Automatic word processing method and computer system for implementing same

Abstract

The invention relates to a method and a system for automatically processing words in syllabic natural languages written in the binary code alphabet. This method is characterised in that it consists in transforming each series of alphabetical characters, each identified by a UNICODE character code and together forming a word, into a specific generic intermediate code by transformation, by the stack processor, of a linear structure of the word into a hierarchical structure of an abstract tree, whose nodes represent the concatenation operation and the fields instantiated by the generic types, and in then using the resulting specific intermediate code to perform a binary differential interpretation between each of the words to be processed, and all of the words of the language concerned, this interpretation being carried out by implementing a quantum measurement.

Description

Description Title of the invention: Automatic text processing method and computer system for its implementation

[0001] The present invention relates to the field of data processing in the form of text (written or spoken) and concerns a method and system for the automatic processing of texts in natural languages. The invention is based on a symbolic approach to characters using a CPU (central processing unit), which is fundamentally different from the currently dominant approach of Large Language Models (LLMs), which uses pixel matrices to represent characters with the implementation of graphics processing units (GPUs).

[0002] The processing of large amounts of data ("big data") in natural languages ​​is confronted with the explosion of combinatorial character formats associated, on the one hand, with a unique UNICODE identifier, and on the other hand, with the arbitrary standards for their encoding / decoding in different proprietary or open operating systems (OS). This situation, which raises the problem of interoperability standards in communication between these OSs as target machines for a natural language source file, implies a definition of the character from both a human and a machine perspective. However, the character is considered as atomic data by high-level languages, which do not access the bits that constitute the binary representation of characters.

[0003] The technical problem of interoperability, which is that of the translation / conversion of character encoding standards between the lowest level of machine language, the level of high-level languages ​​and the highest level of natural languages, is therefore that of the human-machine interface between, on the one hand, the peripheral equipment, consisting either of a keyboard (for written language), or of a microphone associated with a means of automatic speech recognition, and, on the other hand, the pair formed by a CPU and a central memory, in an automatic text processing system in natural languages.

[0004] Currently, the source language is generally translated directly into the target language, with dynamic typing for on-the-fly compilation (J IT compilation) and 'cross-platform' technology (e.g., PCSOFT) depending on the OS.

[0005] We start from the implementation of a symbol table (TSB) for bit sequences - memory words - in the organization of the main memory space and its connection with the bit registers in the processor.

[0006] In natural language processing, as in the lexical analysis phase of a compiler, the source program is read character by character, which can be very time-consuming. Lexical analysis involves incorporating a symbol table (TSB) and functions for managing this table, geared towards searching for and storing lexemes.

[0007] When the lexical analyzer reads a letter, it first saves the letters and numbers in a TampLex buffer.

[0008] Integer AnalLex function;

[0009] Var TampLex array [0...100] of characters;

[0010] C: character;

[0011] The string collected in TampLex is then searched in the symbol table using the Search operation. Since the symbol table was initialized with keywords, the search operation will find these entries if TampLex contains the lexemes of those keywords. If there is no entry for the string contained in TampLex—that is, if Search returns 0—then fTampLex contains the lexeme of a new lexical unit. An entry for this unit is created using Insert. After the insertion has been performed, p is the index of the entry in the symbol table for the string contained in TampLex. This index is passed to the parser by setting ValLex to p, and the lexical unit found in the unilex field of the symbol table is returned.

[0012] Symbol table (TSB)

[0013] Table for storing chains:

[0014] Each string is terminated by an end-of-string character denoted fdc, which cannot appear in lexical units. Each entry in the TSB table, which represents the symbol table, is a record composed of three fields: a lexical pointer that points to the beginning of the lexical unit, a character string, and an attribute field to store attribute values.

[0015] The lexical analyzer of a natural language source therefore currently takes exponentially longer processing time. The calculation of search and insertion procedures in tables is implemented using dispersed addressing via hash functions. However, hash functions involve converting a character into an integer. Furthermore, Unicode character strings are not handled in the same way by the Android and iOS systems (registered trademarks). Android stores a character in 4 bytes (like Linux - registered trademark), while iOS stores a character in 2 bytes (like Windows - registered trademark). The use of the hash function, which manipulates bytes, will not yield the same result depending on the platform, which generates compatibility problems in the encoding and decoding of typographic characters used by all keyboards, particularly those on smartphones.

[0016] The present invention aims to overcome these drawbacks and to offer a versatile interoperability solution in terms of operating system, in a context of interpreting texts in syllabic natural languages ​​with binary alphabetic writing in vowels and consonants, this solution being more secure, and preferably faster and less resource-intensive, than existing solutions.

[0017] To this end, it has as its object a method for the automatic processing of texts in written or oral form in syllabic natural languages ​​with binary alphabetic writing vowels / consonants, this by lexical analysis with the implementation of a stack processor and by recognition of lexical units or words by comparison, a method characterized in that it consists of transforming each series of alphabetic characters, each identified by a UNICODE identifier and corresponding together to a word or a lexical unit, by encoding each of its characters in the form of an eight-bit string in accordance with one of the ISO 8859 standards, preferably the ISO 8859-1 or Latin-1 standard), and by using this bit string to excite in parallel, simultaneously and selectively, for example with the encoding of generic type lexical units: written or oral vowel (V) = 1 and written or oral consonant (C) = 0,the eight inputs of an 8-input, 256-output switching matrix, these operations resulting in the production of a specific generic intermediate code by transformation by the stack processor of a linear structure of the input word or text into a hierarchical structure of an abstract tree, whose nodes represent the concatenation operation and the fields instantiated by the generic types (V and C), then to use the resulting specific intermediate code to perform a binary differential interpretation, of abstract and generic type, between each of the lexical units or words to be processed, on the one hand, and the set of lexical units or words of the language concerned, on the other hand, this binary differential interpretation consisting, by means of elementary comparison circuits, of comparing in parallel between the two bits of the eight pairs of bits of the same rank or position,respectively corresponding to each of the lexical units or each of the words to be processed and the lexical units or words of the language concerned, then to mutually combine the results of the, different elementary comparisons by means of an eight-input AND function, until an identity of bits is reached for all eight pairs of bits, corresponding to a lexicographically exact identity between a character of the lexical unit to be processed or of the word to be processed considered and a character of the alphabet of the language concerned, and this differential interpretation being carried out by implementing a quantum measurement performed by processing a quantum register associated with a (classical) computer register containing said generic intermediate code produced.

[0018] The invention will be better understood from the following description, which relates to a preferred embodiment, given by way of non-limiting example, and explained with reference to the accompanying schematic drawings, in which:

[0019] [Fig. 1] is a general schematic representation of a hybrid architecture of a computer system for the implementation of the process according to the invention;

[0020] [Fig. 2] is a symbolic representation of an example of an abstract tree, here for the lexical expression or word "SEIN";

[0021] [Fig. 3] is a representation of a generic computer lexicographical tree of a "SYLLABLE EXPRESSION" and its associated table;

[0022] [Fig. 4] is a commented graphical representation of the inference rules for the English coding of "MISSION";

[0023] [Fig. 5] is a symbolic representation illustrating the addressing of TSB tables;

[0024] [Fig. 6] is a symbolic representation of a logic comparison circuit forming part of the computer system for implementing the method according to the invention, shown in figure 1;

[0025] [Fig. 7] is a symbolic representation of an elementary comparison circuit that is part of the logic comparison circuit shown in Figure 6.

[0026] Figure 1 and partially Figures 2 to 7, in relation to the various tables presented below, illustrate some aspects of the invention.

[0027] The latter concerns an automatic text processing method in written or oral form in syllabic natural languages ​​with binary alphabetic writing, by lexical analysis with implementation of a stack processor and by recognition of lexical units or words by comparison.

[0028] According to the invention, this method is characterized in that it consists of transforming each series of alphabetic characters, each identified by a UNICODE identifier and corresponding together to a word or lexical unit, by encoding each of its characters in the form of an eight-bit string in accordance with one of the ISO 8859 standards, preferably ISO 8859-1 or Latin-1), and using this bit string to excite in parallel, simultaneously and selectively, with, for example, as encoding of generic lexical units: vowel (V) written or spoken = 1 and consonant (C) written or spoken = 0, the eight inputs of an 8-input, 256-output switching matrix, these operations resulting in the production of a specific generic intermediate code by transformation by the stack processor of a linear structure of the input word or text into a hierarchical structure of an abstract tree,whose nodes represent the concatenation operation and the fields instantiated by the generic types (V and C), then to use the resulting specific intermediate code to perform a binary differential interpretation, of abstract and generic type, between each of the lexical units or words to be processed, on the one hand, and the set of lexical units or words of the language concerned, on the other hand, this interpretation being carried out by implementing a quantum measurement performed by processing a quantum register associated with a (classical) computer register containing said intermediate code produced.

[0029] In accordance with the invention, binary differential interpretation consists, by means of elementary comparison circuits (ECCs), of comparing in parallel the two bits of the eight pairs of bits of the same rank or position, respectively corresponding to each of the lexical units or words to be treated (A) and the lexical units or words (B) of the language concerned.

[0030] Also in accordance with the invention, the method then consists of mutually combining the results of the different elementary comparisons by means of an eight-input "AND" function or circuit, until an identity of bits is obtained for all eight pairs of bits, corresponding to a lexicographically exact identity between a character (A) of the lexical unit or word to be processed considered and a character (B) of the alphabet of the language concerned.

[0031] Thus, in accordance with the aforementioned provisions, the source language is translated into an intermediate language or code that is independent of the target machine and its operating system, and therefore interpretable by any type of existing target machine and operating system. This is achieved with static typing for a compilation model based on analysis and synthesis, where the front end of the compiler translates the source language into an intermediate representation. In this case, a compiler can be created for another machine by modifying only the final part of the compiler: this allows communication between processors working on bit registers and processors working on qubit registers in a hybrid architecture.

[0032] The generation of the aforementioned specific intermediate code allows, on the one hand, quantum simulation on a classical machine and, on the other hand, the implementation of a quantum machine, independently of the technology implemented by the latter.

[0033] The solution proposed by the invention implements a compilation by analysis-synthesis and the production of an intermediate code that involves strong typing of characters into generic types / abstract types / actual types of glyphs or characters, independent of the operating system (OS). Strong typing allows for data modeling through a polynomial, rather than exponential, analysis of the input character stream.

[0034] An example of a concise structure for a table of alphabetic symbols is shown below:

[0035] In accordance with the aforementioned provisions of the method according to the invention, the 8-input, 256-output switching matrix encodes glyphs or characters identified differentially in the UNICODE standard.

[0036] The set of glyphs, which are of a real type and correspond to the characters entered or provided as input, is associated with a finite set of character models of an abstract type called metacharacters. For example, French includes the set of metacharacters [A, E, I, O, U, Y] called the set of metavowels and the set of metacharacters [B, C, D, F, G, H, J, K, L, M, N, P, Q, R, S, T, V, W, X, Z] called the set of metaconsonants, which can be encoded in an indeterminate number of glyphs depending on the standards of the operating systems and the alphabetic writing conventions specific to a given language. As another example, we observe that French associates the real types of glyphs {a, â, in lowercase and uppercase} with the abstract type of the metacharacter [A], whereas German associates the glyphs {a, in lowercase and uppercase}.

[0037] The function linking the real types of glyphs and the abstract types of metacharacters then solves, at the abstract level of metacharacters, the interoperability problem of encoding / decoding typographic characters manipulated by operating systems. Indeed, by generalization, we can infer that, since all natural languages ​​are represented by symbols identified by a unique UNICODE identifier, and since all glyphs of natural languages ​​with alphabetic writing belong to a set of vowels or a set of consonants, all languages ​​with alphabetic writing are isomorphic at the abstract level of their respective metacharacters associated with the DISJOINT sets of metacharacters. Vowels and consonants from the alphabets of the languages ​​in question, interpreted as a generic type, designated by the symbol V (denoting the set of vowels) and the symbol C (denoting the set of consonants), implementable in any computer system with a BINARY alphabet {1, 0}, independently of the encoding / decoding standards of a given operating system. From this arises the binary differential interpretation which associates V with 1 and C with 0.

[0038] In accordance with ISO 8859, which encodes a character on an 8-bit string, or byte, the invention makes it possible to represent the alphabets of all Western European languages ​​at the lowest level of computer systems, that of CPU chips, using diode matrices and switching circuits that represent, in binary code, the lexical units and words of natural languages ​​written with vowels (V) and consonants (C), as well as the instructions relating to them. The method for processing written and spoken texts in natural languages ​​written with alphabets according to the invention thus links the level of binary symbols of low-level assembly languages ​​and the level of symbols of the high-level languages ​​used in the compilation by analysis and synthesis of programming languages, which are by definition artificial.The compilation phases for these two kinds of languages ​​are therefore identical, notably involving the production of an intermediate code, which represents the encoding of the source program for the lexical analysis of the input text for an abstract machine (stack processor), which, in the case of natural alphabetic languages, transforms the linear structure of the input text into the hierarchical structure of an abstract tree, whose nodes represent the concatenation operation and the fields instantiated by the generic type symbols V and C.

[0039] According to another aspect of the invention, partially illustrated by the attached figures and described in more detail below, the processing method according to the invention comprises recognition and storage of the input text to be processed, carried out in two successive operational phases 1) and 2) at the level of a computer system used for the implementation of said method, namely: -1) the explicit management of reading at the level of the central microprocessor (CPU) and the main memory of the computer system used for the implementation of the process, this through an interface which ensures the compatibility of binary signals and controls them by a logic comparison circuit, performing a hardware level interpretation of said signals as an alphabetic human character, namely an element of class "V" or an element of class "C", associated biunivocally with a machine binary character (0 or 1); -2) the construction by a stack processor at the level of an intermediate representation of the abstract tree whose root is the constructed type memory word, via the recursive binary concatenation operation, interpreted as an abstract syllable defined by the normalized configurations of "V" and "C" in any ISO 8859-1 language, instantiated by the concrete syllables realized in each language of the ISO 8859-1 family, this memory word being stored in a computer tree.

[0040] This coding system, implemented by the invention, is thus carried out at the level of the chip circuits and transistors and is therefore implemented below the operating systems (OS), independently of the coding and decoding standards (codec). This ensures hardware security through bit control and software security through strong static typing for information systems distributed across data centers.

[0041] Just as the graphics processing unit (GPU) accelerates character processing by pixels in known systems, so too can the CPU, which directly accesses bits, be accelerated in the context of the present invention by implementing a hybrid architecture of classical and quantum registers, particularly at the hardware level (silicon hybridization). Thus, encoding characters in ISO 8859-1 languages ​​using "lexicographic" comparison logic circuits enables and validates the beneficial implementation of such a hybrid "classical-quantum" physical architecture.

[0042] In accordance with an advantageous feature, the method can consist of providing a matrix in the form of an addressable memory of the pointer array type (I, T), having 8 input rows and 256 output rows output, this matrix corresponding to an address register composed of the two sub-registers of the generic types (V and C) for a given language, and associating the binary generic type (V or C), for any character of this language, with an interpretation valid for a quantum state of a Qbit or qubit, physical and logical, with a superposition of two states with equal probabilities, namely 1 / 2, the registers of the 8-bit strings being correlated to the quantum registers of an 8-dimensional vector space in a hybrid architecture of the quantum physics / classical physics type.

[0043] The 8-input, 256-output switching matrix allows for the encoding / decoding of alphabets of living languages ​​designated by UNICODE universal identifiers, as well as subsets of UNICODE, through a parallel and simultaneous representation of the alphabets of all ISO 8859 languages ​​in a specific (I, T) array interpreted as the model of the alphabet of the language concerned by the ISO 8859 standard. The integer index I [1-N] refers to the set of glyphs occupying consecutive positions in the interval occupied by written uppercase and lowercase vowels, then spoken vowels, followed by the set of glyphs occupying consecutive positions in the interval occupied by written uppercase and lowercase consonants, then spoken consonants. The bit type T [1-8] is established by a general convention of 2 8For each ISO 8859 language, which encodes these characters in 8-bit memory cells, the bytes are: that is, 256 addressable bytes on an 8-bit string, interpreted as the lexeme of the glyph associated with a metacharacter interpreted as a generic lexical unit<V, C> for a given language.

[0044] We then obtain for each language with alphabetic writing an address register composed of 2 INDEX sub-registers accessed in parallel by pointer: the V sub-register and the C sub-register. Indexed addressing will therefore take the time of the iteration traversing with an increment of 8 bits, all the elements of index i from the first element of the alphabetical table of a language until the desired element is accessed.

[0045] The generic binary type <V / C> interpreting the binary set {1, 0} at work in reading / writing alphabetic languages ​​can therefore be associated for each alphabetic character / glyph of the ISO 8859 languages ​​to an interpretation valid for a quantum state with 1 qubit at the same time physical - having as its support the spin of a particle - AND logical with a superposition of 2 states with equal probability:

[0046] probability | V > = probability | C > = 1 / 2.

[0047] Therefore, the quantum state representing an input alphabetic character of type V / C is:

[0048] |'4 J character > = 1 / V2 . |V > + 1 / V2 . | C>

[0049] There is therefore a logical equivalence between the binary computer code {1, 0} representing the generic type lexical units Vowel (V) = 1 and Consonant (C) = 0, and a quantum state associated with a qubit which superimposes these two equiprobable associated states.

[0050] Before measuring the aforementioned quantum state, the state of a qubit is associated with character sets, with glyphs corresponding to character models, and metacharacters associated with generic types (V, C), which constitute two fields of the written and spoken tables (I, T) of ISO 8859 languages. The glyphs and metacharacters are encoded according to the binary generic type.<V, C> This allows for the correlation, in a hybrid architecture with silicon hybridization combining a classical computer and a quantum computer, of the registers of 8-bit strings and the quantum registers of an 8-dimensional vector space.

[0051] This correlation therefore defines a finite-dimensional vector space.

[0052] The state | l The P associated with a quantum bit (qubit) can be described by any unit vector in the 2-dimensional vector space described by |0> and |1> over the field of complexes. The general state of a qubit is given by: where Qo and ai are any two complex numbers with the sole constraint that |M J let >, |0> and |1> be unit vectors in the vector space over the field of complexes - that is, there exists a normalization condition such that

[0053] The state | l P> is in a coherent superposition of the states |0> and |1> with amplitudes Qo and Ch. If Qo or Ch is 0 and the other is 1, we are in the special case where the qubit is in one of the two classical states |0> or |1>. Therefore, the NOT logic gate of classical integrated circuits is equivalent to the NOT transformations of quantum circuits in the definition of reversible operations, particularly relevant in error correction in finite-dimensional computing systems like the one shown here.

[0054] This characteristic can be described by any superposition of quantum states associated with 2 ndifferent classical states (here 8 Qubits), whose sum of squared amplitudes is equal to one:

[0056] 0 <x<2 n 0 <x<2 n

[0057] In the context of quantum computing, the game of 2 n classical states - all possible combinations of tensor product of n individual qubits |0> and |1> - corresponds to the computational basis or classical basis.

[0058] Since the states of the n qubits are limited to the states of the computational basis, the measurement process is identical to the classical "read value" process of x without disturbing the state. Therefore, a quantum computer can be used to simulate a reversible classical computer, limiting only its input register to states of the computational basis and using only unit gates that transform states of the classical basis into new states of the classical basis.

[0059] The hybrid architecture, as defined in the invention and schematically shown in Figure 1, therefore establishes, at the lowest level of the chips, the physical gateway between classical and quantum registers. via an objective measurement of a quantum state: just as the 8 bits, which combine the results of the different elementary comparisons, are assembled by the AND function to give an output signal interpreted as a lexical unit of type V or C, so too are the particles assembled within the atom, which is in a quantum state representing in the wave function the amplitude of the superposition of all possible measurement values ​​simultaneously associated with all the alphabetic glyphs of a given language, a wave function reduced after measurement to a single value (here 1 for V or 0 for C).

[0060] The process can advantageously provide for storing the positive output of the AND function in an array of the type (I, T) of vowels and consonants of the language in question, which thus forms a symbol table (TSB) whose indices correspond to the alphabetical order of the typographic characters, their static type as an abstract type of metacharacter and generic type of lexical unit (V or C), as well as their configuration consisting of 8 bit locations.

[0061] As schematically illustrated in Figures 2 and 3, it may be possible to construct an abstract tree (AA) as a sequence of subtree constructions from bottom to top, whose leaves point to the entries in the symbol table (TSB) and whose nodes represent the result of concatenating the values, corresponding to the characters or strings of characters, associated with the leaves which are the operands of the concatenation operator.

[0062] Advantageously, this syntactic construction of an abstract tree is governed by an alphabetic byte-register constructor algorithm, which produces, from an input consisting of a set or sequence of characters in an array or register pointing to two sub-arrays or sub-registers of vowel (V) and consonant (C), an abstract tree constructed from the bottom up by recursive concatenation. The result of this alphabetic algorithm defines an assembly language, in which the expressions resulting from the concatenation represent memory words at the nodes of the abstract syntactic tree corresponding to the registers of the automaton or the stack-driven processor, and producing the intermediate code of a generic type.

[0063] Preferably, the analysis performed by the automaton or the stack processor corresponds to the implementation of production rules of an underlying algebraic grammar generating the lexical constructions, or memory words, of a language of variable length from 1 to 8 characters, in registers of variable size ranging from 8 bits / 1 byte to 64 bits / 8 bytes, these production rules advantageously providing four string models representing four types of abstract syllabic structures composed by recursive concatenation, corresponding to four recording models (R1, R2, R3 and R4).

[0064] Thus, for input data in the form of a character stream in a TamponLex array pointing to two sub-arrays of vowels V and consonants C, the method according to the invention can include the following operations: 1) Entering the character of the 1st into the TamponLex table èrevowel and any subsequent vowels, and their loading into the vowel register, and evaluation of the attributes attached to the VowelExpression node by calling the CreateSheet and CreateNode functions, the result of which is - either the root node, - either an internal knot, - then calculate the concatenation of the 2 leaf nodes of the higher level node and evaluate the result as a root node or as an interior node. - If internal node, then iterate the concatenation. 2) Input of consonants, if any, into the TamponLex array to the right of the VowelExpression node and their loading into the consonant register, and evaluation of the attributes attached to the ConsonantExpression node by calling the CreateSheet and CreateNode functions, the result of which is - either the root node, - either an internal knot, - then calculate the concatenation of the 2 leaf nodes of the higher level node and evaluate the result as a root node or as an interior node. - If internal node, then iterate the concatenation. 3) Input of consonants, if any, into the TamponLex array to the left of the VowelExpression node and their loading into the consonant register, and evaluation of the attributes attached to the ConsonantExpression node by calling the CreateSheet and CreateNode functions, the result of which is - either the root node, - either an internal knot, - then calculate the concatenation of the 2 leaf nodes of the higher level node and evaluate the result as a root node or as an interior node. - If internal node, then iterate the concatenation.

[0065] The result of the alphabetic algorithm defines an assembly language in which the concatenation results of the expressions represent memory words at the nodes of the abstract syntax tree corresponding to the processor registers.

[0066] According to advantageous provisions of the invention, it is provided that the production of said intermediate code uses three registers for respectively the binary values ​​of the first and second input characters and of the result and provides a code at three addresses consisting of a sequence of instructions, each of which has at most three operands, and that this production of intermediate code consists of a recursive construction of abstract strings carried out by the stack processor which evaluates the postfixed representation of the concatenation: a + b = ab+, resulting recursively in an abstract language of abstract strings constructed on the alphabet of lexical units (V vs C), associated with the real types of the glyph strings of each natural syllabic language with binary vowel / consonant alphabetic writing.

[0067] Memory words define the elements of an assembly language internal to the system of alphabetic vowel-writing languages ​​of the type generic V and generic type consonants C. The alphabetic V / C algorithm produces an intermediate code of generic type considered as a program for an abstract machine, in which each memory location of the syntax tree can play the role of a register.

[0068] More specifically, this intermediate code production uses three registers according to the invention: - a register for the binary value of the first character entered in the input stream, - a register for the binary value of the second, - a register for the binary value of the result.

[0069] Since each input character glyph is associated with both a binary representation and a lexical unit (V vs C), and since each alphabetic language performs concatenation on a restricted set of characters and a restricted set of instructions (RISC), the processor produces an intermediate representation in the form of a three-address code consisting of a sequence of instructions, each with at most three operands. This sequence is viewed as a program for an abstract machine at the level of an assembly language with symbolic addresses (a), (b) for the operands and (r) for the result, and LOAD, CONCAT, COMP, and STOR for the instruction names:

[0070] (has)

[0071] (b)

[0072] (r)

[0073] LOAD a

[0074] LOAD b

[0075] CONCAT abr

[0076] COMP r memory address

[0077] STOR memory address

[0078] This recursively constructs an abstract language of abstract strings built on the alphabet of lexical units (V vs C), associated with the real types of glyph strings in each alphabetic language. The representation of a piece of data is therefore determined by its abstract type and its real type, thus defining object-oriented systems.

[0079] According to the invention, this recursive construction of the strings is carried out by an automaton or stack processor which evaluates the postfixed representation of the concatenation: a + b = ab+.

[0080] The string ab+ to be analyzed by the automaton or the stack processor is the result of the conversion which postfixes the symbolic operator + signifying concatenation.

[0081] Initially the battery is empty.

[0082] When the binary concatenation operator + is encountered, two elements, here b and then a, are popped from the stack and the + operator is applied to them. This yields a value r := a + b, which is then pushed onto the stack. The string to be analyzed is now empty, so the concatenation is complete. The value r, stored in the data register, is then stored in an array at a memory location.

[0083] Since concatenation is recursive, the automaton or stack processor analyzes the expression a + b + c as a postfix expression ab + c +.

[0084] Since `ab+` gives the previous result `r`, which was pushed onto the stack, the expression to be analyzed is `rc+`. When the binary concatenation operator `+` is encountered, two elements are popped from the stack, here `c` and then `r`, and `+` is applied to them. This yields a new value `r' := r + c`, which is then pushed onto the stack. The string to be analyzed is now empty, so the concatenation is complete. The value `r'`, stored in the data register, is then stored in an array in a memory location.

[0085] This method allows the construction of fixed-length strings and the specification of lexical units as variable names structuring the memory space. Because an alphabetic character is encoded using 1 byte, there is isomorphism between the size of a binary register in the processor and the size of the alphabetic character. and the size of a memory location in the data structure of the syntax tree.

[0086] The analysis by automaton or stack processor is therefore equivalent to the production rules of an underlying algebraic grammar generating the lexical constructions of a language of variable length from 1 to 8 characters, in registers of variable size ranging from 8 bits / 1 byte to 64 bits / 8 bytes.

[0087] These rules for producing the underlying grammar advantageously produce 4 models of strings representing 4 types of abstract syllabic structures composed by recursive concatenation: - Recursion stopped: Vowel expression = class R1 - Left recursion: ConsonantExpression + VowelExpression = class R2 - Right recursion: VowelExpression + ConsonantExpression = class R3 - Right-hand recursion and left-hand recursion = consonant expression + vowel expression + consonant expression = class R4

[0088] The number of calls to the CreateSheet and CreateNode functions therefore allows for the definition of complex structures, through inheritance and symmetrically, starting from the elementary class R1 VowelExpression - left-hand inheritance for class R2, - inheritance on the right for class R3, - inheritance on the right then on the left for class R4.

[0089] These derivations generate the lexicographical tree illustrated by figure 3.

[0090] The table associated with this tree is shown below (up to level 3 descending from the tree).

[0091] According to one embodiment of the invention, the method consists in particular of carrying out a lexicographical comparison between the lexical unit or word to be processed and the lexical units or words of the language concerned, serving as models, by comparing in parallel the pairs of bytes of the same rank or position of the two lexical units or words, by implementing an adequate number of comparators, and storing the result if each comparator delivers a signal confirming the identity.

[0092] Advantageously, memory words, from 1 to 8 bytes, are processed by a lexicographic comparator of pairs of bytes of the same rank or position, for example a logic comparison circuit as illustrated in Figure 6, the process consisting of mutually combining the results of the different lexicographic comparisons by means of an AND function with 1 to 8 inputs, until an identity of memory word is obtained for the pairs of bytes, corresponding to an exact lexicographic identity between the lexical unit or word to be processed considered and a memory word of the language concerned.

[0093] Examples of memory / register words are shown below, up to four characters, with the corresponding table of possibilities:

[0094] Length 1, 1 comparator, size 2 1

[0095] Length 2, 2 comparators, size 2 2

[0096] Length 3, 3 comparators, size 2 3

[0097] Length 4, 4 comparators, size 2 4

[0098] Length 5, 5 comparators, size 2 5

[0099] Length 6, 6 comparators, size 2 6

[0100] Length 7, 7 comparators, size 2 7

[0101] Length 8, 8 comparators, size 2 8

[0102] The storage of the word or lexical unit corresponding to the positive result of the lexicographical comparison consists preferably, depending on the number N of places taken by the configuration of the word or lexical unit forming said result, of carrying out a conditional branch with the value of N in the interval [1 - 8] as input and the 8 pointer arrays (I, T) as parallel output for each configuration model of 1 to 8 places pointing to the memory locations associated with the physical and logical records of the words or lexical units of the language concerned.

[0103] These eight pointer arrays (I, T) form a reference basis for the arrays associated with each of the eight memory blocks in the abstract machine. They specify the structure of the input data by allowing or disallowing the concatenation of byte strings; the asterisk (*) character signifies a concatenation error. The unit of The processor controls the circuits that enable or disable the concatenation operation. It also controls the program that derives one structure into another through inheritance, producing two sets of structures: the set of well-formed expressions (BFE) versus the set of erroneous expressions for a given language. Therefore, the concatenation operation performed by the automaton or stack-driven processor constructing the abstract tree (AA) functions as a main program with the 8 Base registers and subprograms that execute using the operands of each alphabet in the format Operand = [[X] + Address].

[0104] The hierarchical organization of memory therefore allows, via parameter passing by reference, the concatenation operation with an indexed addressing system.

[0105] Operand = [[Base] + [X] + Address].

[0106] The conditional branch defining the [Base] is coded in a 3-bit register for the 8 possible values.

[0107] Each pointer (T) defines an addressing as a bijective function of the set of addresses, namely the indices (I), on the set of memory cells implemented in a technology adapted to the binary values ​​of the arrays (I, T) for the registers of the stack processor and the memory locations of the main memory associated with it.

[0108] The data and processing model is inherently that of multiprogramming, with link editing of a memory divided into partitions specific to each language and a master partition that supervises and distributes processor operations and memory accesses.

[0109] The memory therefore comprises a virtual memory address space specified by the Reference Base and a real memory address space, linked by a dynamic address translator, interposed between the central processing unit of the computer system used to implement the process, and the memory access circuits. This translator has a table in which the virtual memory arrays currently in real memory are referenced, along with their addresses (virtual and real).

[0110] This results in a massively parallel intrinsic logical architecture: - at the level of virtual arrays encoding the generic types V and C of ISO 8859 languages, - at the level of virtual arrays pointing to real arrays encoding the metacharacters and glyphs of these same ISO 8859 languages, - under the control of the supervisor who manages the parallel programs for lexical processing of the input stream for each language within its ISO 8859 family. [01 1 1 ] In accordance with a feature of the invention and taking into account the four recording models (R1, R2, R3 and R4) associated with the byte strings of the configuration models corresponding to the words or lexical units to be processed, the conditional branching with its eight possible outputs, which correspond to eight columns for the aforementioned four recording models, creates a single matrix for branching out of the said four recording models.

[0112] Preferably, this unique matrix points to arrays of one to eight columns forming a finite set of data structures represented in virtual memory, itself pointing to the actual addresses of configurations instantiated in real memory, in which each generic type vowel string (V) and each generic type consonant (C) is instantiated by the metacharacters and glyphs of vowels and consonants belonging to the alphabet of each language in the family of languages ​​encoded by the ISO 8859 standard.

[0113] The four registration models are as follows:

[0114] It may be envisaged to use the finite set of data structures as a filter intended to filter or discriminate between well-formed expressions and ill-formed expressions or errors.

[0115] One-place configurations: 1 well-formed expression and 1 error

[0116] Two-place configuration: 3 well-formed expressions and 1 error

[0117] Three-place configuration: 6 well-formed expressions and 2 errors

[0118] Four-place configuration: 10 well-formed expressions and 6 errors

[0119] 5-place configurations: 14 well-formed expressions and 18 errors

[0120] 6-place configurations: 17 well-formed expressions and 27 errors

[0121] 7-place configurations: 17 well-formed expressions and 111 errors

[0122] 8-place configurations: 15 well-formed expressions and 241 errors

[0123] According to one possible practical embodiment, the invention may provide a diode matrix which integrates the finite set of data structures, the conditional branching of said diodes exciting wires by connecting diodes according to the input values ​​of the wires excited or not and the planned switching according to pinout models providing differential interpretations of the controlled identifiers.

[0124] Indeed, planned switching based on pinout models results in differential interpretations of the controlled identifiers: - by constructing an elementary lexicographic comparator for each of the pairs of memory word models [R1, R2, R3, R4] whose configuration is given as input to the matrix, - if the output result is positive, i.e. is a well-formed expression (ebf), by the construction, using the AND function, of a composite lexicographic comparator with N places instantiated by the memory words [R1, R2, R3, R4].

[0125] From the basic registers associated with the memory word model configurations associated with the 8 arrays, the recursive generation of higher-dimensional identifiers is controlled at the level of the single diode matrix:

[0126] A = R1 . (R1 | R2| R3| R4)

[0127] B = R3 . (R1 |R2|R3|R4)

[0128] C = R2 . (R1 |R2|R3|R4)

[0129] D = R4 . (R1 |R2|R3|R4)

[0130] These virtual memory configurations point to the real memory of identifier arrays composed here of two memory words associated with the real type of two "syllables" in the languages ​​coded in their respective ISO 8859 standards, identifiers defined and interpreted recursively left and right by the position and value of the memory words (R1-4) in the configurations of the lexical constructions.

[0131] Identifier configurations are advantageously defined by implementing an appropriate number - specific to each language - of elementary lexicographic comparators of memory words through the AND function.

[0132] These configurations advantageously represent RDBMS-type storage models, where each column receives a differential interpretation linked to its relative position in the resulting configuration of the AND function and the number of lexicographic comparators it implements. This yields a family morphological dictionary. of identifiers interpreted as grammatical words which are the inputs of a higher-level combinatorial unit, constructed by inferences on the prefix, base, suffix fields.

[0133] This generative lexicographic model is illustrated below, as an example, for the word "SEIN", in correlation with figure 4 and figure 2, with coupling of the classical register and the quantum register.

[0134] As a further illustrative example, the inference rules for encoding the word "mission" in English are symbolically represented in Figure 4.

[0135] The associated algorithmic coding is as follows:

[0136] If the value of the attribute attached to the left child of the identifier root node is a pointer to address R4 of the computer tree with a given value, here, CVC / mis / for the ORAL CODE field and CVCC "miss" for the WRITTEN CODE field, and,

[0137] If the value of the attribute attached to the right child of the root identifier node is a pointer to address R2 of the computer tree with the given value / Sn / for the ORAL CODE field and a pointer to address R3 with the given value "ion" for the WRITTEN CODE field,

[0138] Then the value of the attribute of the ORAL CODE field attached to the root node common name identifier is R4 = accent / CVCC / = VmiSn / and, the value of the attribute of the WRITTEN CODE field attached to the root node common name identifier is R4 + R3 = CVCC + VVC = "mission".

[0139] As symbolically shown in Figure 1, and as also partially shown in Figures 6 and 7, the invention also relates to a computer system for implementing the automatic text processing method for natural syllabic languages ​​with binary alphabetic vowel / consonant writing, as described above, these texts being formed of lexical units or words (A).

[0140] This system is characterized by its hybrid architecture, which includes a central processing unit (CPU) and its traditional bit registers. (of a traditional computer) are coupled with a quantum device or computer with several, preferably eight, qubits, and in that it comprises, on the one hand, a logic comparison circuit comprising eight elementary comparison circuits (ECCs), arranged in parallel and capable of comparing with each other the two bits of the eight pairs of bits of the same rank or position, respectively corresponding to each of the lexical units or each of the words (A) to be processed and the lexical units or words (B) of the language concerned, each alphabetic character of which is identified by a UNICODE identifier, and on the other hand, an eight-input AND circuit, combining the results (S') of the different elementary comparisons performed by the eight elementary comparison circuits (ECCs). The output of this AND circuit takes the values ​​0 or 1, corresponding to V or C of a given syllabic natural language with a binary vowel / consonant alphabetic script, of the ISO 8859-1 type.

[0141] Preferably, and as shown in Figure 7, each elementary comparison circuit (ECC) is configured to deliver, for two input signals (A and B) to be compared, an output signal (S') such that S'= (notA AND notB) OR (A AND B).

[0142] In accordance with another feature of the invention, the computer system has a hybrid architecture configured such that, given the bitwise access provided by the elementary comparison circuits (ECCs), the binary data that constitute input data for the quantum circuit and the quantum register forming part of the quantum device are subjected to a quantum measurement of the probability amplitude of the quantum states. This measurement consists of reducing the wave packet to a single value (1 or 0), corresponding, for example, to V or C, respectively. After measurement, this single value is the output value to the traditional registers in mirror registers of the traditional computer associated with the quantum computer. The wave packet reduction corresponds to the wave function as reaching one of the two discrete values ​​associated with classical bits.

[0143] According to another feature of the invention, the hybrid architecture of the computer system is configured such that the classical or traditional processor processes the 8 bits of the register associated with the elementary comparison circuit (ECC) and the quantum processor controls a physical laser reading system of the spin of each of the 8 electrons associated with an 8-qubit quantum register, to enable the quantum processor to measure / read the result of each of the 8 qubits, whose quantum state is a probabilistic distribution of the 2-state system, V = 1 or C = 0, for each input character, the measurement result being recorded in a mirror register of the traditional or classical computer.

[0144] As is clear from the above, the glyph or typographic character is considered herein as an element of the FINITE set of an alphabet consisting of the written symbols (letters) and spoken symbols (phonemes) belonging specifically to the alphabet of each syllabic natural language written in binary vowels and consonants, in particular each natural language of the ISO 8859-1 family. Since this set can represent distinct values ​​within an interval of length 2 8[0; 255], we can define a bijective function, therefore reciprocal, between the human representation of a typographic character (for example, provided by a keyboard key) of a given language (ISO 8859-1) and the binary representation formed by the configuration of an 8-position integer word, each position occupied by a bit with a value of 1 or 0, constituting a byte memory cell. By accessing the bit in this way, it is possible to implement an elementary comparison circuit (see CCE in FIGURE 7) performing the comparison function described above, built with AND, OR, and NOT logic circuits, which compares an electrical signal and provides at output (S') the logic value 1 or 0 associated with the signal of current flowing (value 1) or not flowing (value 0).It is then understood that it is possible, by parallel arrangement of such elementary comparison circuits, to extend this principle to two whole words containing N places, in this case 8, thanks to an 8-place or input AND circuit, arranged as a sequence of elementary circuits (see Figure 6). This allows for the lexicographical comparison of two whole words (A and B). comprising 8 places: the 8 pairs of bits occupying the corresponding positions of the two words entering the 8 elementary comparators, each of which outputs a signal if, and only if, the bits in question are identical. By accepting these 8 elementary outputs (S') as inputs to an 8-input AND gate, we obtain a signal (S) at the output of the latter, which corresponds to the general output of the logic comparison circuit, if, and only if, the two words (a and B) are identical bit by bit, that is to say, "lexicographically" as elements of the set of VOWELS (V) or elements of the set of CONSONANTS (C). By convention, vowels (V) are associated with the value 1 and consonants (C) with the value 0. We thus obtain an equivalent system between the human representation of the alphabetic character and its BINARY representation at the hardware level (microprocessor and transistors), for example in CMOS technology.

[0145] Advantageously, the computer system according to the invention also includes a circuit for detecting and correcting quantum errors by measuring a single qubit whose state is separated from that of other states and which is coupled to two auxiliary qubits.

[0146] Indeed, since the MEASUREMENT of quantum states takes place in a FINITE vector space defined by the number of qubits and the number of associated registers [R 1 -4], the detection and correction of quantum errors can be performed by measuring a single qubit whose state is thus explicitly separated from the other states. This isolated qubit is in the superposition state |'4 J > = Qo |0> + Ch |1 > with probability p = | ax| 2 .

[0147] This single qubit is coupled to two auxiliary qubits initialized in the state |0>: using two cNOT gates, a circuit is constructed that encodes the 1-qubit state Qo |0> + Ch 11 > onto a 3-qubit state Qo |000> + ai 11 11 >. The circuit, which operates for the arithmetic basis states |0> and |1 >, is then extended linearly to any superposition of 2 to 8 states.

[0148] By applying the measurement gates only to the auxiliary qubits, we can extract information about the correlations that exist between the 3 encoding qubits without extracting information about the values particulars of Qo and œ, which may be completely unknown to the user. Information about correlations is therefore intrinsic to a two-state system, according to which a qubit is a two-level quantum system corresponding to the discrete states of classical bits, 0 and 1. During initialization or during reading / measurement, a value equal to |0> or |1> is accessed. Between initialization and reading, the qubit is in a superposition state and is therefore capable of taking on an infinite number of values, those associated with the recursively enumerable numbers of alphabetic language systems. Consequently, the qubit associated with natural language systems encoded in the alphabetic system is both a physical AND logical entity. As a reduction of the wave packet to a single value, the measurement is interpreted as taking values ​​corresponding to a discretization: the value |1> or vector |Vowel V> is opposed to the value |0> or vector |Consonant C>.As physical entities, these values ​​are associated with high spin and low spin. As logical entities, |1> is associated with the FINITE set, that is, the alphabet of UNICODE glyphs corresponding to the vowels used in the texts of a given natural language, and |0> is associated with the FINITE set, the alphabet of glyphs corresponding to the consonants used in these same texts. The information extracted after applying the measurement gates to the auxiliary qubits therefore concerns the dominant, that is, hierarchical, relationships between the qubits of the encoding sequence interpreted as Vowel vs. Consonant, and defines a FINITE set of elementary building blocks interpreted as a controlled assembly of |V> and |C>, considered as atoms isomorphic to the entangled syllables |R> associated with the registers R1, R2, R3, R4 of a given natural language.Quantum measurement is then associated with the process of matching a quantum register and a classical mirror register, in a hybrid architecture, to read (measure) the result of any lexical analysis system of languages ​​with V / C alphabetic writing.

[0149] The coupling of these two types of registers, which forms the basis of the hybrid architecture of the lexical language processing system according to the invention, therefore relies on the controllable subatomic writing of encoded languages. in the ISO 8859 standard. From this proceeds an AI which is itself controllable in the orthographic generation of lexical constructions and their differential interpretation in RDBMS distributed in the cloud (“cloud >>”), as shown in Figure 1.

[0150] For illustrative purposes, a typical alphabetical table is shown below:

[0151] It follows from the above that the invention makes it possible to overcome the limitations of extrinsic UNICODE encoding by implementing intrinsic encoding of natural language alphabets, in which the character encoding fits on 1 byte, since no natural language alphabet reaches the 256 values ​​of a single byte. This new encoding method allows for syntax-driven translation of the internal hierarchical structure of written and spoken characters composing words, represented by the construction of an abstract tree from leaves to root via a stack processor (see Figure 2: example 'SEIN').

[0152] The algorithm for constructing the abstract tree is ordered by the succession of the CreateSheet and CreateNode functions which define the string as the result of, on the one hand, the CreateSheet function which points to the entries in the symbol table (TSB) of the alphabetic characters of a given language specified as Vowel Lexical Unit V vs Consonant Lexical Unit C, and on the other hand, the CreateNode function which represents the binary concatenation operator which returns at the semiotic level a string of characters and at the ontological level an n-tuple type structure of the attributes V and C which compose it.

[0153] The lexical analysis of the input character stream takes in our algorithm the polynomial time of the construction of the abstract tree, via the CreateSheet and CreateNode functions associated with a memory space organized by the arrays allocated to the nodes of the abstract tree and accessed by a direct indexed addressing defined by the bijective function of the set of addresses on the set of memory cells (see figure 3).

[0154] These tables are defined by internal composition laws in each language, which characterize functional dependencies, join dependencies and inclusion dependencies (see figure 4 and associated tables).

[0155] The abstract tree construction algorithm thus produces a set of strings that can be modeled by a priori independent tables in relational databases (RDBMS) for distributed storage in alphabetic language containers in the "cloud" (Figure 5), which represents the bijective relationship between the bit strings of memory words and the bit strings of the stack processor registers in the construction of the abstract tree modeled in Figure 2. However, since the arrays have a finite dimension, known a priori as powers of 2: 2 1 _ 2 8The concatenation construction program operates independently of the number of array elements, using indexed direct addressing to sequentially read the elements—bit strings—of an array. Big data processing is therefore independent of the exponential size of the input data. And, since intermediate code generation in the front end of the compilation allows for the association of bit registers and qubit registers, measuring the qubit registers by simultaneous access, via quantum superposition, to all elements of the quantum register yields an output in a mirror register in the form of bit strings.

[0156] This linguistic data model thus makes it possible to specify an artificial intelligence (AI) controlled by the analysis of the internal syntax of words in natural languages ​​with C / V alphabetic writing, combined with the physical and logical power of coupling quantum and classical registers. It contrasts with existing artificial intelligences implemented in exponential time through training on massive linguistic databases.

[0157] In relation to the invention described above, and more specifically with regard to European languages, which use a finite number of letters in the alphabet, the key corresponds to an elementary or atomic type character, unlike Asian languages ​​which are confronted the problem that there are many more characters (or glyphs) than keys on a computer keyboard, for example.

[0158] The characters of such alphabets (natural syllabic languages ​​with a binary vowel / consonant alphabetic script - ISO 8859-1 type) can be combined to specify a composite structure, a type of aggregate called a syllable type. This composite type allows the definition of a type controller that implements an algebraic grammar by constructing, using a stack processor, an abstract tree from the bottom (the leaves pointing to the generic type characters - Vowel 'V vs Consonant 'O - of an alphabet) upwards (that is, towards the intermediate or terminal node(s)) which represent the recursive binary operation of concatenating the alphabetic characters, each associated with a UNICODE identifier and entered as input.The syllable is therefore the result (or the axiom of algebraic grammar) of the construction of a strongly typed language whose elements are obtained by concatenating an arbitrary number, positive or zero, of syllabic aggregates which each correspond to a MEMORY WORD represented by the strings of typographic characters belonging to the alphabet of a given natural language of the aforementioned type.

[0159] This string construction program, combined with a static type constructor, allows for the front-end compilation of an ISO 8859-1 type alphabetic source language into a generic intermediate language or code of the assembler type, due to the isomorphism between the registers of the implemented stack-based processor and the memory locations represented by arrays. These arrays are traversed to perform a lexicographical comparison of each field of the syllabic structure. It is then the constructed type of the FINITE-DIMENSIONAL array, known a priori, that allows the specification of multiprogrammed low-precision computing units (8 8-bit cores) associated with a hybrid architecture of classical and quantum registers exploiting the superposition and entanglement properties of quantum bits. The syllabic structure as described and used herein therefore has a technical impact on computer processes for inputting and encoding characters, particularly those entered via keyboard.

[0160] In summary and considered as a whole, in relation to the attached figures, the automatic text processing defined herein essentially comprises the following steps, carried out in relation to a computer system comprising a hybrid architecture with a classical or traditional computer associated with a quantum computer: - Creation for each language in the ISO 8859-1 family of a 2-dimensional alphabetical table of written AND spoken symbols (TSB) according to the standard [1 character - 1 string of 8 bits / 1 byte], from the unique identifier of the UNICODE standard, - Inputting the text to be processed character by character, either by keyboard input or phonetic transcription, and eliminating spaces and separator symbols, such as hyphens, apostrophes, punctuation, or similar elements, from the written text. - Input of the 8-bit string into the elementary lexicographic comparator (ECC) circuit and matching with the TSB bit string model of a language from the ISO 8859-1 family, and obtaining at output, for each character, the generic type V = 1 vs C = 0 associated with a TSB glyph, - Glyph storage in field V or C of the TSB of the ISO 8859-1 language, - Loading the bit string associated with the glyph into the stack processor register, - Provision of a hybrid architecture comprising the classical processor which processes the 8 bits of the register associated with the elementary lexicographic comparator (ECC) and a quantum processor which controls a physical laser reading system for the spin of each of the 8 electrons associated with an 8-qubit quantum register, - Measurement or reading by the quantum processor of the result of each of the 8 qubits, whose quantum state is a probabilistic distribution of the 2-state system, V = 1 or C = 0, for each input character and recording of the measurement result in the classical mirror register, - Construction, from pointers to the classic mirror registers, of the abstract tree directed by the algebraic grammar of V and C, specific to each of the languages ​​of the ISO 8859-1 family, by applying the recursive function of concatenation via the CreateLeaf and CreateNode procedures, the result being of composite type Syllable, - Storing this result for each language in a virtual memory (VM) array of generic syllables in CMOS technology pointing to the arrays instantiated in real memory (MR) with the values ​​of the alphabetic characters composing the syllabic structures of each language in the ISO 8859-1 family, - Application of these instantiated syllables as inputs to the composite lexicographical comparison circuit (CCLC) which filters, by comparison with the generic model of syllables in the virtual memory space (MV) pointing to syllables in the real space (MR), the outputs as well-formed expressions (ebf) opposed to errors with regard to the production rules of the syllable grammar defining the syllabic language R: [R1, R2, R3, R4] for each language ISO 8859-1, - Realization, from these types of syllables [ R1 , R2, R3, R4 ] stored in a computer tree, and by application of the recursive function of concatenation, of the construction of the abstract tree of linguistic words of an intermediate language (L*) defined by the morphosyntax of an assembly language for an abstract machine and corresponding to the language of linguistic words obtained by concatenation of an arbitrary number, positive or zero, of generic syllables of the language R, realized in the written and oral instances of the syllables of each language ISO 8859-1 , - Storage of the L* language in a DBMS, dedicated to each ISO 8859-1 language, for the purpose of inferential calculation on the components of linguistic words generated by the production rules of the generative grammar of identifiers of each language.

[0161] The result is a tokenization of lexical constructs, controlled by inference rules that define the attributes of the tables which store words in ISO 8859 language containers distributed in the cloud (FIG 1). The L* language model is therefore defined as a formal language defined by a general grammar of algebraic type whose terminal vocabulary is constituted by the union of the ISO 8859-1 U ISO 8859-2 U ISO 8859-3 U ISO 8859-4 U ISO 8859-5 U ISO 8859-7 languages, i.e. the set of languages ​​with syllabic alphabetic writing with consonants / vowels.

[0162] This definition of linguistic words directed by grammar rules is to be compared with the tokenization done in the "Generative Pretrained Transformer" (GPT) model which is implemented in LLMs, based on GPU processors operating on an infinite volume of linguistic data represented by the pixels of the glyphs opposite to the symbolic value of the glyphs defined in the present invention.

[0163] Evaluating the expressive power of the two representation models allows us to quantify the resources and energy cost required for tokenization / analysis of linguistic words.

[0164] Of course, the invention is not limited to the embodiment described and shown in the accompanying drawings. Modifications remain possible, particularly with regard to the composition of the various elements or by substitution of technical equivalents, without departing from the scope of protection of the invention.

Claims

Demands

1. A method for the automatic processing of texts in syllabic natural languages ​​with binary alphabetic vowel / consonant writing, by lexical analysis with implementation of a stack processor and by recognition of lexical units or words by comparison, a method characterized in that it consists of transforming each series of alphabetic characters, each identified by a UNICODE identifier and corresponding together to a word or a lexical unit, by encoding each of its characters in the form of an eight-bit string in accordance with one of the ISO 8859 standards, preferably ISO 8859-1 or Latin-1, and using this bit string to excite in parallel, simultaneously and selectively, the eight inputs of an 8-input, 256-output switching matrix,These operations result in the production of a specific generic intermediate code by transformation by the stack processor of a linear structure of the input word or text into a hierarchical structure of an abstract tree, whose nodes represent the concatenation operation and the fields instantiated by the generic types (V and C), then to use the resulting specific intermediate code to perform a binary differential interpretation, of abstract and generic type, between each of the lexical units or each of the words to be processed, on the one hand, and the set of lexical units or words of the language concerned, on the other hand, this binary differential interpretation consisting, by means of elementary comparison circuits (ECC), of comparing in parallel between the two bits of the eight pairs of bits of the same rank or position,respectively corresponding to each of the lexical units or each of the words (A) to be processed and the lexical units or words (B) of the language concerned, then to mutually combine the results of the different elementary comparisons by means of an eight-input AND function, until a bit identity is obtained for all eight pairs of bits, corresponding to a lexicographically exact identity between a character of the lexical unit to be processed or of the word to be processed considered and a character of the alphabet of the language concerned, This differential interpretation is further achieved by implementing a quantum measurement performed by processing a quantum register associated with a computer register containing said specific generic intermediate code produced.

2. A method according to claim 1, characterized in that it consists of providing a matrix in the form of an addressable memory of the pointer array type (I, T), having 8 input lines and 256 output lines, this matrix corresponding to an address register composed of the two sub-registers of the generic types (V and C) for a given language, and of associating the binary generic type (V or C), for any character of this language, with an interpretation valid for a physical and logical quantum state of a qubit, with a superposition of two states with equal probabilities, namely 1 / 2, the registers of the 8-bit strings being correlated to the quantum registers of an 8-dimensional vector space in a hybrid architecture of the quantum physics / classical physics type.

3. A method according to claim 1 or 2, characterized in that it consists of storing the positive output of the AND function in an array of the type (I, T) of vowels and consonants of the language in question, which thus forms a symbol table (TSB) whose indices correspond to the alphabetical order of the typographic characters, their static type as an abstract type of metacharacter and generic type of lexical unit (V or C), as well as their configuration consisting of 8 bit locations.

4. A method according to claim 3, characterized in that it consists of constructing an abstract tree (AA) as a sequence of subtree constructions from bottom to top, whose leaves point to the entries of the symbol table (TSB) and whose nodes represent the result of the concatenation of the values, corresponding to the characters or strings of characters, associated with the leaves which are the operands of the concatenation operator.

5. A method according to claim 4, characterized in that this syntactic construction of an abstract tree is driven by an alphabetic byte register constructor algorithm, which produces, from an input consisting of a set or sequence of characters in an array or register pointing to two sub-arrays or sub-registers of vowel (V) and consonant (C), an abstract tree constructed from bottom to top by the recursive concatenation, the result of this alphabetic algorithm defining an assembly language, in which the expressions resulting from the concatenation represent memory words at the nodes of the abstract syntax tree corresponding to the registers of the automaton or the stack processor, and producing the intermediate code of generic type.

6. A method according to claim 5, characterized in that the analysis performed by the automaton or the stack-driven processor corresponds to the implementation of production rules for an underlying algebraic grammar generating the lexical constructions, or memory words, of a language of variable length from 1 to 8 characters, in registers of variable size ranging from 8 bits / 1 byte to 64 bits / 8 bytes, these production rules advantageously providing four string models representing four types of abstract syllabic structures composed by recursive concatenation, corresponding to four record models or classes (R1, R2, R3 and R4), namely: - Recursion stopped: Vowel expression = class R1 - Left recursion: ConsonantExpression + VowelExpression = class R2 - Right recursion: VowelExpression + ConsonantExpression = class R3 - Right-hand recursion and left-hand recursion = consonant expression + vowel expression + consonant expression = class R4

7. A method according to any one of claims 3 to 6, characterized in that it consists of carrying out a lexicographical comparison between the lexical unit to be processed considered or of the word to be processed considered and the lexical units or words of the language concerned, serving as models, by comparing in parallel between the pairs of bytes of the same rank or position of the two lexical units or words, by implementing an adequate number of comparators, and storing the result if each comparator delivers a signal confirming the identity.

8. A method according to any one of claims 5 and 6, characterized in that the memory words, from 1 to 8 bytes, are processed by a lexicographic comparator of pairs of bytes of the same rank or position, the method consisting of mutually combining the results of the different lexicographic comparisons by means of an AND function with 1 to 8 inputs, until an identity of memory word is obtained for the pairs of bytes, corresponding to an exact lexicographic identity between the lexical unit to be processed considered or word to be processed considered and a memory word of the language concerned.

9. A method according to claim 7 or 8, characterized in that the storage of the word or lexical unit corresponding to the positive result of the lexicographical comparison consists, depending on the number N of places taken by the configuration of the word or lexical unit forming said result, of carrying out a conditional branch with the value of N in the interval [1 - 8] as input and the 8 pointer arrays (I, T) as parallel output for each configuration model from 1 to 8 places pointing to the memory locations associated with the physical and logical records of the words or lexical units of the language concerned.

10. A method according to claim 9, characterized in that each pointer (T) defines an addressing as a bijective function of the set of addresses, namely the indices (I), over the set of memory cells implemented in a technology adapted to the binary values ​​of the arrays (I, T) for the registers of the stack processor and the memory locations of the main memory associated therewith. [Claim 1 1] Method according to claim 9 or 10, characterized in that, given the four recording models (R1, R2, R3 and R4) associated with the byte strings of the configuration models corresponding to the words or lexical units to be processed, the conditional branching with its eight possible outputs, which correspond to 8 columns for said four recording models above, creates a single matrix for branching out of said four recording models.

12. A method according to claim 11, characterized in that the single matrix points to tables of one to eight columns forming a finite set of data structures represented in virtual memory, itself pointing to the real addresses of configurations instantiated in real memory, in which each generic type vowel string (V) and each generic type consonant (C) is instantiated by the metacharacters and glyphs of vowels and consonants belonging to the alphabet of each language in the family of languages ​​encoded by the ISO 8859 standard.

13. A method according to claim 12, characterized in that it consists of using the finite set of data structures as a filter for filtering or discriminating between well-formed expressions and malformed expressions or errors.

14. A method according to claim 12 or 13, characterized in that it consists of providing a diode matrix which integrates the finite set of data structures, the conditional branching of said diodes exciting wires by connecting diodes according to the input values ​​of the wires excited or not and the planned switching according to pinout models providing differential interpretations of the controlled identifiers.

15. A method according to any one of claims 6 to 14, characterized in that, with a view to defining an assembly language in which the expressions resulting from concatenation represent memory words at the nodes of the abstract syntax tree corresponding to the registers of the stack processor, for input data in the form of a character stream in a BufferLex array pointing to two sub-arrays of vowels "V" and consonants "C", it consists of performing the following operations: 1) Entering the character of the 1st into the TamponLex table ère vowel and any subsequent vowels, and their loading into the vowel register, and evaluation of the attributes attached to the VowelExpression node by calling the CreateSheet and CreateNode functions, the result of which is - either the root node, - either an internal knot, - then calculate the concatenation of the 2 leaf nodes of the higher level node and evaluate the result as a root node or as an interior node. - If internal node, then iterate the concatenation. 2) Input of consonants, if any, into the TamponLex array to the right of the VowelExpression node and their loading into the consonant register, and evaluation of the attributes attached to the ConsonantExpression node by calling the CreateSheet and CreateNode functions, the result of which is - either the root node, - either an internal knot, - then calculate the concatenation of the 2 leaf nodes of the higher level node and evaluate the result as a root node or as an interior node. - If internal node, then iterate the concatenation. 3) Input of consonants, if any, into the TamponLex array to the left of the VowelExpression node and their loading into the consonant register, and evaluation of the attributes attached to the ConsonantExpression node by calling the CreateSheet and CreateNode functions, the result of which is - either the root node, - either an internal knot, - then calculate the concatenation of the 2 leaf nodes of the higher level node and evaluate the result as a root node or as an interior node. - If internal node, then iterate the concatenation.

16. A method according to any one of claims 1 to 15, characterized in that the generic intermediate code is produced by an alphabetic algorithm in which the concatenation results represent memory words at the nodes of the abstract syntax tree corresponding to the processor registers, and in that the production of said intermediate code uses three registers for respectively the binary values ​​of the first and second input characters and the result and provides a three-address code consisting of a sequence of instructions, each of which has at most three operands, and in that this intermediate code production consists of a recursive construction of abstract strings carried out by the stack processor which evaluates the postfixed representation of the concatenation: a + b = ab+, resulting recursively in an abstract language of abstract strings built on the alphabet of lexical units (V vs C), associated with the real types of glyph strings of each natural syllabic language with binary vowel / consonant alphabetic writing.

17. A method according to any one of claims 1 to 16, characterized in that it comprises -1) the explicit management of reading at the level of the central microprocessor (CPU) and the main memory of the computer system used for its implementation, via an interface which ensures the compatibility of binary signals and controls them by a logic comparison circuit, performing a hardware-level interpretation of said signals as an alphabetic human character, namely an element of class "V" or an element of class "C", uniquely associated with a binary machine character (0 or 1); -2) the construction by a stack processor at the level of an intermediate representation of the abstract tree whose root is the constructed type memory word, via the recursive binary concatenation operation, interpreted as an abstract syllable defined by the normalized configurations of "V" and "C" in any ISO 8859-1 language, instantiated by the concrete syllables realized in each language of the ISO 8859-1 family, this memory word being stored in a computer tree.

18. A method according to any one of claims 1 to 17, characterized in that it essentially comprises the following steps, carried out in connection with a computer system comprising a hybrid architecture with a classical or traditional computer associated with a quantum computer: - Creation, for each language in the ISO 8859-1 family, of a 2-dimensional alphabetical table of written AND spoken symbols (TSB) according to the standard [1 character - 1 string of 8 bits / 1 byte], based on the unique identifier of the UNICODE standard, - Inputting the text to be processed character by character, either by keyboard input or phonetic transcription, and eliminating spaces and separator symbols, such as hyphens, apostrophes, punctuation, or similar elements, from the written text. - Input of the 8-bit string into the elementary lexicographic comparator (ECC) circuit and matching with the TSB bit string model of a language from the ISO 8859-1 family, and obtaining at output, for each character, the generic type V = 1 vs C = 0 associated with a TSB glyph, Glyph storage in field V or C of the TSB of the ISO 8859-1 language, - Loading the bit string associated with the glyph into the stack processor register, - Provision of a hybrid architecture comprising the classical processor which processes the 8 bits of the register associated with the elementary lexicographic comparator (ECC) and a quantum processor which controls a physical laser reading system for the spin of each of the 8 electrons associated with an 8-qubit quantum register, - Measurement or reading by the quantum processor of the result of each of the 8 qubits, whose quantum state is a probabilistic distribution of the 2-state system, V = 1 or C = 0, for each input character and recording of the measurement result in the classical mirror register, - Construction, from pointers to the classic mirror registers, of the abstract tree directed by the algebraic grammar of V and C, specific to each of the languages ​​of the ISO 8859-1 family, by applying the recursive function of concatenation via the CreateLeaf and CreateNode procedures, the result being of composite type Syllable, - Storing this result for each language in a virtual memory (VM) array of generic syllables using CMOS technology, pointing to the arrays instantiated in real memory (MR) with the values ​​of alphabetic characters composing the syllabic structures of each language in the ISO 8859-1 family, - Application of these instantiated syllables as inputs to the composite lexicographical comparison circuit (CCLC) which filters, by comparison with the generic model of syllables in the virtual memory space (MV) pointing to syllables in the real space (MR), the outputs as well-formed expressions (ebf) opposed to errors with regard to the production rules of the syllable grammar defining the syllabic language R: [R1, R2, R3, R4] for each language ISO 8859-1, - Realization, from these types of syllables [ R1 , R2,R3,R4 ] stored in a computer tree, and by application of the recursive function of concatenation, of the construction of the abstract tree of linguistic words of an intermediate language (L*) defined by the morphosyntax of an assembly language for an abstract machine and corresponding to the language of linguistic words obtained by concatenation of an arbitrary number, positive or zero, of generic syllables of the language R, realized in the written and oral instances of the syllables of each language ISO 8859-1.

19. Storage of the L* language in a dedicated RDBMS for each ISO 8859-1 language, for the purpose of inferential calculation on the components of linguistic words generated by the production rules of the generative grammar of identifiers of each language.

20. A computer system for implementing the automatic text processing method for syllabic natural languages ​​with binary alphabetic vowel / consonant writing according to any one of claims 1 to 18, these texts being formed of lexical units or words (A), a system characterized in that it has a hybrid architecture in which a central processing unit (CPU) and its traditional bit registers forming a traditional computer are coupled with a quantum device with several, preferably eight, qubits, and its associated quantum registers, forming a quantum computer, and in that it comprises, on the one hand, a logic comparison circuit comprising eight elementary comparison circuits (ECCs), arranged in parallel and capable of comparing with each other the two bits of the eight pairs of bits of the same rank or position, respectively corresponding to each of the lexical units or each of the words (A) to be processed and the lexical units or words (B) of the language concerned, each of whose alphabetic character is identified by a UNICODE identifier, and on the other hand, an eight-input AND circuit, combining the results (S') of the different elementary comparisons carried out by the eight elementary comparison circuits (ECC).

21. Computer system according to claim 19, characterized in that it comprises a circuit for detecting and correcting quantum errors by measuring a single qubit whose state is separated from that of other states and which is coupled to two auxiliary qubits.

22. Computer system according to claim 19 or 20, characterized in that each elementary comparison circuit (ECC) is configured to deliver, for two input signals (A and B) to be compared, an output signal (S') such that S'= (notA AND notB) OR (A AND B).

23. A computer system according to any one of claims 19 to 22, characterized in that it has a hybrid architecture which is configured such that, given the bit access provided by the elementary comparison circuits (ECC), the binary data which form input data for the quantum circuit and the quantum register forming part of the quantum device are subjected to a quantum measurement on the probability amplitude of the quantum states and consisting of reducing the wave packet to reach a single value (1 or 0), corresponding for example respectively to V or C, and in that, after the measurement, this single value is the output value to traditional registers corresponding to in mirror registers of the traditional computer of the hybrid architecture.

24. A computer system according to any one of claims 19 to 22, characterized in that its hybrid architecture is configured such that the classical or traditional processor processes the 8 bits of the register associated with the elementary comparison circuit (ECC) and the quantum processor controls a physical laser-based spin-reading system for each of the 8 electrons associated with an 8-qubit quantum register, to allow the quantum processor to measure / read the result of each of the 8 qubits, whose quantum state is a probabilistic distribution of the 2-state system, V = 1 or C = 0, for each input character, the measurement result being recorded in a mirror register of the traditional or classical computer.

Images