Download the full version of the ebook at

https://ebookmass.com

Algorithms and Practice Zhu

https://ebookmass.com/product/algorithms-and-
practice-zhu/

Explore and download more ebook at https://ebookmass.com

 Recommended digital products (PDF, EPUB, MOBI) that
you can download immediately if you are interested.

Zhu Xi: Selected Writings Zhu Xi

https://ebookmass.com/product/zhu-xi-selected-writings-zhu-xi/

testbankdeal.com

Asymmetric Organo-Metal Catalysis: Concepts, Principles,

and Applications Liu-Zhu Gong

https://ebookmass.com/product/asymmetric-organo-metal-catalysis-
concepts-principles-and-applications-liu-zhu-gong/

testbankdeal.com

Absolute Beginner's Guide to Algorithms: A Practical

Introduction to Data Structures and Algorithms in
JavaScript Kirupa Chinnathambi
https://ebookmass.com/product/absolute-beginners-guide-to-algorithms-
a-practical-introduction-to-data-structures-and-algorithms-in-
javascript-kirupa-chinnathambi/
testbankdeal.com

Data Structures and Algorithms in C++ 2nd Edition

https://ebookmass.com/product/data-structures-and-algorithms-in-c-2nd-
edition/

testbankdeal.com
Absolute Beginner’s Guide to Algorithms: A Practical
Introduction to Data Structures and Algorithms in
JavaScript (for True Epub) Kirupa Chinnathambi
https://ebookmass.com/product/absolute-beginners-guide-to-algorithms-
a-practical-introduction-to-data-structures-and-algorithms-in-
javascript-for-true-epub-kirupa-chinnathambi/
testbankdeal.com

Chapter 12 - Neural network potentials Jinzhe Zeng & Liqun

Cao & Tong Zhu

https://ebookmass.com/product/chapter-12-neural-network-potentials-
jinzhe-zeng-liqun-cao-tong-zhu/

testbankdeal.com

Introduction to algorithms for data mining and machine

learning Yang

https://ebookmass.com/product/introduction-to-algorithms-for-data-
mining-and-machine-learning-yang/

testbankdeal.com

Machine Learning Algorithms for Signal and Image

Processing Suman Lata Tripathi

https://ebookmass.com/product/machine-learning-algorithms-for-signal-
and-image-processing-suman-lata-tripathi/

testbankdeal.com

Hidden Semi-Markov models : theory, algorithms and

applications 1st Edition Yu

https://ebookmass.com/product/hidden-semi-markov-models-theory-
algorithms-and-applications-1st-edition-yu/

testbankdeal.com
Machine Reading
Comprehension

Algorithms and Practice

Chenguang Zhu
Microsoft Corporation, Redmond, WA, United States
Table of Contents

Cover image

Title page
g

Copyright

About the author

Foreword by Xuedong Huang

Foreword by Zide Du

Preface

Acknowledgment

Recommendation
Part I: Foundation

Chapter 1. Introduction to machine reading comprehension

 Abstract

1.1 The machine reading comprehension task

1.2 Natural language processing

1.3 Deep learning

1.4 Evaluation of machine reading comprehension

1.5 Machine reading comprehension datasets

1.6 How to make an machine reading comprehension dataset

1.7 Summary

References

Chapter 2. The basics of natural language processing

Abstract

2.1 Tokenization

2.2 The cornerstone of natural language processing: word

vectors

2.3 Linguistic tagging

2.4 Language model

2.5 Summary

Reference

Chapter 3. Deep learning in natural language processing

 Abstract

3.1 From word vector to text vector

3.2 Answer multiple-choice questions: natural language

understanding

3.3 Write an article: natural language generation

3.4 Keep focused: attention mechanism

3.5 Summary
Part II: Architecture

Chapter 4. Architecture of machine reading comprehension models

Abstract

4.1 General architecture of machine reading comprehension

models

4.2 Encoding layer

4.3 Interaction layer

4.4 Output layer

4.5 Summary

References

Chapter 5. Common machine reading comprehension models

Abstract

5.1 Bidirectional attention flow model

 5.2 R-NET

5.3 FusionNet

5.4 Essential-term-aware retriever–reader

5.5 Summary

References

Chapter 6. Pretrained language models

Abstract

6.1 Pretrained models and transfer learning

6.2 Translation-based pretrained language model: CoVe

6.3 Pretrained language model ELMo

6.4 The generative pretraining language model: generative pre-

training (GPT)

6.5 The phenomenal pretrained language model: BERT

6.6 Summary

References
Part III: Application

Chapter 7. Code analysis of the SDNet model

Abstract

7.1 Multiturn conversational machine reading comprehension

model: SDNet
 7.2 Introduction to code

7.3 Preprocessing

7.4 Training

7.5 Batch generator

7.6 SDNet model

7.7 Summary

Reference

Chapter 8. Applications and future of machine reading

comprehension

Abstract

8.1 Intelligent customer service

8.2 Search engine

8.3 Health care

8.4 Laws

8.5 Finance

8.6 Education

8.7 The future of machine reading comprehension

8.8 Summary

References
Appendix A. Machine learning basics

A.1 Types of machine learning

A.2 Model and parameters

A.3 Generalization and overfitting

Appendix B. Deep learning basics

B.1 Neural network

B.2 Common types of neural network in deep learning

B.3 The deep learning framework PyTorch

Index
Copyright
Elsevier
Radarweg 29, PO Box 211, 1000 AE Amsterdam, Netherlands
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United
Kingdom
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States

Copyright © 2021 Beijing Huazhang Graphics & Information Co.,

Ltd/China Machine Press. Published by Elsevier Inc. All rights
reserved.

No part of this publication may be reproduced or transmitted in any

form or by any means, electronic or mechanical, including
photocopying, recording, or any information storage and retrieval
system, without permission in writing from the publisher. Details on
how to seek permission, further information about the Publisher’s
permissions policies and our arrangements with organizations such
as the Copyright Clearance Center and the Copyright Licensing
Agency, can be found at our website:
www.elsevier.com/permissions.

This book and the individual contributions contained in it are

protected under copyright by the Publisher (other than as may be
noted herein).

Notices
Knowledge and best practice in this field are constantly changing.
As new research and experience broaden our understanding,
 Visit https://ebookmass.com
now to explore a rich
collection of eBooks and enjoy
exciting offers!
changes in research methods, professional practices, or medical
treatment may become necessary.

Practitioners and researchers must always rely on their own

experience and knowledge in evaluating and using any
information, methods, compounds, or experiments described
herein. In using such information or methods they should be
mindful of their own safety and the safety of others, including
parties for whom they have a professional responsibility.

To the fullest extent of the law, neither the Publisher nor the
authors, contributors, or editors, assume any liability for any injury
and/or damage to persons or property as a matter of products
liability, negligence or otherwise, or from any use or operation of
any methods, products, instructions, or ideas contained in the
material herein.

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library

Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of
Congress

ISBN: 978-0-323-90118-5

For Information on all Elsevier publications visit our website at

https://www.elsevier.com/books-and-journals

Publisher: Glyn Jones

Editorial Project Manager: Naomi Robertson
Production Project Manager: Punithavathy Govindaradjane
Cover Designer: Mark Rogers

Typeset by MPS Limited, Chennai, India

About the author
Dr. Chenguang Zhu is a Principal Research Manager in the
Microsoft Corporation. Dr. Zhu obtained his PhD in Computer
Science from Stanford University, United States. He is leading efforts
in the research and productization of natural language processing in
Azure Cognitive AI. He is proficient in artificial intelligence, deep
learning, and natural language processing, specializing in machine
reading comprehension, text summarization, and dialogue
understanding. He has led teams to win the first place in the SQuAD
1.0 Machine Reading Comprehension Competition held by Stanford
University, and reach human parity in the CoQA Conversational
Reading Comprehension Competition. He has 40 papers published
in top AI and NLP conferences, such as ACL, EMNLP, NAACL, and
ICLR, with more than 1000 citations.
Foreword by Xuedong Huang
Xuedong Huang Dr. , CTO of Artificial Intelligence, Microsoft, Redmond, WA,
United States

There are two levels of intelligence. One is perceptual intelligence,

which enables the computer to see, hear, and touch. In these areas,
artificial intelligence has made many breakthroughs, such as speech
recognition, speech synthesis, and computer vision. A higher level is
cognitive intelligence, which requires computers to understand and
analyze concepts, relationships, and logic. At this level, artificial
intelligence is still in its infancy.
As an important medium for human communication and
information dissemination, language is at the core of human
intelligence. From the Turing Test in the 1950s to the deep learning
era today, the understanding and application of natural language
has always been a hot research topic. If AI is the crown, speech and
language technologies are the jewels on top of the crown. It is fair to
say that if computers can fully understand human language, we’ll
have achieved strong artificial intelligence.
In recent years, machine reading comprehension has become one
of the most popular and cutting-edge directions in natural language
processing research. It has significant scientific and practical values
to enable computer models to read articles, analyze semantics, and
answer questions like humans. Machine reading comprehension
technology can automate plenty of time-consuming and laborious
text analysis work and greatly improve the productivity of many
applications, ranging from intelligent customer service to search
engines, from automatic essay scoring to intelligent finance.
 With the development of deep learning technology, the research of
machine reading comprehension has made tremendous progress. In
some specific tasks, the answers of computer models are already
comparable to the human level. Some media reports have even
claimed that the computer is superior to humans in reading
comprehension. However, existing models are still far from a
genuine and thorough understanding of text. In many cases, these
models still rely on simple matching of words and phrases, rather
than on a thorough understanding of the syntactic structure and
semantics.
In general, there are three key factors to the success of an artificial
intelligence system: platform, data, and algorithm. As the computing
power and magnitude of data continue to soar, the exploration and
improvement of algorithms have become a hotly contested spot for
artificial intelligence research.
Currently there are very few books on the market which have a
complete introduction to machine reading comprehension. Dr.
Chenguang Zhu in our team has worked deeply in this direction for
many years and has led the team to achieve top places in a number
of international contests. The purpose of his book is to objectively
show the field of machine reading comprehension to readers. The
book includes a detailed introduction to the latest research results
and thoughts on the future directions of machine reading
comprehension. I hope this book will inspire readers to work
together to achieve human-level machine reading comprehension.
Foreword by Zide Du
Zide Du, Former Researcher at the Institute of Computing Technology, Chinese
Academy of Sciences, Beijing, China, Secretary General of the Chinese Computer
Federation, Beijing, China

Natural language processing aims to solve the problem of

understanding and generating natural language. Natural language is
the jewel in the crown of artificial intelligence. It is one of the most
important abilities of computers, but is also a challenging direction
to study. Every human language has its own grammar, but because
of the different styles of usage, coupled with factors such as dialects
and idioms, the resulting forms of language have a large variation.
These variations usually do not interfere much with the
communication between humans, but it is very difficult for
computers to understand. This is because the current von Neumann
computer architecture is good at handling information with clear
rules, but is less capable of handling constantly evolving forms of
information.
Over the years, researchers have proposed and developed many
methods, ranging from rule-based linguistic techniques to models
based on statistical machine learning. In recent years, researchers
have developed end-to-end deep learning frameworks for natural
language processing, including word embeddings, attention
mechanisms, encoder–decoder architecture, and the recent,
pretraining models. These techniques have greatly improved
models’ ability to understand text and brought new interesting ideas
to natural language processing.
Machine reading comprehension is one of the most popular and
cutting-edge research topics in natural language processing. Reading
is the basic means for people to obtain information. Without reading
there is no understanding, and without understanding one cannot
communicate. There are already many chatbots in the market, but
people often find them responding off the point. The reason is that
the current technology is a black box approach based on text
matching. So the chatbot does not really understand what people
mean. As we know, humans communicate with context, so that we
can easily understand what other people are talking about via
referencing. However, it is very difficult to make machines
understand the context. In order to solve these problems, researchers
have proposed many ways to improve models’ ability to understand
dialog and articles. Moreover, the release of many reading
comprehension datasets have played an important role in promoting
the development of technology.
In addition to its research values, machine reading comprehension
has many meaningful applications. For example, machine-generated
article summaries can save a lot of time of reading the full text, and
the QA system can accurately find answers to user questions from a
large number of documents. Machine reading comprehension is also
the basis for translation and dialog, which are of great value to
computer-assisted services.
Chenguang’s book systematically introduces the key technologies
and progress in this area as well as existing challenges. I believe that
readers will have a clear understanding of the research and
application of this field after reading this book.
During high school, Chenguang participated in the National
Olympiad in Informatics (NOI) organized by the China Computer
Federation (CCF), winning gold medals in the national competition.
He was also a candidate for the Chinese team for the International
Olympiad in Informatics. As I was the chairman of NOI, I got to
know him back then. He later went to Tsinghua University to study
computer science, got a PhD at Stanford University, and now works
at Microsoft on natural language processing. We rarely see each
other, but constantly keep in touch. I think he is a talented young
scholar with a rigorous and very sensible style. I am therefore very
happy to discuss various issues with him. He asked me to write the
 ppy
foreword for his new book and I am very glad to see his research
progress. I also want to express my congratulations to him.
Preface
Chenguang Zhu
Reading is an important means for humans to acquire knowledge
and understand the world. For thousands of years, language has
been the carrier of human civilization, containing a wealth of
information, experience, and wisdom. Language’s nature of high
concentration makes reading ability a vital intelligence. The famous
science fiction writer, Mr. Liu Cixin, described the efficiency of
human language communication in his novel The Rural Teacher:

You’re trying to tell us that a species that has no memory inheritance,

communicates with each other in sound waves at an incredible rate of 1
to 10 bits per second, can create a 5B-level civilization?! And this
civilization has evolved on its own without any cultivation from
external advanced civilization?!

It is estimated that the average reading speed of humans is about

two to three times the speed of speech. Thus, under a rough
estimate, even if a person reads for 8 hours a day for 50 years, he or
she can only get about 1.5 GB of information out of reading.
However, human civilization has far exceeded this magnitude.
Therefore reading is a complex process of abstracting text into
concepts, ideas, and derived knowledge through understanding.
In today’s wave of artificial intelligence, it is very important to
enable the computer learn to read. On the one hand, reading ability
lies at the core of human intelligence, which is indispensable in the
ultimate form of artificial intelligence. On the other hand, with the
explosion of text data, computer models can automate the process of
 Visit https://ebookmass.com
now to explore a rich
collection of eBooks and enjoy
exciting offers!
text understanding, save a lot of cost and time, and have a wide
range of applications in many industries.
Thus in recent years machine reading comprehension (MRC) has
been one of the most cutting-edge research topics in natural
language processing. The goal of the study is to teach computers to
read articles and answer related questions like humans. Lots of AI
technologies have been applied to this field, and there have emerged
many MRC tasks. I was fortunate enough to be among the first
group of researchers in this field, designing and implementing
several models to win the first place in the SQuAD competition
hosted by Stanford University, and surpass human level
performance for the first time in the conversational MRC
competition CoQA.
However, as computer models outperform humans on more
datasets, many media reports use headlines like Computer has beaten
humans in reading, which contributes to the so-called claim that AI
has replaced humans. As a researcher in this area, I deeply feel that
computers are far from humans’ ability of reading and
understanding. Although current reading comprehension models
have achieved huge progress compared with a decade ago, they are
above the human level only in specific datasets under various
constraints. Studies have shown that the performance of these
models will significantly drop when a confusing sentence is
appended to the article, while humans find it very easy to judge.
In contrast to the boom of machine reading comprehension
research, there are so far no books on the market in this field. Most
progress is published in the form of academic papers, and little
information can be found about the application of MRC in industry.
Therefore the purpose of this book is to objectively show the status
quo of machine reading comprehension research. Starting from the
basic modules, the architecture of models, and to cutting-edge
algorithms, the book details the design and implementation of
machine reading comprehension models. There are numerous code
examples in Python and PyTorch by the author, showcasing the
model building process, which has a high practical value. All the
code is available at https://github.com/zcgzcgzcg1/MRC_book_en. In
 p g g g g
addition, the book introduces the landing of machine reading
comprehension technology in various industrial applications, such
as intelligent customer service and search engines, and points out
the challenges and future directions of MRC research.
Although currently machines are still inferior to humans in terms
of reading ability, we can leverage the high speed and large storage
of computers to overtake humans. As said in the ancient Chinese
proverb, When one learns 300 poems of the Tang Dynasty by heart, he is
sure to be able to write poetry. Nowadays, computers can read a million
poems in a split second. So there is every reason to expect a
breakthrough in machine reading comprehension. For example, the
BERT model combines the merits of massive data and large models
to make breakthroughs in many areas of natural language
processing including machine reading comprehension. I hope that
this book can inspire readers to make computers achieve and exceed
the reading ability of humans in the near future.
The book is divided into three parts, with a total of eight chapters.
The first part is Foundation (Chapters 1–3), which introduces the
basics of machine reading comprehension and key technologies.
These include the definition of the MRC task, natural language
processing techniques used in MRC models, and related deep
learning network modules, such as how to represent articles and
questions in computers, how to answer multiple-choice questions,
and how to generate freestyle answers.
The second part is Architecture (Chapters 4–6), which introduces
the basic model architecture and popular models for machine
reading comprehension. It also analyzes the state-of-the-art
pretrained models, for example, GPT and BERT, that have had a
revolutionary impact on machine reading comprehension research.
The third part is Application (Chapter 7: Code Analysis of SDNet
Model and Chapter 8: Applications and Future of Machine Reading
Comprehension), including the code analysis of the SDNet model
which won first place in the CoQA MRC competition in 2018, the
process of landing machine reading comprehension technology in
various industrial applications, as well as challenges and future
directions of MRC research.
Random documents with unrelated
content Scribd suggests to you:
the mythic type of anecdote, the Fable, is judged by precisely
opposite criteria.
398. See pp. 143 et seq.
399. The Fates of the Greeks are represented as spinning,
measuring out and cutting the thread of a man’s destiny, but not as
weaving it into the web of his life. It is a mere dimension.—Tr.
400. See p. 129.
401. The evolution of meaning in the Classical words pathos and
passico corresponds with this. The second was formed from the first
only in the Imperial period, and carried its original sense in the
“Passion” of Christ. It was in the early Gothic times, and particularly
in the language of the Franciscan “Zealots” and the disciples of
Joachim of Floris, that its meaning underwent the decisive reversal.
Expressing thenceforward a condition of profound excitement which
strained to discharge itself, it became finally a generic name for all
spiritual dynamic; in this sense of strong will and directional energy it
was brought into German as Leidenschaft by Zesen in 1647.
402. The Eleusinian mysteries contained no secrets at all.
Everyone knew what went on. But upon the believers they exercised
a strange and overpowering effect, and the “betrayal” consisted in
profaning them by imitating their holy forms outside the temple-
precinct. See, further, A. Dieterich, Kleine Schriften (1911), pp. 414
et seq.
403. See Vol. II, pp. 345 et seq.
404. The dancers were goats, Silenus as leader of the dance wore
a horsetail, but Aristophanes’s “Birds,” “Frogs” and “Wasps” suggest
that there were still other animal disguises.
405. See pp. 283 et seq.
406. As the student of cultural history to-day is not necessarily
familiar with technical Greek, it may be helpful to reproduce from
Cornish’s edition of Smith’s “Greek and Roman Antiquities,” s.v.
“Tragoedia,” the following paragraph, as clear as it is succinct:
“Tragedy is described by Aristotle (Poet., VI, 2) as effecting by
means of pity and terror that purgation [of the soul] (κάθαρσις) which
belongs to [is proper for] such feelings.”... Tragedy excites pity and
terror by presenting to the mind things which are truly pitiable and
terrible. When pity and terror are moved, as tragedy moves them, by
a worthy cause, then the mind experiences that sense of relief which
comes from finding an outlet for a natural energy. And thus the
impressions made by Tragedy leave behind them in the spectator a
temperate and harmonious state of the soul. Similarly Aristotle
speaks of the enthusiastic worshippers of Dionysus as obtaining a
κάθαρσις, a healthful relief, by the “lyric utterance of their sacred
frenzy.”—Tr.
407. The evolution of ideals of stage-presentation in the minds of
Æschylus, Sophocles and Euripides successively is perhaps
comparable with that of sculptural style which we see in the
pediments of Ægina, of Olympia and of the Parthenon.
408. It must be repeated that the Hellenistic shadow-painting of
Zeuxis and Apollodorus is a modelling of the individual body for the
purpose of producing the plastic effect on the eye. There was no
idea of rendering space by means of light and shade. The body is
“shaded” but it casts no shadow.
(Contrast with this Dante’s exact and careful specification of the
time-of-day in every episode of the Purgatorio and the Paradiso,
sublimely imaginative as these poems are.—Tr.)
409. The great mass of Socialists would cease to be Socialists if
they could understand the Socialism of the nine or ten men who to-
day grasp it with the full historical consequences that it involves.
410. See p. 239 et seq.
411. See p. 68.
412. See Vol. II, p. 363, note.
413. As we increase the powers of the telescope we find that the
number of newly appearing stars falls off rapidly towards the edges
of the field.
414. The thrill of big figures is a feeling peculiar to Western
mankind. In the Civilization of to-day this significant passion for
gigantic sums, for indefinitely big and indefinitely minute
measurements, for “records” and statistics, is playing a conspicuous
part.
(Our very notation of number is ceasing to rest on sense-
standards. Science has carried number, as ordinarily written, so high
and so low that it now uses a movable base for its numerical
statements. For example, a number in astronomy is written, not as
3,450,000,000 but as 3.45 × 109, one relating to ordinary experience
as 3.45 (i.e., 3.45 × 100) and one in electromagnetic theory, not as
0.00000345 but as 3.45 × 10-6. Under this system the conceptual
unit may be as large or as small, compared with the unit of daily
experience, as the region of thought in which the calculation is taking
place requires. And different conceptual worlds can be connected as
to number [say, a number of kilometres brought into an order of
thought that deals with millimetres] by simply changing the ten-
power.—Tr.)
415. In stellar calculations even the mean radius of the earth’s
orbit (1.493 × 1013 cm.) hardly suffices as unit, as the distance of a
star of one second parallax is already 206,265 such units away from
us; star-distances are reckoned therefore either in light-years or in
terms of the unit distance of a star of this standard parallax.—Tr.
416. As early as the second millennium before Christ they worked
from Iceland and the North Sea past Finisterre to the Canaries and
West Africa. An echo of these voyagings lingers in the Atlantis-saga
of the Greeks. The realm of Tartessus (at the mouth of the
Guadalquivir) appears to have been a centre of these movements
(see Leo Frobenius, Das unbekannte Afrika, p. 139). Some sort of
relation, too, there must have been between them and the
movements of the “sea peoples,” Viking swarms which after long
land-wanderings from North to South built themselves ships again on
the Black Sea or the Ægean and burst out against Egypt from the
time of Rameses II (1292-1225). The Egyptian reliefs show their
ship-types to have been quite different from the native and the
Phœnician; but they may well have been similar to those that Cæsar
found afterwards among the Veneti of Brittany. A later example of
such outbursts is afforded by the Varyags or Varangians in Russia
and at Constantinople. No doubt more light will shortly be thrown on
the courses of these movement-streams.
417. Here there is no need to postulate firearms (as distinct from
gunpowder used in fireworks) in the Chinese Culture. The archery of
the Chinese and Japanese was such as only the British 14th-century
archery could match in the Western and nothing in the Classical.
It should be noted also that it was in our 14th Century that—quite
independently of gunpowder—archery and the construction of siege-
engines reached their zenith in the West. The “English” bow had
long been used by the Welsh, but it was left to Edward I and Edward
III to make it the tactical weapon par excellence.—Tr.
418. See Vol. II, pp. 626 et seq.
419. Half as long again as Nelson’s Victory and about the same
length as the last wooden steam three-deckers (e.g., Duke of
Wellington) of the mid-19th Century.—Tr.
420. See Vol. II, pp. 207 et seq., and Chapter IV B.
421. See Vol. II, p. 80.
422. I.e., adherents of the various syncretic cults. Sec Vol. II, pp.
212 et seq.
423. This applies even more forcibly to the other “long-range”
episode, that of the Ten Thousand (Xenophon, Anabasis I).—Tr.
424. In this place it is exclusively with the conscious, religio-
philosophical morale—the morale which can be known and taught
and followed—that we are concerned, and not with the racial rhythm
of Life, the habit, Sitte, ἦθος, that is unconsciously present. The
morale with which we are dealing turns upon intellectual concepts of
Virtue and Vice, good and bad; the other, upon ideals in the blood
such as honour, loyalty, bravery, the feeling that attributes nobility
and vulgarity. See Vol. II, 421 et seq.
425. The original is here expanded a little for the sake of clarity.—
Tr.
426. After what has been said above regarding the absence of
pregnant words for “will” and “space” in the Classical tongues, the
reader will not be surprised to hear that neither Greek nor Latin
affords exact equivalents for these words action and activity.
427. See Vol. II, pp. 293 et seq.
428. “He who hath ears to hear, let him hear”—there is no claim to
power in these words. But the Western Church never conceived its
mission thus. The “Glad Tidings” of Jesus, like those of Zoroaster, of
Mani, of Mahomet, of the Neo-Platonists and of all the cognate
Magian religions were mystic benefits displayed but in nowise
imposed. Youthful Christianity, when it had flowed into the Western
world, merely imitated the missionarism of the later Stoa, itself by
that time thoroughly Magian. Paul may be thought of as urgent; the
itinerant preachers of the Stoa were certainly so, as we know from
our authorities. But commanding they were not. To illustrate by a
somewhat farfetched parallel—in direct contrast to the physicians of
the Magian stamp who merely proclaimed the virtues of their
mysterious arcana, the medical men of the West seek to obtain for
their knowledge the force of civil law, as for instance in the matter of
vaccination or the inspection of pork for trichina.
429. For the Buddhist Four Truths see Ency. Brit., XI ed., Vol. IV, p.
742. English translation of Kant’s Kritik der praktischen Vernunft by
T. K. Abbott.—Tr.
430. See p. 201.
431. See p. 205 and 222 et seq.
432. See Vol. II, p. 334.
433. The philosophy and dogma of charity and almsgiving—a
subject that English research seems generally to have ignored—is
dealt with at length in Dr. C. S. Loch’s article Charity and Charities,
Ency. Brit., XI ed.—Tr.
434. Not only as local sovereigns enforcing order, like the good
Bishop Wazo of Liége who fought down his castled robber-barons
one by one in the middle of the 11th Century, but even as high
commanders for the Emperor in distant Italy. The battle of Tusculum
in 1167 was won by the Archbishops of Köln and Mainz. English
history, too, contains the figures of warlike prelates—not only leaders
of national movements like Stephen Langton but strong-handed
administrators and fighters. The great Scots invasion of 1346 was
met and defeated by the Archbishop of York. The Bishops of Durham
were for centuries “palatines”; we find one of them serving on pay in
the King’s army in France, 1348. The line of these warlike Bishops in
our history extends from Odo the brother of William the Conqueror to
Scrope, archbishop and rebel in Henry IV’s time.—Tr.
435. A paraphrase of the opening of “John Tanner’s Revolutionist’s
Handbook,” Ch. V.—Tr.
436. See Vol. II, pp. 116 et seq.
437. Rousseau’s Contrat Social is paralleled by exactly equivalent
productions of Aristotle’s time.
438. The first on the atheistical system of Sankhya, the second
(through Socrates) on the Sophists, the third on English sensualism.
439. See Vol. II, pp. 441 et seq.
 440. It was many centuries later that the Buddhist ethic of life gave
rise to a religion for simple peasantry, and it was only enabled to do
so by reaching back to the long-stiffened theology of Brahmanism
and, further back still, to very ancient popular cults. See Vol. II, pp.
378, 285.
441. The articles Buddha and Buddhism in the Ency. Brit., XI ed.,
by T. W. Rhys Davids, may be studied in this connexion.—Tr.
442. See “The Questions of King Milinda,” ed. Rhys Davids.—Tr.
443. Of course, each Culture naturally has its own kind of
materialism, conditioned in every detail by its general world-feeling.
444. To begin with, it would be necessary to specify what
Christianity was being compared with it—that of the Fathers or that
of the Crusades. For these are two different religions in the same
clothing of dogma and cult. The same want of psychological flair is
evident in the parallel that is so fashionable to-day between
Socialism and early Christianity.
445. The term must not be confused with anti-religious.
446. Note the striking similarity of many Roman portrait-busts to
the matter-of-fact modern heads of the American style, and also
(though this is not so distinct) to many of the portrait-heads of the
Egyptian New Empire.
447. See Vol. II, pp. 122 et seq.
448. The original is here very obscure; it reads: “... es ist der
‘Gebildete,’ jener Anhänger eines Kultus des geistigen Mittelmasses
und der Offentlichkeit als Kultstätte.”—Tr.
449. See P. Wendland, Die hellenist.-röm. Kultur (1912), pp. 75 et
seq.
450. See Vol. II, pp. 318 et seq.
451. See Vol. II, pp. 269 et seq.
452. Compare my Preussentum und Sozialismus, pp. 22 et seq.
453. See Vol. II, pp. 324 et seq., 368 et seq.
454. See Vol. II, p. 345. It is possible that the peculiar style of
Heraclitus, who came of a priestly family of the temple of Ephesus, is
an example of the form in which the old Orphic wisdom was orally
transmitted.
455. See Vol. II, p. 307.
 456. Here we are considering only the scholastic side. The mystic
side, from which Pythagoras and Leibniz were not very far, reached
its culminations in Plato and Goethe, and in our own case it has
been extended beyond Goethe by the Romantics, Hegel and
Nietzsche, whereas Scholasticism exhausted itself with Kant—and
Aristotle—and degenerated thereafter into a routine-profession.
457. Zeno the Stoic, not to be confused with Zeno of Elea, whose
mathematical fineness has already been alluded to.—Tr.
458. Neue Paralipomena, § 656.
459. Even the modern idea that unconscious and impulsive acts of
life are completely efficient, while intellect can only bungle, is to be
found in Schopenhauer (Vol. II, cap. 30).
460. In the chapter “Zur Metaphysik der Geschlechtsliebe” (II, 44)
the idea of natural selection for the preservation of the genus is
anticipated in full.
461. See Vol. II, pp. 36 et seq.
462. This began to appear in 1867. But the preliminary work Zur
Kritik der politischen Ökonomie came out in the same year as
Darwin’s masterpiece.
463. Vol. II, p. 625. See, for example, Leonard, Relativitäts-Prinzip,
Aether, Gravitation (1920), pp. 20 et seq.
464. See Vol. II, pp. 369 et seq., 624 et seq.
465. See p. 57.
466. E.g., in Boltzmann’s formulation of the Second Law of
Thermodynamics: “the logarithm of the probability of a state is
proportional to the entropy of that state.” Every word in this contains
an entire scientific concept, capable only of being sensed and not
described.
467. See Vol. II, p. 369.
468. See Vol. II, pp. 382 et seq.
469. E. Wiedermann, Die Naturwissensch. bei den Arabern
(1890). F. Struntz, Gesch. d. Naturwissensch. im Mittelalter (1910),
p. 58.
470. An order of encyclopædists and philosophers; see Ency. Brit.,
XI ed., Vol. II, p. 278a.—Tr.
471. M. P. E. Berthelot, Die Chemie im Altertum u. Mittelalter
(1909), pp. 64 et seq. (The reference is evidently to a German
version; Berthelot published several works on the subject, viz., Les
origines de l’Alchémie [1885]; Introduction à l’étude de la chimie des
anciens et du moyen âge [1889]; Collection des anciens alchimistes
grecs [1887, translations of texts]; La chimie au moyen âge [1893].—
Tr.
472. For the metals, “mercury” is the principle of substantial
character (lustre, tensility, fusibility), “sulphur” that of the attributive
generation (e.g., combustion, transmutation). See Struntz, Gesch. d.
Naturwissensch. im Mittelalter (1910), pp. 73 et seq.
(It seems desirable to supplement this a little for the non-technical
reader, by stating, however roughly and generally, the principle and
process of transmutation as the alchemist saw them. All metals
consist of mercury and sulphur. Remove “materiality” from common
mercury (or from the mercury-content of the metal under treatment)
by depriving it (or the metal) of “earthness,” “liquidness” and
“airiness” (i.e., volatility) and we have a prime, substantial (though
not material) and stable thing. Similarly, remove materiality from
sulphur (or the sulphur-content of the metal treated) and it becomes
an elixir, efficient for generating attributes. Then, the prime matter
and the elixir react upon one another so that the product on
reassuming materiality is a different metal, or rather a “metallicity”
endowed with different characters and attributes. The production of
one metal from another thus depends merely on the modalities of
working processes.—Tr.)
473. See Vol. II, pp. 370, 627.
474. See Vol. II, pp. 314 et seq.
475. See the article under this heading, and also that under
Alchemy, Ency. Brit., XI ed.—Tr.
476. During the Gothic age, in spite of the Spanish Dominican
Arnold of Villanova (d. 1311), chemistry had had no sort of creative
importance in comparison with the mathematical-physical research
of that age.
477. For even Helmholtz had sought to account for the
phenomena of electrolysis by the assumption of an atomic structure
of electricity.
478. Which in their physical aspect are individual centres of force,
without parts or extension or figure. (For their metaphysical aspect,
see Ency. Brit., XI edition. Article Leibniz, especially pp. 387-8.—Tr.)
479. M. Born, Aufbau der Materie (1920), p. 27.
(So many books and papers—strict, semi-popular and frankly
popular—have been published in the last few years that references
may seem superfluous, the more so as the formulation of this central
theory of present-day physics. The article Matter by Rutherford in the
Ency. Brit., XIIth edition (1922), and Bertrand Russell, The A.B.C. of
Atoms, are perhaps the clearest elementary accounts that are
possible, having regard to the scientist’s necessary reservations of
judgment.—Tr.
480. See p. 231.
481. See p. 172.
482. See p. 121 and Vol. II, pp. 11 et seq.
483. See p. 169.
484. See p. 166 and Vol. II, p. 18.
485. See p. 152.
486. See p. 116 et seq., pp. 151 et seq.
487. See Vol. II, pp. 369 et seq.
488. J. Goldziher, Die islam. und jüd. Philosophie (“Kultur der
Gegenwart,” I, V, 1913), pp. 306 et seq.
489. See Vol. II, pp. 27 et seq., 427 et seq.
490. And it may be asserted that the downright faith that Haeckel,
for example, pins to the names atom, matter, energy, is not
essentially different from the fetishism of Neanderthal Man.
491. See p. 126.
492. Compare Vol. II, pp. 38 et seq.
493. See Vol. II, p. 305.
494. See Vol. II, pp. 343 et seq., and p. 346.
495. E. Mogk, Germ. Mythol., Grundr. d. Germ. Philos., III (1900),
p. 340.
496. See Vol. II, p. 241 et seq., 306 et seq.
497. See p. 268.
498. The pantheistic idea of Pan, familiar in European poetry, is a
conception of later Classical ages, acquired in principle from Egypt.
—Tr.
499. Few passages in the Acts of the Apostles have obtained a
stronger hold on our imagination than Paul’s meeting with the altar of
“the Unknown God” at Phalerum (Acts XVII, 23). And yet we have
perfectly definite evidence, later than Paul’s time, of the plurality of
the gods to whom this altar was dedicated. Pausanias in his guide-
book (I, 24) says: “here there are ... altars of the gods styled
Unknowns, of heroes, etc.” (βωμοί δε θεῶν τε ὀνομαζομένων
Ἀγνώστων καὶ ἡρῴων ... κ.τ.λ.). Such, however, is the force of our
fixed idea that even Sir J. G. Frazer, in his “Pausanias and Other
Studies,” speaks of “The Altar to the Unknown God which St. Paul,
and Pausanias after him, saw.” More, he follows this up with a
description of a dialogue “attributed to Lucian” (2nd Cent. A.D.) in
which the Unknown God of Athens figures in a Christian discussion;
but this dialogue (the Philopatris) is almost universally regarded as a
much later work, dating at earliest from Julian’s time (mid-4th Cent.)
and probably from that of Nicephorus Phocas (10th Cent.).—Tr.
500. Wissowa, Religion und Kultus der Römer (1912), p. 38.
501. See Ency. Brit., XI ed., article Great Mother of the Gods.—Tr.
502. In Egypt Ptolemy Philadelphus was the first to introduce a
ruler-cult. The reverence that had been paid to the Pharaohs was of
quite other significance.
503. See Vol. II, pp. 241 et seq.
504. Significantly enough, the formula of the oath sworn by this
stone was not “per Jovis lapidem” but “per Jovem lapidem.”—Tr.
505. The Erechtheum, similarly, was a group of cult-sites, each
refraining from interference with the others.—Tr.
506. Juppiter Dolichenus was a local deity of Doliche in
Commagene, whose worship was spread over all parts of the Empire
by soldiers recruited from that region; the tablet dedicated to him
which is in the British Museum was found, for example, near
Frankfurt-on-Main.
Sol Invictus is the Roman official form of Mithras. Troop-
movements and trade spread his worship, like that of Juppiter
Dolichenus, over the Empire.—Tr.
507. To whom the inhabitants of “Roman” Carthage managed to
attach even Dido.—Tr.
508. Wissowa, Kult. und. Relig. d. Römer (1912), pp. 98 et seq.
509. Wissowa, Relig. u. Kult. der Römer (1912), p. 355.
 510. The symbolic importance of the Title, and its relation to the
concept and idea of the Person, cannot here be dealt with. It must
suffice to draw attention to the fact that the Classical is the only
Culture in which the Title is unknown. It would have been in
contradiction with the strictly somatic character of their names. Apart
from personal and family names, only the technical names of offices
actually exercised were in use. “Augustus” became at once a
personal name, “Cæsar” very soon a designation of office. The
advance of the Magian feeling can be seen in the way in which
courtesy-expressions of the Late-Roman bureaucracy, like “Vir
clarissimus,” became permanent titles of honour which could be
conferred and cancelled. In just the same way, the names of old and
foreign deities became titles of the recognized Godhead; e.g.,
Saviour and Healer (Asklepios) and Good Shepherd (Orpheus) are
titles of Christ. In the Classical, on the contrary, we find the
secondary names of Roman deities evolving into independent and
separate gods.
511. Diagoras, who was condemned to death by the Athenians for
his “godless” writings, left behind him deeply pious dithyrambs.
Read, too, Hebbel’s diaries and his letters to Elise. He “did not
believe in God,” but he prayed.
512. See Vol. II, p. 376.
513. See Vol. II, p. 244.
514. Livy XL, 29.—Tr.
515. In the famous conclusion of his “Optics” (1706) which made a
powerful impression and became the starting-point of quite new
enunciations of theological problems, Newton limits the domain of
mechanical causes as against the Divine First Cause, whose
perception-organ is necessarily infinite space itself.
516. As has been shown already, the dynamic structure of our
thought was manifested first of all when Western languages changed
“feci” to “ego habeo factum,” and thereafter we have increasingly
emphasized the dynamic in the phrases with which we fix our
phenomena. We say, for instance, that industry “finds outlets for
itself” and that Rationalism “has come into power.” No Classical
language allows of such expressions. No Greek would have spoken
of Stoicism, but only of the Stoics. There is an essential difference,
too, between the imagery of Classical and that of Western poetry in
this respect.
517. The law of the equivalence of heat and work.—Tr.
518. See p. 307.
519. Original: “Keine dem abendländischen Geist natürliche Art
der Deutung mechanischer Tatsachen, welche die Begriffe Gestalt
und Substanz (allenfalls Raum und Masse) statt Raum, Zeit, Masse,
und Kraft zugrunde liegt.”
520. See foot-note, p. 314.—Tr.
521. See p. 355.
522. See Vol. II, p. 618.
523. See M. Planck, Entstehung und bisherige Entwicklung der
Quantentheorie (1920), pp. 17-25.
524. Which in many cases have led to the supposition that the
“actual existence” of atoms has now at last been proved—a singular
throw-back to the materialism of the preceding generation.
525. This sentence follows the original word for word and phrase
for phrase. Its significance depends wholly on the precise meaning
to be attached to such words as “dead,” “free,” “latent,” and to
attempt any sharper formulation of the processes in English would
require not only the definition of these (or other) basic terms but also
extended description of what they imply.
The Second Law of Thermodynamics is something which is
absorbed by, rather than specified for, the student. Elsewhere in this
English edition, indications have been frequently given to enable the
ordinary student to follow up matters referred to more allusively in
the text. But in this difficult domain such minor aids would be
worthless. All that is possible is to recommend such students to
make a very careful study of some plain statement of the subject like
Professor Soddy’s “Matter and Energy” (especially chapters 4 and 5)
and to follow this up—to the extent that his mathematical knowledge
permits—in the articles Energy, Energetics and Thermodynamics in
the Ency. Brit., XI ed.—Tr.
526. See foot-note, p. 157.
527. The application of the idea of “lifetime” to elements has in fact
produced the conception of “half-transformation times” [such as 3.85
days for Radium Emanation.—Tr.].
 528. The text of this paragraph has been slightly condensed, as in
such a field as this of philosophical mathematics partial indications
would serve no useful purpose. The mathematical reader may refer
to the articles Function, Number, and Groups in the Ency. Brit., XI
ed.—Tr.
 INDEX

Prepared by David Μ. Matteson

Aachen Minster, and style, 200

Abaca, Evaristo F. dall’, sonatas, 283
Abel, Niels H., mathematic problem, 85
Absolutism, contemporary periods, table iii
Abydos, 58n.;
contemporaries, table ii
Abyssinia, cult-buildings, 209
Academy, contemporaries, table i
Acanthus motive, history, 215
Acheloüs, as god, 403
Achilles, archetype, 203, 402
Acre, battle, 150
Acropolis, contemporaries, table ii. See also Parthenon
Act, and portrait, 262, 266, 270
Action, in Western morale, 342
Actium, battle, 381
Activity, as Western trait, 315, 320;
as quality of Socialism, 362-364
Actuality, as test of philosophy, 41;
significance, 164
Adam de la Hale. See La Hale
Addison, Joseph, type, 254
Adolescence, initiation-rites as symbol, 174n.
Adrastos, cult, 33n.
Ægina temple, sculpture, 226, 244
Æschines, portrait statue, 270
Æschylus, tragic form and method, 129, 320, 321;
and architecture, 206;
and motherhood, 268;
and deity, 313;
morale, 355
Æsthetics, and genius in art, 128
Æther, contradictory theories, 418
Agamemnon, contemporaries, table iii
Aggregates, theory, 426
Aglaure, cult, 406
Ahmes, arithmetic, 58
Ahriman, Persian Devil, 312
Aim, and direction, 361;
nebulousness, 363
Aksakov, Sergei, and Europe, 16n.
Albani, Francesco, linear perspective, 240;
colour, 246
Albani villa, garden, 240
Albert of Saxony, Occamist, 381
Alberti, Leone B., gardening, 240
Alcamenes, contemporary mathematic, 78;
period, 284
Alchemy, as symbol, 248;
as Arabian physics, 382, 383;
process of transmutation, 382n.;
and substance, 383;
and mechanical necessity, 393
Alcibiades, and Napoleon, 4;
and Classical morale, 351;
condemnation, 411
Alcman, music, 223
Alembert, Jean B. le R. d’, mathematic, 66, 78;
and time, 126;
mechanics and deism, 412
Alexander the Great, analogies, 4;
and Dionysus legend, 8;
romantic, 38;
and economic organization, 138;
expedition as episode, 147;
himself as epoch, 149;
as conqueror, 336;
morale, 349;
as paradox, 363;
deification, 405;
contemporaries, table iii
Alexander I of Russia, and Napoleon, 150
Alexandria, as a cultural left-over, 33, 73n., 79;
contemporaries, 112;
collections of University, 136n.;
as irreligious, 358
Alfarabi, and extension, 178;
and dualism, 306;
contemporaries, table i
Algebra, defined, significance of letter-notation, 71;
Diophantus and Arabian Culture, 71-73;
Western liberation, 86;
 contemporaries, table i.
See also Mathematics
Algiers, origin of French war, 144n.
Alhambra, courtyard, 235
Alien, and “proper”, 53
Alkabi, and extension, 178
Alkarchi, contemporaries, table i
Al-Khwarizmi, mathematic, 72;
contemporaries, table i
Alkindi, and dualism, 307;
contemporaries, table i
Allegory, motive and word, 219n.
Almighty, philosophical attitude toward, 123. See also Religion
Alphabet, and historical consciousness, 12n. See also Language
Alsidzshi, mathematic, 72
Altar of the Unknown God, Paul’s error, 404
Amarna art, contemporaries, table ii
Ambrosian chants, and Jewish psalmody, 228
Amenemhet III, pyramid, 13;
portrait, 108, 262
Amida, and Arabian art, 209
Analogies, superficial and real historical, 4, 6, 27, 38, 39;
necessity of technique, 5
Analysis, and Classical mathematic, 69;
in Western mathematic, 74, 75;
inadequacy as term, 81;
and earlier mathematics, 84;
contemporaries, table i.
See also Mathematics
Anamnesis, and comprehension of depth, 174
Ananke, and Tyche, 146
Anarchism, basis, 367, 373
Anatomy, in Classical and Western art, 264;
Michelangelo and Leonardo, 277
Anaxagoras, and ego, 311;
on atoms, 386;
and mechanical necessity, 392, 394;
condemnation, 411
Anaximander, and chaos, 64;
popularity, 327
Ancestral worship, cultural basis, 134, 135n.
Ancient History, as term, 16
Anecdote, and Classical tragedy, 318;
Western, 318n.
Angelico, Fra, and the antique, 275
Anthesteria, 135n.
Antigone, and Kriemhild, 268
Antiphons, and Jewish psalmody, 228
Antisthenes, character of Nihilism, 357;
and diet, 361
Antonello da Messina, Dutch influence, 236
Apelles, contemporaries, table ii
Aphrodisias Temple in Caria, as pseudomorphic, 210
Aphrodite, as goddess, 268;
in Classical art, 268
Apocalypses, and world-history, 18n.;
contemporaries, table i
Apollinian soul, explained, 183. See also Classical Culture
Apollo Didymæus Temple, form-type, 204
Apollo of Tenea, contemporaries, table ii
Apollodorus of Athens, unpopularity, 35;
painting, 283, 325n.
Apollodorus of Damascus, Roman architecture, 211
Apollonius Pergæus, and infinity, 69;
mathematic, 90
Appius Claudius, contemporaries, table iii
Arabesque, algebraic analogy, 72;
period, 108;
spun surface, 196;
character, 203, 212;
as symbol, 215, 248;
end-art, 223;
contemporaries, table ii
Arabian Culture, and polar idea of history, 18;
mathematic, significance of algebra, 63, 71-73;
expressions, 72;
and Late-Classical, 73, 209, 212, 214;
and Marycult, 137;
prime symbol, cavern, 174, 209, 215;
soul and dualism, 183, 305-307, 363;
“inside” architectural expression, 184, 199, 200, 224;
religious expression, 187, 188, 312, 401;
and Russian art, 201;
autumn of style, 207;
art as single phenomenon, 207-209;
art research, 209;
dome space-symbolism, 210-212;
ornamentation, 212;
fetters, 212;
emancipation, hurry, 213;
and mosaic, 214;
arch-column, 214;
 Acanthus motive, 215;
and portraiture, 223, 262;
architecture in Italy, 235;
music, 228;
and Renaissance, 235;
gold as symbol, 247;
political concept, 335;
will-lessness, 309, 311;
art and spectator, 329;
and world-history, 363;
nature idea, chemistry, 382-384, 393;
religion in Late-Classical, 407;
spiritual epochs, table i;
art epochs, table ii
Arabian Nights, as symbol, 248
Arbela, battle, 151
Arcadians, provided history, 11
Arch, and column, 214, 236
Archæology, and historical repetition, 4;
cultural attitude, 14, 132, 254;
significance, 134.
Archery, Eastern and Western, 333n.
Archimedes, style, 59;
and infinity, 69;
mathematical limitation, 84, 90;
contemporaries, 112, 386;
and metaphysics, 366;
and motion, 377;
as creator, 425
Architecture, ahistoric symbolism of Classical, 9, 12n.;
symbolism of Egyptian, 69, 189, 202;
transition to and from Arabian, 72, 73;
Rococo as music, 87, 231, 285;
as early art of a Culture, mother-art, 128, 224;