is responsive to her total relevant evidence.
As I want to use the notion, a person’s belief can be responsive to the
evidence even if she lacks conscious access to the reasoning process. Of
course, when subjects do report their reasoning process, if the process they
report is a process that is responsive to the evidence, this gives us good
reason to think that they are in fact making judgments in ways that are—to
some extent—evidentially rational. Still, such conscious access is not
12  
necessary for responding to the evidence.⁶ Algorithms might be responsive

to the evidence even when unconscious.
In addition to its prominence in analytic epistemology, evidentialism also
coheres reasonably well with the notion of rationality that anchors discus-
sions in naturalized epistemology.⁷ Perhaps the best-known statement of
rationality in that literature comes from Ed Stein:
[T]o be rational is to reason in accordance with principles of reasoning that

are based on rules of logic, probability theory and so forth. If the standard
picture of reasoning is right, principles of reasoning that are based on such
rules are normative principles of reasoning, namely they are the principles
we ought to reason in accordance with. (Stein 1996: 4)
On this familiar description, one must reason “in accordance” with the
principles of probability theory to be rational.
What exactly is it to reason “in accordance” with the rules of logic and
probability theory? Much of the literature on reasoning is conspicuously
vague about this (for discussion, see Nichols & Samuels 2017). This much is
obvious, though: reasoning is a process. To clarify the nature of processes,
and the nature of reasoning processes in particular, we can draw on David
Marr’s influential account of levels of analysis in cognitive science.
3.2 Processes
Marr explains levels of analysis in terms of their guiding questions. The first
level is the computational level, and its basic questions are: “What is the goal
of the computation, why is it appropriate, and what is the logic of the
strategy by which it can be carried out?” The second level, representation
and algorithm, asks: “What is the representation for the input and output,
and what is the algorithm for the transformation?” The third level, that of
hardware implementation, won’t occupy us here, but it asks, “How can the
representation and algorithm be realized physically?” (Marr 1982: 25).
⁶ Insofar as evidentialism requires that beliefs are responsive to evidence, the view is typically
taken to be at odds with simple forms of reliabilism (for discussion, see Goldman & Beddor
2016). I won’t try to engage this issue in the book, but reliabilists might maintain that the
inferences that I promote are justified insofar as they are based on reliable processes.
⁷ Research on heuristics and biases has been used to challenge the idea that people are
rational in this evidentialist way (see, e.g., Stich 1990).
  :    13
Marr illustrates different levels of analysis with the example of a lowly

cash register. At the computational level, the process we find in a cash
register is addition, which is a function that takes two numbers as input
and yields one number, in ways specified by an arithmetic theory (22).
Addition is what the cash register does. The computational level also
concerns the question why addition is used for this device, rather than,
say, division or multiplication? Here, Marr says, we can draw on our
intuitions about what is appropriate for the context. In the context of
exchanging groceries for money, it seems most appropriate to sum the
costs of the items purchased (22–3). Intuitions are just one resource for
gleaning the why of psychological processes. Evolutionary psychologists
emphasize adaptationist considerations to explain why a certain process is
appropriate for a context (Cosmides & Tooby 1995: 1202).⁸
Marr’s second level of analysis involves the actual representations and
algorithms of the process (Marr 1982: 23). The computation of addition
can be carried out in different ways. One dimension of flexibility is the
representational system itself. One might use different kinds of symbols to
represent numbers, e.g., binary, Arabic, or hash marks. The other dimen-
sion of flexibility is the algorithms that are deployed. Importantly, the
kinds of algorithms that are appropriate will be constrained by the kinds of
representations in play. With hash marks, one can use concatenation for
addition (|||| + ||| = |||||||), but obviously this would be a disaster for Arabic
numerals (4 + 3 = 43?). Even if the representations are fixed as, say,
Arabic numerals, different algorithms can be used to carry out addition.
One common algorithm for addition mirrors the “carrying” algorithm
people learn in grade school—add the least significant digits, carry if
necessary, move left, and repeat. Another algorithm for addition uses
“partial sums,” separately adding the 1 place-value column, the 10 place-
value column, the 100 place-value column, and then summing these partial
sums. These are different processes at the algorithmic level but not at the
computational level.
⁸ Although Marr doesn’t mention it, sometimes we might identify what the process is, even if
the purpose is unclear or somehow inapt. Imagine that we observe the cash register receive two
inputs—$2 and $3—and it generates $6 as output; then, when another $2 item is entered, the
output is $12. Eventually it becomes clear that what the machine is doing is multiplication. Why
it’s doing multiplication is not because that’s the right process in this context (based on our
intuitions or evolution). Perhaps the machine was hacked or perhaps there’s a short circuit that
remapped + to *.
14  
3.3 Rational Processes
We can now recruit Marr’s distinctions to impose greater specificity in

characterizing rational processes.⁹ On one construal, for a system to be “in
accordance” with the rules of logic and probability theory is for the process
to execute algorithms that encode the rules of logic and probability theory.
So, for instance, if modus tollens is a proper rule of logic, and a given process
deploys a modus tollens algorithm to transition from ((P ! Q) and ~Q) to
~P, then that process is rational on the algorithmic construal of accordance.
In that case, we can say the transition from input to output is algorithmically
rational.¹⁰
Another way to understand “accordance” with the rules of logic and
probability theory is in terms of the computational-level description. On
that approach, a process accords with logic and probability theory when
what the process computes—the input–output profile of the process—
corresponds to the function of the relevant logico-probabilistic rules. In
that case, we can say that the process is computationally rational. As we
saw above, the computational-level description of a process is neutral about
the actual algorithms involved in the transition from input to output.¹¹
Often in cognitive science, what we really want to capture is the algorith-
mic level process. But it’s also the case that often we settle for less. We can
start by trying to show that a process is rational at the computational level,

with the hope that this can eventually be filled out with an algorithmic
⁹ This discussion is based on joint work with Richard Samuels (Nichols & Samuels 2017).
¹⁰ Richard Samuels and I distinguish strong and weak versions of the thesis that some
cognitive process is in algorithmic accordance with logic and probability theory (Nichols &
Samuels 2017). Strong algorithmic accordance requires that there is an isomorphism between
analytic probability theory and the inferential process being evaluated. The idea is that an
algorithm is only rational if it proceeds as prescribed by the mathematics of probability theory.
Samuels and I suggest that this looks to be an excessively demanding way to characterize
rational psychological processes. Instead, we argue for weak algorithmic accordance, which
allows that a process can be rational when the algorithm implements a good Bayesian approxi-
mation method (2017: 24–5). What makes for a good approximation method might vary by
context. Although I won’t discuss weak algorithmic accordance further in this book, the notion
of weak algorithmic accordance makes it rather easier for a psychological process to count as
rational. By contrast, strong algorithmic rationality imposes severe demands on the requirement
for rationality (2017: 22).
¹¹ Not all processes characterized at the computational level need conform to logic and
probability theory. For instance, part of the visual system of the housefly is described at the
computational-level as having the goal of landing (Marr 1982: 32–3); but it needn’t be the case
that the transition from perceptual stimulus to landing behavior in the housefly conforms to the
laws of probability.
  :    15
analysis (see, e.g., Xu & Tenenbaum 2007b: 270).¹² Making a case that moral
learning is characterized by a process that is rational at the computational
level is a step toward an algorithmic analysis, and often the most we can
hope for at this point is a computational analysis.¹³ However, for some of
the inferences that I will promote in this book, I will suggest that we do have
the beginnings of an algorithmic account, reflected, for instance, in the
explanations offered by the experimental subjects.
4. Statistical Learning
4.1 Statistical Learning
Some kinds of statistical learning involve extremely sophisticated techniques

which require enormous computational resources, but other kinds of stat-
istical learning are humble and familiar forms of inference. Imagine you’re
on a road trip with a friend and you have been sleeping while he drives. You
wake up wondering what state you’re in. You notice that most of the license
plates are Kansas plates. You can use this information to conclude that you are in
Kansas. This is a simple form of statistical learning. You are consulting samples
of license plates (the ones you see), and using a principle on which samples
reflect populations (in this case, the population of license plates). This, together
with the belief that Kansas is the only state with a preponderance of Kansas
plates warrants your new belief that you are in Kansas.
Early work on statistical reasoning in adults indicated that people are
generally bad at statistical inference (e.g., Kahneman & Tversky 1973). For
instance, people seem to neglect prior probabilities when making judgments
about likely outcomes. In a striking experiment, Kahneman and Tversky
(1973) presented one group of participants with the following scenario:
A panel of psychologists have interviewed and administered personality

tests to 30 engineers and 70 lawyers, all successful in their respective fields.
¹² Griffiths and colleagues (2015) suggest that the transition from the computational to the
algorithmic level can be facilitated by an intermediate level which adverts to resource con-
straints.
¹³ Note that we don’t need to qualify that a Marrian rational process is merely pro tanto
rational. This is because computational and algorithmic rationality are defined narrowly in
terms of the function of the process. For example, the algorithm for addition is defined in terms
of a restricted class of inputs and outputs and the dedicated transitions between them.
16  
On the basis of this information, thumbnail descriptions of the 30 engin-

eers and 70 lawyers have been written. You will find on your forms five
descriptions, chosen at random from the 100 available descriptions. For
each description, please indicate your probability that the person described
is an engineer, on a scale from 0 to 100. (241)
Another group of participants got the same scenario, but with the base rates
reversed—in this condition there were said to be 30 lawyers and 70 engin-
eers. Participants were then given the five descriptions (allegedly chosen at
random) mentioned in the instructions. One of the descriptions fits with a
stereotype of engineers:
Jack is a 45-year-old man. He is married and has four children. He is

generally conservative, careful and ambitious. He shows no interest in
political and social issues and spends most of his free time on his many
hobbies which include home carpentry, sailing, and mathematical puzzles.
(241)
With this description in hand, subjects are supposed to indicate how likely it
is (from 0 to 100) that Jack is an engineer. Kahneman and Tversky found
that participants in both conditions gave the same, high, probability esti-
mates that Jack is an engineer. The fact that there were 70 engineers in one
condition and 30 in the other had no discernible effect on subjects’ responses
(1973: 241). Another description was designed to be completely neutral
between the lawyer and engineer stereotype:
Dick is a 30-year-old man. He is married with no children. A man of high

ability and high motivation, he promises to be quite successful in his field.
He is well liked by his colleagues. (142)
When given the description that was neutral with respect to the stereotypes,
obviously participants should go with the base rates; instead, in both con-
ditions, they tended to say that there was a 50 percent chance that the person
was an engineer (1973: 242).
This study is representative of the broad pattern of results in the
Heuristics & Biases tradition, which has exposed numerous ways in which
people make mistakes in statistical reasoning. In the wake of this pessimistic
line of research, a new wave of cognitive psychology celebrates people’s basic
abilities in statistical inference. A wide range of cognitive phenomena have been
  :    17
promoted as instances of rational inference, including categorization (Kemp

et al. 2007; Smith et al. 2002), word learning (Xu & Tenenbaum 2007b), and
parsing (Gibson et al. 2013). In addition, since the late 2000s, work in devel-
opmental and cognitive psychology suggests that children actually have an
early facility with statistical reasoning. I’ll present two sets of findings from this
emerging research.
It is normatively appropriate to draw inferences from samples to popu-
lations when samples are randomly drawn from that population, but typic-
ally not otherwise. To see whether children appreciated this aspect of
statistical inference, Xu and Garcia (2008) had infants watch as an experi-
menter reach into an opaque box (without looking in the box) and pull out
four red ping-pong balls and one white one. In that case, it’s statistically
appropriate to infer that the box has mostly red balls. In keeping with this,
when infants were then shown the contents of the box, they looked longer
when the box contained mostly white balls than when the box contained
mostly red balls. Xu and Denison (2009) then investigated whether the
nature of the sampling made a difference. At the beginning of the task, the
experimenter showed a preference for selecting red balls over white ones.
Then, the experimenter selected balls from an opaque container as in the
experiment reported above. But in this study, for one condition the experi-
menter was blindfolded while in the other she had visual access to the
contents of the box. Xu and Denison found that babies were more inclined
to expect the population of balls to resemble the sample in the blindfolded
condition as compared to when the experimenter could see the balls she was
choosing. It seems that the babies were sensitive to whether or not the
sampling was random. Building on these findings, Kushnir and colleagues
found that children use sampling considerations to draw inferences about
preferences. When a puppet took five toy frogs from a population with few
frogs, the children tended to think the puppet preferred toy frogs; but
children tended not to make this inference when a puppet took five toy
frogs from a population that consisted entirely of frogs (Kushnir et al. 2010).
For a second example, consider another feature of good probabilistic
reasoning: If you have priors (e.g., you know the percentage of the popula-
tion afflicted with a certain disease) then you should use those priors (e.g., in
making inferences from a person’s symptoms to whether they have the
disease); furthermore, if you get new information, you should update the
priors. Girotto and Gonzalez (2008) explored such reasoning in children
using a task with chips of different shapes and colors. Figure 1.1 depicts a
18  
Figure 1.1 Stimuli for

probabilistic inference task
(adapted from Girotto &
Gonzalez 2008: 328)
stylized version of the materials—four black circles, 1 black square and 3

white squares.
In the “prior” task, kids were shown just the square chips and told “I’m
going to put all the chips in this bag . . . . I will shake the bag and I will take a
chip from it without looking.” Then they were told that if the chip was black
they would give a chocolate to Mr. Black (a puppet) and if the chip was white
it would go to Mr. White (another puppet). Then they were asked which
puppet they choose to be. Of the 4th grade school children, 91 percent
correctly chose the puppet most likely to win the chocolate (332). Girotto
and Gonzalez explored whether the child can also update in this kind of task.
The child was shown all eight chips and asked which is more likely to win
(with black advantage 5:3). Children tend to say correctly that black is more
likely to win. Then all eight chips are put in the bag and the experimenter
reaches in. He says, “Ah, listen. I’m touching the chip that I have drawn and
now I know something that might help you to win the game. I’m touching
the chip that I have in my hand and I feel that it is [a square]” (331). The kids
are allowed to revise their judgment, and they tend to answer correctly that
now it’s more likely that white will win (334). Note that children succeed in
these tasks with no training—they produce the correct response immedi-
ately. Subsequent work found that children and adults in two pre-literate
Mayan groups also succeed in these tasks (Fontanari et al. 2014).
These are just two examples of ways in which even young children seem
to make statistically appropriate inferences. We will see several further
examples in the course of the book. The lesson of this work is that despite
the foibles that have been revealed by the Heuristics and Biases tradition,
people, including very young children, possess a substantial competence at
probabilistic reasoning.
4.2 Statistical Learning and Rationality
The dominant paradigm for explaining these results is Bayesian learning

(see, e.g., Perfors et al. 2011). The advocates of this view stress the rational
  :    19
nature of Bayesian inference. For example, Amy Perfors and colleagues

write:
Bayesian probability theory is not simply a set of ad hoc rules useful for
manipulating and evaluating statistical information: it is also the set of
unique, consistent rules for conducting plausible inference . . . . Just as
formal logic describes a deductively correct way of thinking, Bayesian
probability theory describes an inductively correct way of thinking.
(Perfors et al. 2011: 313)
For many experiments, it’s not obvious that the data support the view that
children are engaged in a form of Bayesian updating (see, e.g., Nichols &
Samuels 2017). But there is little doubt that the inferences in the tasks
reviewed in Section 4.1 are plausible candidates for meeting familiar notions
of evidential rationality. Critically, in the above tasks, the child makes
inferences that are appropriate given the evidence to the agent. For instance,
in the ping-pong ball task, the infant is right to infer from a random sample
of mostly red balls that the population is mostly red. All of the evidence
she has supports this conclusion.¹⁴ In addition, these tasks take exactly
zero training. The normatively appropriate pattern appears on the first
(and only) trial. Much prominent work in Bayesian psychology claims
only to be giving an analysis of people’s judgments at the computational

level (e.g., Xu & Tenenbaum 2007b: 270). But at least in some cases, an
algorithmic analysis is also plausible, and one can get evidence for this
from people’s reports of their reasoning process (see, e.g., De Neys
& Glumicic 2008; Ericsson & Simon 1984). It’s likely that for tasks like
the chips task from Girotto and Gonzalez, adults would be quite capable of
articulating the reasoning process they actually go through, which might
well provide evidence that their reasoning process is algorithmically
rational.
¹⁴ One potential worry about the rationality of everyday inferences from samples is that the
samples might be unrepresentative. It is plausible that when people have evidence that the
sample is unrepresentative, if they ignore this in their statistical inferences, their inferences are
rationally compromised. However, when a person has no evidence that a sample is unrepre-
sentative, it seems uncharitable to declare their inferences from the sample to be rationally
corrupt. That is, when there is no evidence that a sample is unrepresentative, it’s reasonable for a
learner to make inferences as if it is representative. Indeed, if we were so cautious as to withhold
inferences on the bare possibility that a sample is unrepresentative, we would rarely make
inferences. To suggest such inferential caution borders on recommending skepticism.
20  
4.3 Statistical Learning and Empiricism
I’ll be promoting a kind of empiricist account of moral learning, in terms of

statistical inference. I’ll discuss this at some length in Chapter 7, but for now
I want to provide a bit of the intellectual background that informs the
discussion. In contemporary cognitive science, the debate between empiri-
cists and nativists is all about acquisition (see, e.g., Cowie 2008; Laurence &
Margolis 2001). Take some capacity like the knowledge of grammar. How is
that knowledge acquired? Empiricists about language acquisition typically
maintain that grammatical knowledge is acquired from general purpose
learning mechanisms (e.g., statistical learning) operating over the available
evidence.¹⁵ Nativists about language acquisition maintain instead that there
is some domain-specific mechanism (e.g., a specialized mechanism for
acquiring grammar) that plays an essential role in the acquisition of lan-
guage. In the case of grammatical knowledge, debate rages on (e.g., Perfors
et al. 2011; Yang et al. 2017). But it’s critical to appreciate that there is some
consensus that for certain capacities, an empiricist account is most plausible
while for other capacities, a nativist account is most plausible.
On the empiricist end, research shows that infants can use statistical
evidence to segment sequences of sounds into words. The speech stream is
largely continuous, as is apparent when you hear a foreign language as
spoken by native speakers. How can a continuous stream of sounds be

broken up into the relevant units? In theory, one way that this might be
done is by detecting “transitional probabilities”: how likely it is for one
sound (e.g., a syllable) to follow another. In general, the transitional prob-
abilities between words will be lower than the transitional probabilities
within words. Take a sequence like this:
happy robin
As an English speaker, you will have heard “PEE” following “HAP” more
frequently than ROB following PEE. This is because “HAPPEE” is a word in
English but “PEEROB” isn’t. This sort of frequency information is ubiqui-
tous in speech. And it could, in principle, be used to help segment a stream
into words. When the transitional probability between one sound and the
next is very low, this is evidence that there is a word boundary between the
¹⁵ Note that empiricists allow that these general-purpose learning mechanisms themselves
might be innate; after all, we are much better at learning than rocks, trees, and dust mites.
  :    21
sounds. In a groundbreaking study, Jenny Saffran and colleagues (1996)

used an artificial language experiment to see whether babies could use
transitional probabilities to segment a stream. They created four nonsense
“words”:
pabiku tibudo golatu daropi
These artificial words were strung together into a single sound stream,
varying the order between the words (the three orders are depicted on
separate lines below, but they are seamlessly strung together in the audio):
pabikutibudogolatudaropi
golatutibudodaropipabiku
daropigolatupabikutibudo
By varying the order of the words, the transitional probabilities are varied
too. Transitional probabilities between syllabus pairs within a word (e.g.,
bi-ku) were higher than between words (e.g., pi-go) (p = 1.0 vs. p = 0.33).
After hearing two continuous minutes of this sound stream, infants were
played either a word (e.g., pabiku) or part word (e.g., pigola). Infants listened
longer (i.e., showed more interest) when hearing the part word, which
indicates that they were tracking the transitional probabilities.¹⁶ This ability
to use statistical learning to segment sequences isn’t specific to the linguistic
domain. It extends to segmenting non-linguistic tones (Saffran et al. 1999)
and even to the visual domain (Kirkham et al. 2002). Perhaps humans have
additional ways to segment words, but at a minimum, there is a proven
empiricist account of one way that we can segment streams of continuous
information into parts using statistical learning.
Nativists can claim victories too, however. Birdsong provides a compel-
ling case. For many songbirds, like the song sparrow and the swamp
sparrow, the song they sing is species specific. It’s not that the bird is born
with the exact song it will produce as an adult, but birds are born with a
“template song” which has important elements of what will emerge as the
adult song. One line of evidence for this comes from studies in which birds
are reared in isolation from other birds. When the song sparrow is raised in
¹⁶ Listening time was measured by how long babies looked towards the source of the sound,
which was either on the left or the right side of the room.
22  
isolation, it will produce a song that shares elements of the normal adult
song sparrow song; the same is true for the swamp sparrow. And critically,
the song produced by the isolate song sparrow differs from the song pro-
duced by the isolate swamp sparrow. This provides a nice illustration of a
nativist capacity. It’s not that experience plays no role whatsoever—the
specific song that the bird produces does depend on the experience. But
there is also an innate contribution that is revealed by the song produced by
isolate birds. The template gives the bird a head start in acquiring the
appropriate song (see, e.g., Marler 2004).
The examples of birdsong and segmentation of acoustic strings show that
it’s misguided to think that there is a general answer to the nativist/empiri-
cist debate. There are numerous nativist/empiricist debates, and it’s import-
ant to evaluate the disputes on a case by case basis. The cases I offer in Part II
of this book are all empiricist learning stories, based on principles of
statistical inference. Importantly, however, the arguments in Part II make
no claim to a thoroughgoing empiricism. The work starts with learners who
already have facility with concepts like agent, intention, and cause. It also
starts with the presumption that learners have the capacity for acquiring
rules. I argue that, given those resources and the evidence available to
children, their inferences are rational. This is all consistent with the nativist
claim that the acquisition of the concept of agent (for example) depends on
innate domain-specific contributions and that there is an innate capacity to

learn rules (Nichols 2006: 355–8).
4.4 A Schema for Statistical Learning Accounts
Let’s now turn to the characteristics of such learning accounts. Suppose we

want to argue that some concept (for example) was acquired via statistical
inference over the available evidence. Several things are needed.
(1) One needs a description of the concept (belief, distinction, etc.) the
acquisition of which is to be explained. We can call the target concept
the acquirendum (Pullum & Scholz 2002). Part of the work here is to
argue that we do in fact have the concept or distinction or belief that
is proposed as the acquirendum (A).
(2) Insofar as statistical learning is a form of hypothesis selection, one
needs to specify the set of hypotheses (S) that the learner considers in
  :    23
acquiring A. This set of hypotheses will presumably include A as well

as competing hypotheses.
(3) One will also need an empirical assessment of the evidence (E) that is
available to the learner. For this, one might consult, inter alia, corpus
evidence on child-directed speech.
(4) The relevant statistical principle(s) (P) need to be articulated. These
principles should make it appropriate for a learner with the evidence
E and the set of hypotheses S to infer A.
(5) Finally, a complete theory of acquisition would tie this all together by
showing that the learner in fact does use the postulated statistical
principle P and the evidence E to select A among hypotheses S.
Few empiricist theories of acquisition manage to provide convincing evi-

dence for all of these components. The last item in particular is well beyond
what most learning theorists hope to achieve. For instance, (5) would
require extremely fine-grained longitudinal analyses of the evidence avail-
able to individual children and their use of that evidence in learning. In place
of this daunting demand, a learning theorist might aim for a weaker goal—to
show that learners are capable of using the relevant kind of evidence to make
the inferences that would be appropriate given the postulated statistical
principles. That is, instead of trying to capture the learner’s actual acquisi-
tion of the concept, one might settle for something a good deal weaker:
(5*) Show that when the learner is given evidence like E, she makes
inferences that would be appropriate if she were deploying the postulated
statistical principles P.
This requirement is of course weaker than (5) in that it doesn’t try to show
the actual transition that occurs when a child acquires the concept. Rather,
the goal is to show that learners are appropriately sensitive to the evidence.¹⁷
In addition, (5*) is intended to make a weak claim about the precision and
accuracy of the probabilistic representations. I don’t argue (I don’t even
believe) that people make precisely accurate probabilistic inferences from
the evidence. Rather, the goal of this book is to argue that, for a range of
important elements in our moral psychology, when people learn those
¹⁷ The term “sensitive” has a technical meaning in analytic epistemology (e.g., Nozick 1981),
but I intend the ordinary notion on which being sensitive roughly means responding appro-
priately under different conditions.
24  
elements, they make inferences that are roughly the kinds of inferences that
they should make, given the evidence.¹⁸
Our next task is to determine the kinds of representations that are
implicated in moral judgment. That will be crucial to characterizing the
acquirenda, and it will be the focus of the next chapter.
¹⁸ This modest defense of human rationality is reflected in some of Tania Lombrozo’s work.
For instance, in her lovely work on simplicity, she shows that people are responsive to evidence
in a Bayesian fashion but they overweight the importance of simplicity (2007).
2
The Wrong and the Bad
On the Nature of Moral Representations
1. Introduction
It is bad when a puppy falls off of a cliff. It is wrong when a person throws a
puppy off a cliff. There are some obvious differences between these unfor-
tunate events. For instance, the latter involves a person and an action. Any
account of the mental representation of these events would have to register
these differences. But there might also be a more subtle difference in the
corresponding moral representations. The representation of the badness of
the puppy falling is naturally taken to be a representation of the value of the
event. It’s a bad value. The representation in the second case will also involve
a value representation since the puppy is injured in that scenario too. But it’s
possible that the characteristic representation for the second scenario
involves something more than registering a bad value. It might involve a

structured representation of a rule against injuring innocents, composed of
abstract concepts like impermissible, harm, and knowledge.
The idea that moral judgments of wrongness implicate structured rules is
hardly new.¹ But many theories of moral judgment try to make do with a
much more austere set of resources. It is a familiar pattern in cognitive
science to seek low-level explanations for apparently high-level cognitive
phenomena—witness classic disputes about symbolic processing. Some
influential connectionist approaches attempt to explain cognition with no
recourse to symbols (McClelland et al. 1986). We find a related trend in
accounts of moral judgment that exclude rules in favor of lower level factors.
In low-level accounts of moral judgment, the primitive ingredient is typic-
ally some kind of simple value representation. Blair’s account of the moral/
¹ In philosophy, particularism is set in opposition to rule-based approaches to morality (e.g.,

Dancy 2009). It’s often unclear the extent to which particularism is supposed to make descrip-
tive claims about human psychology. That is, it’s unclear whether particularists mean to deny
that rules play a causal role in ordinary moral judgment. But if particularists do mean to deny
this, then their view runs into significant difficulties (see Zamzow 2015).
26  
conventional distinction is based on the distress associated with seeing

others in distress (Blair 1995). Greene’s account of responses to dilemma
cases is based on alarm-like reactions of ancient emotions systems (Greene
2008). Cushman (2013) and Crockett (2013) seek to explain judgments in
dilemmas by appeal to the kinds of value representations implicated in habits.
Railton (2014) draws on more sophisticated value representations, but still
stops well short of anything like rules framed over abstract categories.
Low-level accounts are often attractive because they build on processes
that are uncontroversially present in the organism. In the case of moral
judgment, few dispute that humans find it aversive to witness suffering;
similarly, it’s widely acknowledged that humans come to find certain kinds
of actions aversive through reinforcement learning. Thus, if we can explain
moral judgment in terms of some such widely accepted low-level processes, then
we have no need to appeal to such cognitive extravagances as richly structured
rules constructed from abstract concepts like impermissible and harm.
Despite its tough-minded appeal, the race to lower levels can neglect the
very phenomena we want to understand. Trying to explain human cognition
without adverting to symbolic processing makes it difficult to capture core
phenomena like the systematicity and inferential potential of thought
(Fodor & Pylyshyn 1988). Similarly, I argue, it is difficult to capture the
distinctive nature and specificity of wrongness judgments without adverting
to structured rules. Before we get to the arguments, I want to explicate the

operative notions of value representation and rule representation.
2. Value Representations
One ubiquitous feature of navigating the world involves representing certain

actions as good and others as bad. A hungry rat will regard an outcome of
getting a cracker as having a certain amount of positive value, and it will
regard an outcome of getting a grape as having even greater positive value.²
By contrast, rats regard an outcome of being immersed in a pool as having
a certain amount of negative value. Generally speaking, if an organism
regards a certain outcome as having a positive value, then it will have some
motivation to bring that outcome about, and if it regards a certain outcome
² The value associated with the particular outcome (e.g., of getting a grape) is sometimes
called the “instantaneous utility function” in expected utility theory and the “reward” in
reinforcement learning.
      27
as having a negative value, then it will have some motivation to avoid

that outcome.
Value representations fit neatly into venerable work on reinforcement
learning. Early work on maze learning revealed that rats have an impressive
ability to build models. For instance, in one task, satiated rats were put in a
Y maze that had food at the end of the left branch and water at the end of the
right branch. Although the rats weren’t hungry or thirsty, they were led to
run down each branch twice a day for seven days. After this, the rats were
made either hungry or thirsty and then put into the maze. The hungry rats
tended to go to the left branch (where the food had been on the previous
days) and the thirsty rats tended to go to the right branch (where the water
had been on the previous days). This shows that simply by running the
branches the rats built a model of the maze, including what was in the maze
(food and water) and where it was (left/right). This is just one simple
example, but it’s representative of the rat’s talent at learning models of its
environment (see, e.g., Tolman 1948).
Value representations also fit neatly into an expected utility framework.
On that familiar framework, an agent considers both the value attached to a
particular outcome and the probability that a certain action will lead to that
outcome. In the simple case, the system just multiplies the value associated
with an outcome by the probability that the action will lead to that outcome.
The system computes this expected utility for each action under consider-
ation, and then selects the action that has the highest expected utility. For
instance, a rat might come to expect that pressing a lever will lead to an
outcome in which the rat gets a grape, and this will motivate the rat to press
the lever. A rat might also take into account the relative probabilities of
certain outcomes given certain actions. Let’s say the rat values the outcome
of getting a grape at 0.7, and values the outcome of getting a cracker at 0.5.
The rat’s past familiarity with a maze has led the rat to have a model of
the maze specifying that if it goes left there is a 0.1 chance of a grape and a
0.9 chance of nothing; if it goes right there is a 0.6 chance of a cracker and
a 0.4 chance of nothing. In that case, even though the rat prefers the grape
to the cracker, it assigns a higher expected value to the action of going right.
In all of these cases, we think of the rat as being guided by a value assigned to
an outcome. Thus, these are dubbed outcome-based value representations
(e.g., Cushman 2013).
Outcome-based value representations are the kinds of value representa-
tions that underlie goal-directed behavior. However, sometimes organisms
represent an action itself as having a positive (or negative) value, without
28  
representing some goal or good consequence that will result from the action.
The basic idea is familiar from habitual actions, like absentmindedly putting
your pen to your mouth. When you do this, you usually don’t have a goal in
mind that is served by putting your pen to your mouth. Rather, the behavior
is produced by a system that places values on actions themselves. Such
representations are action-based value representations.
While outcome-based value representations draw on the organism’s
knowledge of models, action-based value representations are not integrated
with models. They are simply values assigned to particular kinds of actions.
This means that the kinds of procedures by which action-based value
representations are acquired don’t depend on the agent learning a model
of its environment. Accordingly, this kind of learning is sometimes called
“model-free” reinforcement learning. Instead of constructing a model, the
agent simply develops a value for particular actions given a situation. For
instance, after getting food from pushing a lever several times, a rat might
come to assign a positive value to lever-pushing itself. Such action-based
value representations drive habitual behavior, and this behavior can persist
even when the original goal of the behavior is undermined.
For a simple action like pushing a lever to get food, it can be hard to
determine whether the act is driven by an outcome-based value representa-
tion (obtain food) or an action-based value representation (hit the lever).
We can distinguish these explanations when the goal is “devalued.” In a

characteristic devaluation experiment, a rat first learns that pushing the lever
is the way to get food. The rat is then removed from the cage, fed until it is
completely satiated, and put back into the cage. In some conditions, rats will
immediately start pushing the lever even though they don’t eat the food that
tumbles out. This habit-like behavior can also be observed if a hungry rat is
led to the clear knowledge that there is no food available, such that pressing
the lever will not lead to food. Nonetheless, under certain conditions, the rat
will still press the bar, out of habit.
As we will see, several theorists have tried to explain moral judgment in
terms of simple value representations and reinforcement learning (Sections
4 and 5 below). These accounts are intended to explain moral judgment with
minimal resources. In keeping with the examples in this section, the kinds of
processes and representations invoked in these theories are primarily rooted
in research on rat learning. Rats are great and all, but I will argue that to
explain moral judgment, we need to go beyond rodent psychology. Moral
judgment involves more than just outcome-based and action-based value
representations; moral judgment also involves complex representations of
      29
rules. Of course we can use the label “value representation” in a way that
would allow us to say that complex rules are associated with values. An
expected utility model can assign a subjective cost for breaking a prohibition
rule. In such a case, we might say that there is a representation that assigns
negative value to actions that violate the rule, and we can call this a “value
representation.” But the interest of the recent work on value representations
and moral judgment has been the attempt to explain moral judgment with
more austere resources that do not incorporate complex rule representa-
tions, and I will accordingly use “value representation” to pick out repre-
sentations of actions or outcomes that are not specified in terms of complex
rule representations. Now I will turn to the task of saying more fully what a
rule representation is.
3. Rule Representations
Which kinds of representations are implicated in moral judgment? This

question is naturally pitched at Marr’s level 2, where the goal is to charac-
terize the representations and algorithms that an organism actually uses to
solve problems. Of course, these representations must be neurally imple-
mented in some way, but that is not a question that will occupy us here. I will
argue that in order to explain moral judgment, we need to appeal to rule

representations as characterized by the following three claims:
(1) Rule representations are partly composed of concepts, including

abstract concepts like , , and .
(2) Rule representations are structure dependent. That is, the same set of
concepts will constitute different rules depending on how the con-
cepts are structured in the rule.
(3) Rule representations are (at least typically) not hypothetical.
The structure dependence of rules means that if Xs are different from Ys,
then the rule that it’s impermissible to put Xs on Ys is different from the rule
that it’s impermissible to put Ys on Xs. The claim that the rules are not
hypothetical draws on Kant’s distinction between hypothetical and non-
hypothetical imperatives. Hypothetical imperatives are rules that serve one’s
interests like “Put oil in your car.” This imperative applies to us because we
desire to prevent our engine from seizing up. If for some reason we want our
engine to seize up (say, because we’re conducting an engine test) then the
Another random document with
no related content on Scribd:
Gering, rue de Sorbonne, avec son enseigne: au Soleil d’Or; mais, lorsqu’il
se fut associé avec son compatriote Philippe Cruzenach, il porta son
imprimerie, rue Saint-Jacques, à l’enseigne de Sainte Barbe. Jean Lambert,
qui était libraire rue Saint-Séverin, en 1492, ne conserva pas son enseigne de
la Corne de cerf, quand il vint s’établir, comme imprimeur, devant le collège
Coqueret, où il jugea convenable de pendre l’Image de Saint Claude au-
dessus de sa boutique. Jean Trepperel, imprimeur-libraire, qui imprima et
publia un grand nombre de romans de chevalerie, avait d’abord pour
enseigne l’image de Saint Laurent, quand il demeurait sur le pont Notre-
Dame, en 1491; il changea de saint et d’enseigne, en allant demeurer rue
Saint-Jacques, en 1498, à l’Image Saint Yves; il n’y resta pas longtemps et
passa dans la rue de la Tannerie, à l’enseigne du Cheval noir; en 1502, sa
veuve tenait boutique, rue Neuve-Notre-Dame, à l’Écu de France. On
s’explique ces changements d’enseigne par ces changements de domicile.
C’était, d’ailleurs, une habitude presque générale chez les imprimeurs et
les libraires, du moins à la fin du XVᵉ siècle et au commencement du XVIᵉ.
Jean Philippi, imprimeur allemand, établit ses presses, en 1495, rue Saint-
Jacques, à l’Image de Sainte Barbe; puis, il les transporta bientôt, rue Saint-
Marcel, à l’enseigne de la Sainte-Trinité. Un autre imprimeur allemand, Jean
Higman, imprimait d’abord, en 1489, dans la maison de Sorbonne, cet asile
tutélaire de l’imprimerie; il alla demeurer, au Clos-Bruneau, près les Écoles
de Décret, à l’enseigne des Lions; ensuite, il se fixa rue Saint-Jacques, à
l’Image Saint Georges. La rue Saint-Jacques sera désormais, comme au
moyen âge, le grand marché de la librairie. Jean Driart s’y installe, à
l’enseigne des Trois Pucelles, en 1498, pour y imprimer la plus belle édition
du Mystère de la Destruction de Troie la grant. Antoine Nidel, qui était
maître ès arts avant de se faire imprimeur, ne s’éloigne pas du collège
Coqueret et du collège de Montaigu en créant une imprimerie, à l’enseigne
de la Cathédrale, dans le quartier des relieurs, qu’on appelait le Mont Saint-
Hilaire, entre la rue Saint-Hilaire et la rue des Sept-Voies. Deux imprimeurs
s’établissent, la même année, Félix Baligant, sur la montagne Sainte-
Geneviève, près le collège de Reims, à l’Image Saint Étienne, et Geoffroy de
Marnef, rue Saint-Jacques, près l’église Saint-Yves, à l’enseigne du Pélican.
Ce dernier s’était associé à ses deux frères Jean et Enguilbert, en réunissant
leurs trois enseignes dans une marque collective, qui représentait leur
association «par trois symboles, dit La Caille, des grues qui font leur nid en
volant, un perroquet qui parle, un pélican qui donne la vie à ses petits, avec
trois bâtons» où sont les initiales de leur trois prénoms.
Il faut maintenant citer les imprimeurs et les libraires qui ouvrent leurs
ateliers et leurs boutiques, avec enseignes, sous le règne de Louis XII. C’est,
en 1498, le libraire Jean Petit, rue Saint-Jacques, à l’enseigne du Lion
d’Argent, et dans la même rue, près des Mathurins, à la Fleur de lys d’Or. Ce
libraire est un de ceux qui, de son temps, ont fait imprimer le plus de livres,
puisqu’il entretenait, à lui seul, les presses de quinze imprimeurs. Il avait
réuni, dans sa marque de libraire, le lion et la fleur de lis, qui figurèrent l’un
après l’autre dans les deux enseignes de ses boutiques. Thomas du Guernier,
qui exerça pendant plus de vingt-six ans comme imprimeur et comme
libraire, et qui imprima un grand nombre de romans de chevalerie, demeurait
d’abord, rue de la Harpe, à l’Image Saint Yves; il resta dans la même rue, en
transportant son établissement près le Pilier vert, à l’enseigne du Petit
Cheval blanc. Denis Roce, qui ne fut que libraire, demeurait rue Saint-
Jacques, près l’église Saint-Benoît, où la plupart des libraires avaient leur
sépulture; il prit pour enseigne l’Image Saint Martin, et pour marque, par
allusion à son nom de Roce, un rosier portant des roses, avec son écusson
armorié, soutenu par deux griffons.
Les imprimeurs allemands et belges qui venaient s’établir à Paris y
faisaient encore fortune. Jodocus Badius Ascensius, après avoir été
professeur au collège de Lyon, se fit imprimeur à Paris, vers 1502, rue Saint-
Jacques, «au-dessus de l’église Saint-Benoît, près du Gril», à l’enseigne des
Trois Luxes (brochets); il devint libraire, pour vendre les excellentes éditions
de classiques latins, qu’il publiait lui-même. Nicolas Wolff, de Bade, qui
mettait sur ses éditions: Impressi arte et industria ingeniosissimi viri N. W.
Allemani, imprima d’abord dans le cloître Saint-Benoît, aux Trois Tranchoirs
d’Argent; ensuite rue de la Harpe, à l’enseigne des Rats. Thomas Kées, de
Wesel, imprimait à l’enseigne du Miroir, près le collège des Lombards.
Alexandre Aliate, Belge, n’exerça que quelques mois, en 1500, rue Saint-
Jacques, devant le Collège de France, à l’Image Sainte Barbe. Thielman
Kerver, arrivé d’Allemagne en 1500, fonda une des plus importantes
maisons d’imprimerie et de librairie, qui devait durer plus d’un siècle; il
demeurait d’abord sur le pont Saint-Michel, à l’enseigne de la Licorne;
ensuite, rue des Mathurins et rue Saint-Jacques, toujours avec la même
enseigne, qu’il ne changea que vers la fin de sa vie, pour prendre celle du
Gril; mais ses successeurs reprirent celle de la Licorne, qu’ils faisaient
figurer dans leur marque typographique. Berchtold Rembolt, de Strasbourg,
avait adopté l’enseigne d’Ulric Gering, le Soleil d’Or, en créant son
imprimerie rue Saint-Jacques; mais, en allant s’établir près du Collège de
France, il prit pour enseigne l’Image Saint Christophe. Sa veuve, qui épousa
en 1513 Claude Chevallon, rentra, avec le Soleil d’Or, dans la maison que
son premier mari avait habitée rue Saint-Jacques. Claude Chevallon était si
fier de son enseigne, qu’il la mit dans sa marque, au-dessus de son écusson,
soutenu par deux chevaux dressés debout, pour faire allusion par un rébus à
son nom de Chevallon (cheval long).
Les autres imprimeurs et libraires qui parurent du temps de Louis XII
étaient tous Français, mais ils n’exercèrent pas longtemps, à l’exception de
quatre ou cinq. Gaspard Philippe imprimait, en 1502, rue de la Harpe, aux
Trois Pigeons; Nicolas de la Barre, en 1509, rue de la Harpe, devant l’Écu de
France, aux Trois Saumons; puis, rue des Carmes, devant le collège des
Lombards, à l’Image de Saint Jean l’Évangéliste; Antoine Bonnemère, en
1508, rue Saint-Jean-de-Beauvais, devant les grandes Écoles de Décret, à
l’Image Saint Martin; Guillaume le Rouge, en 1512, rue Saint-Jean-de-
Latran, à la Corne de Daim; Nicolas des Prez, en 1508, rue Saint-Étienne-
du-Mont, au Miroir. Il y eut plusieurs imprimeurs célèbres dont les débuts
datent de ce règne-là, mais qui n’acquirent toute leur réputation que sous le
règne suivant. François Grandjon, très habile graveur et fondeur de
caractères, de même que ses deux frères Jean et Robert, demeurait rue Saint-
Jacques, près l’église Saint-Yves, à l’Image Saint Claude, et devant le
couvent des Mathurins, à l’Éléphant; Gilles de Gourmont, le premier qui
imprima du grec à Paris, et qui devint imprimeur du roi, habitait rue Saint-
Jacques, aux Trois Couronnes, mais il remplaça cette enseigne par celle de
l’Écu de Cologne, sans changer de domicile. Ses deux frères, Jean et Robert,
imprimeurs comme lui, demeuraient au Clos-Bruneau, près le collège
Coqueret, à l’enseigne des Deux Boules. Guillaume Anabat, demeurant rue
Saint-Jean-de-Beauvais, depuis 1500 jusqu’en 1505, à l’enseigne des
Connils (lapins), imprimait de superbes livres d’heures, remplis d’images
gravées sur bois, pour Gilles Hardouyn, demeurant au bout du Pont-au-
Change, à l’enseigne de la Rose, et pour Germain Hardouyn, qui avait sa
boutique de libraire devant le Palais, à l’Image Sainte Marguerite. Enfin,
Guillaume Nyverd, successeur de Pierre Le Caron, en 1516, conserva cette
fameuse imprimerie, dans le même local, rue de la Juiverie, à l’enseigne de
la Rose, qui céda la place, on ne sait pourquoi, à l’Image Saint Pierre.
Le nombre des libraires augmentait à mesure que diminuait celui des
imprimeurs. Il y eut un libraire du roi, Guillaume Eustace, qui demeurait, en
1508, rue de la Juiverie, à l’enseigne des Deux Sagittaires, qui furent
reproduits dans toutes les marques de ce libraire, jusqu’à ce que sa boutique
fût transférée, dans la rue Notre-Dame, à l’enseigne de l’Agnus Dei. Il faut
citer, parmi les autres libraires de ces temps-là, Poncet Lepreux, qui fit
imprimer à ses frais une immense quantité de livres en tous genres, depuis
1498 jusqu’en 1552; pendant 55 ans, il n’eut que deux domiciles, rue Saint-
Jacques, devant les Mathurins, à l’enseigne du Loup, et même rue, au
Croissant. Citons encore deux autres libraires de la même époque, Jean
Lefèvre, rue Saint-Jacques, à l’enseigne du Croissant; Jacques Ferraboue, en
1514, sur le Petit-Pont, devant l’Hôtel-Dieu, à l’enseigne du Croissant.
Galiot du Pré, chef d’une nombreuse famille de libraires qui se succédèrent
jusqu’en 1570, ouvrit boutique, sur le nouveau pont Notre-Dame, en 1512, à
l’enseigne de la Galée, par allusion à son prénom de Galiot, qui signifiait
aussi un navire à voiles et à rames. Il avait donc fait poser son enseigne dans
sa marque, avec cette devise de bon augure: Vogue la Galée.
Sous François Iᵉʳ, où le nombre des livres imprimés se multiplie à l’infini,
la place est prise et disputée par les fils et les descendants des anciens
imprimeurs et libraires; nous n’avons donc à mentionner que les nouveaux
venus dans l’imprimerie et la librairie, avec de nouvelles enseignes.
Commençons par les imprimeurs, qui la plupart travaillaient pour les
libraires et ne publiaient plus de livres. On ne peut imaginer combien de
poésies et de vieux romans de chevalerie, d’ouvrages de morale et de
philosophie, furent mis au jour sous le règne de François Iᵉʳ. Guillaume
Lebret imprimait, avec Jean Réal, en 1538, dans le Clos-Bruneau, à la Corne
de Cerf, puis à la Rose rouge. Jean Réal se sépara de son associé en 1548, et
imprima seul, rue du Mûrier, à l’Image Sainte Geneviève; puis, rue
Traversine, au Cheval blanc. Pierre Gromors imprimait, à l’enseigne de la
Cuillère, près l’église Saint-Hilaire, en 1521; au Phénix, près le collège de
Reims, en 1538. Simon de Colines, qui était à la fois imprimeur et libraire,
de 1510 à 1550, changea souvent de demeure et d’enseigne: en 1526, il avait
sa boutique, près le collège de Beauvais, au Soleil d’Or; en 1527, sa marque
de libraire nous fait supposer qu’il avait pris pour enseigne: aux Connils
(lapins); en 1531, son enseigne représentait le Temps, barbu et poilu,
maniant sa faux, avec cette devise: Hanc aciem sola retundit virtus. Jean
Messier, associé d’abord avec Jean du Pré, imprimait en 1517, rue des
Poirées, près le collège de Cluny, à l’Image Saint Sébastien; Nicolas
Savetier, rue des Carmes, à l’Homme sauvage; Guillaume de Bozzozel, rue
Saint-Jacques, près des Mathurins, au Chapeau rouge; Nicolas Prévost, rue
Saint-Jacques,
à l’Image Saint Georges; Pierre Leber, rue des Amandiers, à l’enseigne de la
Vérité; Gilles Couteau, rue Grenier-Saint-Ladre, à l’enseigne du Grand
Couteau; Pierre Regnault, rue Saint-Jacques, aux Trois Couronnes de
Cologne; Pierre Sergent, rue Notre-Dame, à l’Image Saint Nicolas; Denis
Janot, qui fut nommé imprimeur du roi, en 1543, rue Notre-Dame, contre
l’église Sainte-Geneviève-des-Ardents, à l’Image Saint Jean-Baptiste. Enfin,
Michel Vascosan, qui fut libraire et imprimeur, de 1530 à 1576, et qui peut
être considéré comme le plus excellent imprimeur du XVIᵉ siècle, demeurait
rue Saint-Jacques, à l’enseigne de la Fontaine, là où Jodocus Badius avait
établi sa presse, à son arrivée à Paris, en 1502; aussi, Vascosan avait-il
représenté dans sa marque d’imprimeur cette presse avec l’inscription:
Pressum ascensianum.
Voici maintenant les enseignes de quelques-uns des meilleurs libraires de
cette époque: Alain Lotrian, de 1518 à 1539, rue Notre-Dame, à l’Écu de
France; Regnault Chaudière, 1516-1551, rue Saint-Jacques, à l’Homme
sauvage; Pierre Bridoux, rue de la Vieille-Pelleterie, à l’enseigne du
Croissant; Jean Hérouf, rue Notre-Dame, à l’Image de Saint Nicolas;
Nicolas Chrestien, près le collège de Coqueret, à l’Image Saint Sébastien;
Jean Roigny, rue Saint-Jacques, au Basilic, et ensuite aux Quatre Éléments;
Ambroise Girauld, 1526-1546, rue Saint-Jacques, au Lion d’Argent; même
rue, au Roi David; même rue, au Pélican; Jean Macé, 1531-1582, au Mont
Saint-Hilaire, à l’Écu de Bretagne; puis au Clos-Bruneau, à l’Écu de
Guyenne; Vincent Sertenas, 1538-1554, rue Notre-Dame, à l’Image Saint
Jean l’Évangéliste; puis, même rue, à la Corne de Cerf; Étienne Robert, dit
le Faucheur, libraire et relieur du roi, sur le pont Saint-Michel, à la Rose
blanche. Geoffroy Tory, peintre, graveur et libraire, 1512-1550, le premier
qui obtint un privilège du roi pour l’impression d’un nouveau livre d’heures,
demeura successivement sur le Petit-Pont, joignant l’Hôtel-Dieu, rue Saint-
Jacques, devant l’Écu de Bâle et, même rue, devant l’église de la Madeleine,
et conserva toujours sa fameuse enseigne du Pot cassé, qu’il avait dessinée
lui-même sur sa marque de libraire, avec cette touchante devise: Non plus,
qui rappelait sa douleur à la mort de sa fille, etc.
Je m’arrête ici, en me reprochant de m’être laissé entraîner trop loin dans
cette aride et monotone énumération
d’imprimeurs et de libraires, sans autres détails que leurs adresses et leurs
enseignes. J’ai tenu à montrer ainsi que ces enseignes et ces adresses
changeaient souvent, dans l’exercice de leur profession; que leur industrie et
leur commerce étaient concentrés dans un petit nombre de rues du quartier
de l’Université, et que leurs enseignes différaient souvent de leurs marques
typographiques. On nous permettra[137] d’ajouter à ces renseignements un
peu arides, malgré leur intérêt, une nomenclature sommaire des marques ou
enseignes que certains imprimeurs et libraires de Paris ont mises sur des
livres sortis de leurs presses ou de leurs boutiques, quoique ces livres ne
portent pas leurs noms, ni l’indication du lieu d’impression. Ce n’étaient pas
là des éditions clandestines, mais c’étaient bien des éditions anonymes, et
nous avouons ne pas comprendre le but et l’intention des éditeurs, qui se
cachaient ainsi sous leurs enseignes.
L’Abel, de l’Angelier.—L’Abraham, de Pacard.—L’Amitié, de Guillaume
Julien.—Le Basilic et les Quatre Éléments, de Roigny.—Le Bellérophon, de
Perier.—La Bonne Foi, de Billaine.—Le Caducée, de Wechel.—Le
Cavalier, de Pierre Cavalier.—Le Cordon du Soleil, de Drouart.—Le Chêne
vert, de Nicolas Chesneau.—Les Cigognes, de Pâris.—Le Saint-Claude,
d’Ambroise de la Porte.—Le Cœur, de Huré.—Le Compas, d’Adrien Perier.
—La Couronne d’Or, de Mathurin du Puis.—L’Éléphant, de François
Regnault.—Les Épis mûrs, de Du Bray.—L’Espérance, de Gorbin.—
L’Étoile d’Or, de Benoît Prevost.—La Fontaine, de Vascosan.—Autre
Fontaine, des Morel.—La Galère, de Galiot du Pré.—L’Hercule, de Vitré.—
La Licorne, de Chappelet.—Autre Licorne, de Kerver.—Le Loup, de Poncet
le Preux.—Le Lys blanc, de Gilles Beys.—Le Lys d’Or, d’Ouen Petit.—Le
Mercure arrêté, de David Douceur.—Le Mûrier, de Morel.—Le Grand
Navire, de la Société des libraires de Paris, pour les impressions des Pères de
l’Église.—L’Occasion, de Fouet.—L’Olivier, des Estienne.—Autre Olivier,
de Chappelet.—Autre Olivier, de Patisson.—Autre Olivier, de Pierre
l’Huillier.—La Paix, de Jean Heuqueville.—La Palme, de Courbé.—Le
Parnasse, des Ballard.—Le Pégase, de Wechel.—Autre Pégase, de Denis du
Val.—Le Pélican, de Girault.—Le Phénix, de Michel Joly.—La Pique
entortillée d’une branche et d’un serpent, de Frédéric Morel.—Autre, de
Jean Bien-né.—Le Pot cassé, de Geoffroy Tory.—La Presse ou
l’Imprimerie, de Badius Ascencius.—La Rose dans un cœur, de Corrozet.—
La Ruche, de Robert Fouet.—La Salamandre, de Denis Moreau.—La
Samaritaine, de Jacques du Puis.—Le Saturne, ou le Temps, de Colines.—Le
Sauvage, de Buon.—Le Serpent mosaïque, de Martin le Jeune.—Le Soleil,
de Guillard.—La Toison d’Or, de Camusat.—La Trinité, de Meturas.—La
Vérité, de David.—La Vertu, de Laurent Durand.—Les Vertus théologales,
de Savreux.—La Vipère de saint Paul, de Michel Sonnius.
Cette liste aurait pu être aisément doublée, mais, telle qu’elle est, elle doit
suffire pour indiquer des enseignes qui étaient alors tellement bien connues,
qu’elles suppléaient en quelque sorte aux noms des imprimeurs et des
libraires qui les avaient adoptées et mises en honneur.
XII
ENSEIGNES DES ACADÉMIES, DES THÉÂTRES, DES

LIEUX PUBLICS, DES TRIPOTS ET DES MAUVAIS
LIEUX
E N rassemblant les notes qui m’ont servi à préparer mon édition du Livre
commode des Adresses de Paris, publié en 1691 et 1692 par le sieur de
Blegny, sous le pseudonyme d’Abraham du Pradel, je m’étais beaucoup
préoccupé de rechercher quelles étaient les enseignes des lieux publics de
Paris, tels que les étuves, les théâtres, les jeux de paume, car ces enseignes
devaient exister encore, du moins la plupart, à la fin du XVIIᵉ siècle; et il est
certain que, dans le siècle précédent, il n’y avait pas à Paris une seule maison
qui n’eût son enseigne, soit nominale, soit figurée. Ainsi les académies (nous
ne parlons pas des Académies royale des sciences, des inscriptions et belles-
lettres, des beaux-arts ou de sculpture et de peinture, encore moins de
l’Académie française), si nombreuses depuis le règne de Louis XIII, eurent
incontestablement des enseignes, et pourtant nous n’avons pas réussi à les
découvrir, soit pour les académies des jeux ou des brelans publics, soit pour
les académies proprement dites, instituées par des particuliers pour
l’éducation de la noblesse. Ces académies n’étaient que des manèges
d’équitation, des salles d’escrime, des écoles de musique, de danse, etc., et
par conséquent elles s’annonçaient aux passants par des enseignes
permanentes et par des écriteaux indicatifs. Notre édition du Livre
commode[138], où les enseignes des marchands ne sont signalées qu’en très
petit nombre, ne donnera donc pas les enseignes des académies, ni celles des
autres lieux publics, enseignes qu’une enquête plus heureuse que la nôtre
parviendra peut-être à mettre au jour.
Il en est une, cependant, que je viens de découvrir en feuilletant la
Bibliothèque de l’École des Chartes et qui a été commune à toutes les
académies de danse au XVIIᵉ siècle. Quand Louis XIV, qui aimait la danse au
point de la pratiquer en maître, eut créé l’Académie royale de danse, en
1662, le roi des ménétriers, Guillaume du Manoir, joueur de violon du
cabinet du roi et l’un des vingt-cinq de la grande bande, intenta un procès
aux maîtres à danser, et ce, au nom de la communauté et de la confrérie de
Saint-Julien des ménétriers. Ce procès en Parlement ne dura pas moins de
trente ans et fut terminé par une déclaration du roi, du 2 novembre 1692, en
vertu de laquelle la communauté de Saint-Julien fut maintenue dans la
jouissance de son privilège de donner, concurremment avec l’Académie
royale de danse, soit des lettres de maîtrise de joueur d’instruments, soit des
leçons de danse. Voici un curieux extrait du factum pour Guillaume du
Manoir: «De tous temps, la plupart des maîtres à danser ont eu et ont encor
un violon pour enseigne, non pas pour désigner qu’ils montrent à jouer de
cet instrument, mais pour marquer qu’ils montrent la danse et la liaison qu’il
y a entre la danse et le violon; et, de fait, au-dessous du violon qui leur sert
d’enseigne on écrit toujours ces mots: Céans on montre à danser, et, de plus
encor, lorsqu’on veut exprimer qu’un écolier va apprendre cet exercice ou
celui de l’épée, on dit vulgairement qu’il va à la salle[139].»
Cette citation nous permettra de supposer l’existence d’une autre
enseigne pour les académies où l’on apprenait l’escrime. Leur enseigne avait
été une épée tenue par une main gantée, et cette enseigne était placée sans
doute au-dessus de la porte de toutes les maisons où il y avait une salle
d’armes pour l’exercice de l’épée. Nous avions remarqué, en effet, dans les
notes de Berty sur la topographie des quartiers du Louvre et du bourg Saint-
Germain, plusieurs maisons de l’Épée, et même une maison de l’Épée
rompue. Il en devait être de même des enseignes de tous les maîtres joueurs
d’instruments, qui avaient une salle de musique: les joueurs de luth
exposaient pour enseigne un luth; les joueurs de cor de chasse, un cor; les
joueurs de hautbois, un hautbois, etc. D’après ce système, une académie
pour l’équitation prenait pour enseigne un cheval ou une tête de cheval,
sinon une selle, un mors, une bride. Quant aux académies des jeux, comme
elles avaient été défendues, la plupart n’avaient garde de se trahir par
quelque autre signe extérieur qu’une simple lanterne, qui devait être
indispensable en un temps où les rues de Paris n’étaient pas éclairées.
Si l’Académie royale de danse, dont nous venons de parler plus haut,
avait au moins un violon pour enseigne, on peut supposer que l’Académie
royale de musique se donna aussi le luxe d’une enseigne et y mit en montre
tous les instruments de son orchestre, lorsque cette académie, créée par Lully
en vertu des lettres patentes que le roi lui avait accordées au mois de mai
1672, s’établit d’abord dans une salle provisoire, qui n’était qu’un jeu de
paume, «près Luxembourg, vis-à-vis Bel-Air», suivant l’adresse que nous
fournit l’opéra des Fêtes de l’Amour et de Bacchus, représenté et imprimé en
1672. Lully avait attribué certainement une enseigne à son théâtre, puisqu’il
en avait appliqué une à sa maison, qu’il bâtissait simultanément au coin de la
rue Sainte-Anne et de la rue Neuve-des-Petits-Champs. Cette vieille maison
existe encore, avec sa décoration que nous avons appelée une enseigne
lyrique: «Au-dessus de la haute fenêtre qui occupe le milieu de la principale
façade, disons-nous dans un de nos ouvrages[140], se voient comme sculptés
dans la pierre plusieurs des attributs qui rappellent le premier propriétaire.
Ce sont des instruments de musique, une timbale, des trompettes, une
guitare, etc. Des masques de théâtre servent de clefs de voûte aux cintres du
rez-de-chaussée et sont une allusion à l’origine de la fortune de celui qui fit
bâtir cette demeure.»
L’enseigne d’un théâtre était aussi un éclairage de lanternes, mais ces

lanternes portaient peut-être une inscription qui devenait lumineuse le soir
des représentations. Le Duchat avait dit, dans une note de la préface de son
édition des œuvres de Rabelais, commentées par lui (1711), que le nom du
Théâtre ou Jeu des Pois pilés venait d’une enseigne: «Espèces de farces
morales connues sous le nom de Poids pilés, et appelées de la sorte parce
qu’à la maison où on les représentoit, à Paris, pendoit pour enseigne une pile
de poids à peser.» (Voir Fœneste, liv. III, chap. X.) Mais Le Duchat s’est
démenti lui-même dans une note sur ce passage des Aventures du baron de
Fœneste, par Agrippa d’Aubigné, où il représente les pois pilés comme une
purée de pois, à laquelle on faisait allusion avec dédain en parlant des
premières et grossières farces du théâtre français. Nous ne sommes pas
éloigné de croire que Le Duchat avait raison dans sa première interprétation
du nom de ce théâtre des Poids pilés, qui avait son siège aux Halles et
probablement dans une grande maison, une espèce de halle fermée où était le
Poids du Roi. Cette halle ou maison était située dans la rue de la Buffeterie
ou des Lombards. Il suffira de citer un seul document contemporain de
l’époque où le jeu des pois pilés fut établi dans le quartier des Halles par la
joyeuse bande des Enfans sans souci: «Lettre de l’an 1471, du quatorzième
octobre, par laquelle Marguerite de la Roche-Guyot vend au Chapitre le
Poids le Roi, avec le lieu où il se tient en la rue de la Buffeterie, alias des
Lombards, avec le Poids de la Cire, deux mille sept cens soixante-quatorze
livres, 12 sous[141].»
On peut donc dire avec certitude que l’enseigne du Poids du Roi était une
pyramide de poids empilés, par ordre de grosseur, en commençant par les
plus gros et en finissant par les plus petits. Il y avait d’ailleurs plusieurs
maisons du même genre pour le pesage des marchandises. Berty, dans sa
Topographie du vieux Paris, cite une maison des Balances, en 1372, dans la
rue du Coq. La rue des Billettes avait aussi pris son nom de l’enseigne des
billes, ou du billot, qu’on pendait à la maison où se payait quelque péage au
profit du roi ou de la ville[142].
Les étuves, qui s’étaient tant multipliées depuis le XIIIᵉ siècle, avaient
également des enseignes, mais nous n’en avons pu découvrir que deux[143],
l’une dans le Compte des confiscations de Paris en 1421, pour les Anglais
qui étaient alors maîtres de la ville, l’autre dans le Compte du Domaine de
Paris pour un an fini à la saint Jean-Baptiste 1439: «Maison en laquelle a
estuves à femmes, scise rue de la Huchette, où est l’enseigne des Deux
Bœufs, faisant le coin de ladite rue, près de l’Abreuvoir du Pont-
Neuf.»—«Hôtel de l’Arbalestre, dans la rue de la Huchette, tenant du côté
du Petit-Pont à l’hostel de Pontigny, et du côté du pont Saint-Michel, à
l’hostel des Bœufs, etc.; à l’Arbalestre, il y avoit estuves pour hommes, et
aux Bœufs, estuves pour femmes.» Sauval, qui nous offre ce renseignement
précieux par sa rareté, avait dit dans son Histoire des Antiquités de la ville de
Paris[144]: «Vers la fin du siècle passé, on a cessé d’aller aux étuves.
Auparavant elles étaient si communes, qu’on ne pouvoit faire un pas sans en
rencontrer.» De ces étuves, il était resté jusqu’à nos jours des noms de rues:
rues des Vieilles-Étuves-Saint-Martin, rue des Vieilles-Étuves-Saint-Honoré,
la ruelle des Étuves, près la rue de la Huchette, etc. Mais les enseignes de ces
étuves, enseignes qui étaient sans doute curieuses pour l’histoire des mœurs,
n’avaient pas même laissé un souvenir.
Des étuves et des maisons de baigneurs aux maisons de débauche, il n’y
avait qu’un pas. Le savant Duméril, dans son ouvrage sur les Formes du
mariage, soutient qu’au moyen âge un bouchon de paille servait d’enseigne
aux prostituées, et aujourd’hui même un lien de paille est encore employé
dans les rues pour indiquer un objet quelconque à vendre. Mais les mauvais
lieux avaient d’autres enseignes plus caractéristiques, la prostitution formant
un corps de métier. Ainsi, l’abbé Perau, dans ses additions à la Description
historique de la ville de Paris, par Piganiol de la Force[145], affirme que le
nom du quartier du Gros-Caillou n’a pas eu d’autre origine qu’une maison
de débauche: «Il faut dire, à présent, l’origine de ce nom singulier de Gros
Caillou, qui lui fut donné après son nom très ancien de la Longray. Dans le
lieu où est aujourd’hui sise l’église, étoit une maison publique de débauche,
à laquelle un caillou énorme servoit d’enseigne, et l’on fut obligé
d’employer la poudre pour le détruire et élever en sa place la croix qui y est
aujourd’hui et l’église à la place de la maison.» L’abbé Perau n’a pas eu
égard au nom primitif de la Longray, qui peut avoir été le nom vulgaire
d’une pierre druidique, d’une espèce de dolmen: le long grais ou gray. En
effet, Jaillot, dans ses Recherches sur la ville de Paris, dit que ce gros caillou
était une borne servant de limite entre les seigneuries de Saint-Germain-des-
Prés et de Sainte-Geneviève. L’enseigne de la maison de débauche n’en était
pas moins la tradition populaire du quartier.
On connaît du moins avec certitude les enseignes d’un grand nombre de
jeux de paume, qui ont été souvent nommés dans l’histoire et dont Adolphe
Berty a recueilli les noms seulement pour les quartiers du Louvre et du
bourg Saint-Germain.
Le Journal d’un bourgeois de Paris, sous le règne du roi Charles VII,
mentionne le jeu de paume du Petit Temple rue Grenier-Saint-Lazare, où l’on
voyait la belle Margot jouer à la paume avec la main nue, en guise de
raquette. L’Illustre Théâtre, où Molière fit ses débuts de comédien avec la
troupe des Béjart, s’installa, en 1643, dans le jeu de paume des Métayers,
situé près de la porte de Nesle; il fut transporté ensuite au jeu de paume de la
Croix noire, situé rue des Barrés, près du port Saint-Paul; puis enfin, il alla
terminer sa carrière dramatique, en 1646, au jeu de paume de la Croix

blanche, rue de Bucy. Ce sont là les faits les plus intéressants de l’histoire
des jeux de paume, que nous ne nommerons pas tous, car on n’a pas relevé
encore leurs annales dans tous les quartiers du vieux Paris, où ils furent si
nombreux au XVIIᵉ siècle, lorsque la paume était l’exercice favori de la
jeunesse. Ils ont disparu, la plupart, depuis que le jeu de billard, qui a
remplacé la paume, s’est intronisé dans les estaminets et les cafés. Un des
derniers jeux de paume qu’on ait vus à Paris, dans la rue Saint-Victor, avait
une enseigne fort curieuse pour l’archéologie, puisqu’elle représentait deux
joueurs en action, avec le costume qu’ils portaient il y a cent cinquante ans
environ. Nous attribuerions aussi à un jeu de paume, sinon à un marchand de
raquettes et d’éteufs ou de balles de paume, l’enseigne d’une maison: A la
Raquette, que nous avons fait dessiner, il y a vingt ans, au coin de la rue
Charlemagne et de la rue des Nonnains-d’Hyères.
Contentons-nous de donner comme spécimen la liste des jeux de paume

désignés par leurs enseignes, tels que Berty les a trouvés dans l’ancien bourg
de Saint-Germain-des-Prés.
Rue des Boucheries. Maison et jeu de paume du Dauphin, aboutissant à la
rue des Quatre-Vents, en 1523.—Jeu de paume du Château de Milan.
Rue de l’Ancienne-Comédie. Jeu de paume de l’Écu de Savoie, en 1523.
—Jeu de paume de l’Écu, en 1592, dans le grand hôtel de l’Écu de France.
Le jeu de paume de l’Écu de Savoie occupait un espace de terrain si
considérable, qu’on y bâtit au XVIIIᵉ siècle plusieurs maisons, dont les
enseignes furent l’Écritoire, la Talemouse, la Tour d’Argent, le Champ des
Oiseaux, l’Ane vert, les Clefs, et la Rose rouge. Ces noms d’enseigne
n’annonçaient pas des maisons très respectables.
Le jeu de paume de l’Étoile, qui existait en 1547, fut remplacé, à la fin du
XVIIᵉ siècle, par la nouvelle salle de la Comédie française, au nº 14 de la rue
de l’Ancienne-Comédie, qui lui doit son nom.
Rue de Seine. Les jeux de paume furent plus nombreux dans cette rue-là
que dans tout le quartier Saint-Germain. Voici les noms des principaux: Jeu
de paume de Fort Affaire, 1588.—Jeu de paume des Deux Anges, 1593.—
Jeu de paume des Trois Cygnes, 1595.—Jeu de paume du Soleil d’Or, 1595.
—Jeu de paume de la Bouteille, 1600.—Jeu de paume Saint-Nicolas, 1617.
—Jeu de paume des Trois Torches, 1687.
Terminons par une remarque qui a échappé à Berty, le dépisteur de tous
ces jeux de paume: c’est que le jeu de paume des Métayers, près de la porte
de Nesle, où Molière parut sur la scène pour la première fois avec la troupe
de l’Illustre Théâtre, a subsisté bien plus longtemps qu’en 1790. Nous
sommes presque certain qu’il conservait sa première destination, sans avoir
changé d’aspect, en 1818, ayant toujours ses deux entrées, l’une dans la rue
de Seine et l’autre dans la rue Mazarine. Il ne fut détruit qu’en 1823, lors de
l’ouverture du passage du Pont-Neuf; mais on reconnaît, au nº 42 de la rue
Mazarine, l’entrée et l’allée obscure qui conduisaient au jeu de paume. En
1818, tous les habitants du quartier, fidèles gardiens de la tradition,
appelaient encore ce vieux jeu de paume le Théâtre de Molière.
XIII
LES VIEILLES ENSEIGNES
O N s’est demandé souvent autrefois proverbialement: «Où vont les

vieilles lunes?» On aurait pu se demander aussi: «Que deviennent les
vieilles enseignes?»
Sans doute, la pluie, la sécheresse, le soleil, l’humidité et la poussière
faisaient leur œuvre sur ces enseignes, exposées à toutes les intempéries de
l’air et des saisons, pendant de longues années; mais si on ne les repeignait
pas, si on ne les nettoyait pas de temps à autre, on les changeait trois ou
quatre fois dans un siècle, et les vieilles enseignes n’étaient pas condamnées
à faire des fagots pour allumer du feu. Il y avait sans doute des vendeurs et
des acheteurs pour ces vieilles enseignes, qui passaient d’une maison à une
autre et servaient tour à tour à recommander différentes industries et
différents commerces; car l’enseigne n’avait pas toujours un rapport direct et
caractérisé avec la profession de l’artisan, qui la choisissait par caprice ou
par hasard. C’étaient aussi les mêmes enseignes qu’on voyait répétées dans
le même quartier et dans la même rue, elles ne différaient souvent que de
couleur: s’il y avait trois Croix, trois Lions, trois Chevaux, trois Pots, trois
Cages, trois Paniers, à côté l’un de l’autre, chacune de ces enseignes se
distinguait par une couleur spéciale, de manière à ce que les mêmes signes
distinctifs, représentés et dénommés dans plusieurs enseignes voisines, ne
fussent jamais confondus entre eux, puisqu’ils n’avaient pas d’autre objet
que de désigner une maison ou une boutique; ainsi la Croix d’Or n’était pas
la Croix d’Argent, le Grand Lion n’était pas le Petit Lion, le Cheval blanc
n’était pas le Cheval rouge, le Pot d’Étain n’était pas le Pot de Cuivre, la
Cage bleue n’était pas la Cage noire, le Panier vert n’était pas le Panier
fleuri.
Les signes distinctifs des enseignes ne variaient donc pas à l’infini,
comme on paraît le croire, et leur ressemblance même n’avait rien qui pût
déplaire au marchand ou au propriétaire. Il suffisait qu’il n’y eût pas deux
enseignes absolument semblables dans la même rue. Un changement
d’enseigne ne pouvait être déterminé que par une circonstance indépendante
de l’enseigne elle-même, car ordinairement l’ancienneté d’une enseigne en
faisait la valeur. Aussi, nous avons remarqué que si la maison changeait
d’enseigne deux ou trois fois en un siècle, la boutique n’en changeait pas, à
moins de changer de destination commerciale. Ce sont là des raisons qui
nous font penser que les enseignes ne se détruisaient pas, en cessant
d’appartenir à telle maison ou à telle boutique, et que l’acquéreur ne
manquait pas, pour les transporter d’un lieu à un autre et pour leur donner
une nouvelle existence en les attachant à un nouveau commerce ou à un
nouveau local. Cependant, nous n’avons pas réussi à découvrir quels étaient
les marchands qui à une époque reculée vendaient les vieilles enseignes
d’occasion, en les faisant réparer et repeindre.
Ce n’est qu’au XVIIIᵉ siècle que nous trouvons ces marchands-là: «Chez
les marchands de ferraille du quai de la Mégisserie sont des magasins de
vieilles enseignes, dit Mercier[146], propres à décorer l’entrée de tous les
cabarets et tabagies des faubourgs et de la banlieue de Paris. Là, tous les rois
de la terre dorment ensemble: Louis XVI et Georges III se baisent
fraternellement, le roi de Prusse couche avec l’impératrice de Russie,
l’Empereur est de niveau avec les Électeurs; là, enfin, la tiare et le turban se
confondent. Un cabaretier arrive, remue avec le pied toutes ces têtes
couronnées, les examine, prend au hasard la figure du roi de Pologne,
l’emporte, l’accroche et écrit dessous: Au Grand Vainqueur.» Mais il ne
s’agit ici que de têtes peintes représentant des portraits de rois et de reines,
qui étaient en faveur, à ce qu’il paraît, auprès des cabaretiers de Paris et de la
banlieue. Ces marchands de ferraille avaient à leur disposition les peintres
d’enseigne pour rafraîchir et enjoliver la marchandise au plus juste prix: «Un
autre gargotier demande une impératrice; il veut que sa gorge soit
boursouflée, et le peintre, sortant de la taverne voisine, fait présent d’une
gorge rebondie à toutes les princesses de l’Europe.» La police, si tracassière
et si épineuse pour tant de sujets indifférents, ne prenait pas sous sa
protection ces pauvres souverains, auxquels le peintre donnait «un air hagard
ou burlesque, des yeux éraillés, un nez de travers, une bouche énorme.» On
se contentait d’exiger que la légende de l’enseigne ne fût pas injurieuse.
«Quand je vois, ajoute philosophiquement Mercier, toutes ces vieilles
enseignes, pêle-mêle confondues, comme on les change, comme on les
marchande; quand je songe aux destinées qui promènent de cabarets en
cabarets ces grotesques portraits de souverains; au vent qui les ballotte, aux

Rational Rules Towards A Theory of Moral Learning Shaun Nichols Full Chapter PDF

Uploaded by

Copyright:

Available Formats

You might also like

Rational Rules Towards A Theory of Moral Learning Shaun Nichols Full Chapter PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Rational Rules Towards A Theory of Moral Learning Shaun Nichols Full Chapter PDF

Uploaded by

Copyright:

Available Formats

Rational Rules: Towards a Theory of

Moral Learning Shaun Nichols

The Evolution Of Moral Progress: A Biocultural Theory

From Deep Learning to Rational Machines [converted]

From Deep Learning to Rational Machines 1st Edition

Einstein's Unfinished Dream: Practical Progress Towards

Classics of Moral and Political Theory 5th Edition,

Moral Error Theory 1st Edition Wouter Floris Kalf

A Translation Theory of Knowledge Transfer: Learning

The Uses of Delusion: Why It's Not Always Rational to

British Library Cataloguing in Publication Data

For Sarah and Julia

I. RATIONALITY AND RULES

II. STATISTICAL LEARNING OF

III. PHILOSOPHICAL IMPLICATIONS

My ﬁrst book in moral psychology, Sentimental Rules, emphasized the role

rational learning from the evidence. To locate this in contemporary moral

ﬂexible moral learners, with no innate constraints on the kinds of rules

form of rational sentimentalism, where rational learning and emotions both

matters through conversations, discussions in Q&A, and comments on

1.1 Stimuli for probabilistic inference task 18

4.1 Complete list of violations for overhypothesis study 90

Footbridge: Is it permissible for Frank to push a man off a footbridge and

Footbridge-Allow: Is it permissible for Jeff not to pull a lever that would

Bystander: Is it permissible for Dennis to pull a lever that redirects a

Authority dependence: If the teacher didn’t have a rule against hitting,

  :    5

Disagreement: If John and Mark make different judgments about whether

2. A Challenge to Common-Sense Morality

2.1 Debunking Normative Ethics

reasoning. System 2 processes are ﬂexible, domain general, sensitive to new

  :    7

two processes: the moral condemnation is driven by an affective intuition

[T]he consequentialist weighing of harms and beneﬁts is a weighing

tional reactions that are not responsive to rational considerations, we should

2.2 Debunking Metaethics

and the best-known proposals serve as debunking explanations. In his

² In addition, recent work indicates that Bystander is just as emotionally arousing as

  :    9

Integral emotions may contain information that should appropriately

universalism. However, there is a pressing limitation of the study.

3.1 The Many Rationalisms

“Rationalism” is used in several strikingly different ways in philosophy. For

their evidence. I will argue that several key distinctions in common-sense

  :    11

rationality directly to action. Evidentialist approaches to rationality can, of

necessary for responding to the evidence.⁶ Algorithms might be responsive

[T]o be rational is to reason in accordance with principles of reasoning that

  :    13

Marr illustrates different levels of analysis with the example of a lowly

3.3 Rational Processes

We can now recruit Marr’s distinctions to impose greater speciﬁcity in

start by trying to show that a process is rational at the computational level,

  :    15

4.1 Statistical Learning

Some kinds of statistical learning involve extremely sophisticated techniques

A panel of psychologists have interviewed and administered personality

On the basis of this information, thumbnail descriptions of the 30 engin-