This book regards the conceived language machine as the latest development of language media system tools, and language machines can also be analyzed and compared in the history of computing machines.

When Turing conceived the Turing machine, he began by observing the process of people’s calculation on paper with pens. The Turing machine device conceived was used to simulate the behavior of human beings in the calculation process. Turing machine models calculation problems by establishing rules about symbols, internal states, and how input and state determine output. Simply put, given the input, Turing machine can output regularly. The idea of Turing machine is more from the machine, which is reflected in the fact that the atomic operation of Turing machine is based on a character: reading a character, making judgment and processing, and the output is only a character or no output. This is unnatural to understand from a human perspective. The object of concern in manual calculation operation is various expressions, which may be a single symbol only under certain circumstances. In this book, we regard calculation as equivalence in expression-substitution as a calculation operation. In an abstract sense, Turing machine’s output from input to rule can also be regarded as a replacement operation. Only the substitution of a single symbol is not a human computing behavior pattern. The effect is the same problem, usually Turing machine needs to establish more rules.

As mentioned earlier, Turing machine is equivalent to Chomsky’s type 2 grammar. The basis of Chomsky’s grammar theory, such as production rules, is also a substitution operation in nature. Chomsky’s grammar theory forms the grammatical basis of modern programming languages. A few months before Turing submitted his paper, the American mathematician alonzo church (June, 1903–August, 1995) submitted his paper "Explanation on Deterministic Problems", which also proved that there was no solution to the deterministic problems. Qiu Qi’s argument is based on Lambda calculus. Qiu Qi’s Lambda calculus is equivalent to Turing machine’s computing power. By studying the theory of Lambda calculus, we can find that the operation of calculus is obviously embodied in recursive symbol replacement. Through more investigation of the history of computing machines, other computing machine models based on similar replacement operation designs can be found. Turing machine thought does not explain where the rules from input to output come from, which is regarded as open. Church’s Lambda calculus theory, Chomsky grammar theory and other theories and models are also considered from a specific technical point of view. In this book, calculation is not studied as an independent topic, but in the context of the use of symbols. The replacement rules on which calculation is based can be classified as equivalence relations in knowledge or facts, including existing equivalence relations and potentially derivable equivalence relations. The background on which this book is based enables machines to be used as tools for language media systems.

Modern computer is an engineering realization of turing machine, and turing machine is not only realized by modern computer. The processing of modern computers is ultimately decomposed into arithmetic and logical calculations. Turing machine theory itself has no such requirement, which is only determined by the technical route. The consistency of binary system and its calculation in symbol and physics has brought modern computers. In the concept of language machine, the process based on substitution-transformation can have special advantages in terms of the efficiency of calculation execution. Because it can operate directly on any level of expression, it reduces the process of decomposition back to atomic operation. A substitution-transformation operation in expression corresponds to many times of atomic arithmetic, logical calculation operations and corresponding internal communication of modern computers, and theoretically, the consumption of space and time will be smaller.

Using the present computer, it is possible to simulate the ability to perform substitution-transformation operations at the expression level, or to package a programming language based on substitution-transformation operations. For the goal of language machine, these schemes have only limited verification significance. First of all, this brings some kind of cycle: virtual replacement-conversion operation based on logical and arithmetic calculation, and logical and mathematical calculation based on replacement-conversion operation. Conceptually, this book thinks that substitution-transformation is a more basic operation than logical and arithmetic calculation. Secondly, it sacrifices the possible efficiency advantages of language machines. The conception, design and implementation of language machine may draw lessons from some current computer technologies, and the overall use mode is difficult to be compared with the use of von Neumann machine. It can be predicted that the language machine will be realized, and its ability will not exceed that of a universal Turing machine, which does not constitute a key evaluation of the language machine. What the language machine pursues is the effects that can be brought by the new man-machine collaboration, which are first of all for people.

If the language machine is finally realizable, will it be dedicated or can it be used as a new universal machine? It is too early to think about such a problem now. It can be seen that the discussion in this book is basically limited to the scope of theoretical knowledge. By comparison, the application of computers is mainly in the fields of technology and engineering. In the problem solving in these fields, many problems are not typical calculation problems; Empirical rules are equally important for solving all kinds of problems, and they are not all deducible from theory. This book is to understand the language machine from the perspective of language tools, and to support people’s use of symbols is always the first. On the other hand, once the machine shows some potential, people will try to maximize the potential of the machine, which is easier to carry out and may be counterproductive to the goal of using human-oriented symbols by the machine.

In application, the depth of language mechanism and content of language machine is related to the problems considered by artificial intelligence at present, so the two can also be compared. Fundamentally, their ideological starting points are different. The primary goal of language machine is to replace tools such as paper, pen, ink and computer to support the construction and use of human language and realize more effective use of human language with the help of tools. Contemporary artificial intelligence generally uses machines to simulate the intellectual behavior of the human brain, so that more jobs that were originally done by people can be replaced by machines. This book regards human intelligence as developing or expanding with the use of symbols, rather than a roughly fixed model that can be imitated and replaced by machines.

While better cooperating with the use of symbols by human beings, the knowledge possessed by human beings and machines can be more and more similar, and the knowledge possessed by machines can exceed the capacity of any human individual; In the past, more and more symbolic regular operations performed by people can be performed by machines, and they are more efficient; At this point, it is no problem to say that the behavior shown by the machine is intelligent. Let’s look at a concrete example, the relationship between category and instance. A is a category, and X is an example under the category. We say that X belongs to A, and the word "belonging" can be understood by interpretation. It is doubtful whether the understanding formed by this explanation is really effective, and it is impossible for us to explain the machine like this. Now, we can realize such a mechanism: X is an example under category A, which contains any proposition of A. You can replace A with X, and the proposition still holds. This substitutable relationship can be actually implemented by the machine in a specific context. From the operational sense, the language machine shows that it has understood that "X is an example under category A". Machines don’t understand whether symbolic units have meaning or not, just like people do. On the other hand, an important sign of understanding is to know how to operate symbols.

A popular view is that modern computers are good at dealing with structured and regular knowledge, but they are not good at dealing with unstructured and open knowledge, such as common sense. With the development of machines in the latter direction, we can understand and imitate human intelligence more. First of all, this book does not think that a perfect understanding of structured rule knowledge has been established, and this theory has been disintegrated in machine application. Modern computers began with computing as an independent theme. To this end, this book has established its own theory, the core point is: knowledge is expressed based on equivalence relation, and calculation is based on equivalence relation. On this basis, we discuss how to build knowledge in the machine, thus making the machine have deduction ability. What is important is the structured and regular knowledge, which corresponds to the expansion of human intelligence and cognition by symbols. From the perspective of the division of labor between machines and humans, it is a reasonable way to improve or innovate tools to enhance this expansion.

The emphasis on common sense makes natural language a core theme of artificial intelligence. Through the establishment of a complex language model and statistical calculation on a large number of corpus, the machine also shows a certain language processing ability. From the economic point of view, these advances can be of great value. Most white-collar jobs are fixed jobs and do not need real creativity. It is possible that these processes are close to some parts of human intelligence, but this will not be the core part. The deductive ability of a language machine, ideally, can calculate all possible conclusions from the currently known theories and facts. This ability is the part of human intelligence expanded by symbols. This book only re-understands and realizes this part of intelligence as a more general way.

When the system we build shows that we can name, define, form new cognition and express, and ask questions, we can probably say that it is close to the core of human intelligence. This may require more systematic research, and a continuous understanding from perceptual experience to symbol system and from natural language to abstract symbol system can be established before we can find out whether there are some forms of intelligence that always depend on it. This book is more concerned about whether we can develop new forms and ways of using symbols, so that human beings can use symbols more efficiently.

## 关于作者