FR3087555A1 - DEVICE FOR AUTOMATIC TEXT PROCESSING BY COMPUTER - Google Patents

Abstract

An automatic computer word processor that receives a tokenized sentence, and uses a rectifier, matcher, and combiner repeatedly, to produce a tree-like structure that describes the syntactic and semantic links of the sentence on the basis of a non-monotonic probabilistic logic.

Description

MEREMIND1.FRD Device for automatic word processing by computer The invention relates to the field of automatic processing of a text by computer, and more particularly to the field of natural language processing, or NLP (Natural Language Processing).

There are two main branches of natural language processing: linguistic methods and automatic training-based methods.

10 The methods of the first branch are based on the theory of languages derived from Chomsky's theories.

However, this branch has only been the subject of implementations based on largely manual and complex heuristics and has never known a computer application giving satisfactory results for a general application.

15 The foundations of artificial intelligence based on automatic training were laid in the 1960s.

In the last five years, according to technological developments, and with the explosion of the quantities of data accessible to carry out automatic training (MIL for machine learning in English, DL for Deep 20 Learning in English and NN for Neural Networks in English) , this field has grown exponentially.

In general, automatic workouts are based on the determination by a machine of a statistical model determined on the basis of a workout, the parameters of which are set by the person who programs the workout, and on the basis of a practice game.

In practice, this means that the designer understands the training principles and parameters as well as the data on which the training is based, but not the result, which is called the inference model.

Thus, once the training is finished, it is the inference model that is used to make the predictions on the input data that one wishes to process, without the designer being able to see it other than as. a black box.

2 Applied to NLP, these methods are mainly based on the vectorization of words, with models of the Bag-of-words type, or even Word2Vec, and the treatment of these vectors as an automatic training problem.

These models are also based on a syntactic analysis sequence followed by a semantic analysis.

Thus, the sentences are first cut out syntactically, then a meaning is applied to the result.

Nevertheless, the "black box" nature of the inference models produced is contradictory to the objective pursued in the NLP.

Indeed, language has a meaning, it is even its foundation.

Moreover, this meaning is not expressed only by syntax or by semantics, but by a combination of the two.

For these reasons, and contrary to what one might think, the solutions of this second branch are also based on a large quantity of heuristics, which have the additional defect of being devoid of sense or of perceptible logic, because they are created to satisfy the inference engine whose organization is not understood There is therefore a need for a stable automatic word processor, functional in the most general sense of the term, and which does not rely exclusively on heuristics.

The invention improves the situation.

To this end, the invention provides a device for automatic computer word processing, comprising a memory arranged to receive text data to be analyzed in the form of tokens each comprising a string of characters and a unique token identifier, a base of concept data 25 associating character strings and concept identifiers, at least some of the concept identifiers being associated with one another, model lexical construct data and model structural construct data, each comprising one or more conditions of application to a characteristic and one or more conclusions constituting elements to be applied to a characteristic, and a database of observations associating at least two concept identifiers, a type of relation and an observation value indicating a probability of veracity the type of relationship between the at least two concept identifiers, the device being arranged to work repetitively on a transient comprising lexical features and structural features produced by applying model lexical constructions and model structural constructions, the transient being initialized with lexical features comprising for each token a concept identifier whose frequency is most important in the concept database and which is associated with the token character string.

This device further comprises: a rectifier arranged to determine for each lexical characteristic of a transient a list of concept identifiers associated with the concept identifier of this lexical characteristic, to determine a set of observations corresponding to the identifiers of concept of the lists thus determined, and to apply a non-monotonic probabilistic logical inference engine to determine the concept identifier of each list such that the observation values associated with these concept identifiers minimize a cost function defined in applying a multivalued logic operator to one or more rules taken from the content of the characteristics of the transient and instantiated with corresponding observation values, and to replace the concept identifiers of the lexical characteristics of the transient by the concept identifiers thus determined, 20 - a pairer arranged to determine among the constructio ns structural models those whose condition (s) apply to one or more of the characteristics of the transient, and to return the list of structural constructions with the characteristic (s) to which their conditions apply, the device being furthermore arranged to classify the structural constructs associated with each characteristic by frequency of use, the first structural construct to be applied to the transient and the others forming a list of options, and a combiner arranged to sequentially perform the selection of a structural construct to be applied. applied to the transient, storing a copy of the transient with the list of options associated with the structural construction, and applying the structural construction to be applied to the transient to the characteristic (s) of the transient at which this structural construction has been associated by 4 the pairer, and to repeat r this sequential execution on the transient thus modified with the structural construction having to be applied to the following transient.

This device is furthermore arranged to determine, after the execution of the combiner, whether the characteristics of the produced transient define a tree of which all the nodes are linked to each other and without a cycle, to return this tree if this is the case, and to repeat the execution of the rectifier, the pairer and the combiner on the last transient produced otherwise, and is further arranged, when the pairer does not return any structural construct to be applied to the transient, to replace the current transient with the copy of the most recent transient, and to run the combiner with the first construct of the option list as a structural construct to apply to the transient.

This device is particularly advantageous because it makes it possible to solve the problems described above.

Indeed, it relies entirely on the application of an algorithm whose rules are linked to linguistics.

In various variants, the device may have one or more of the following characteristics: - the non-monotonic probabilistic logic inference engine comprises an optimizer using the algorithm of multipliers with alternating directions, - the device further comprises a filter designed to determine , for each structural construction to be applied and the associated list of options a set of rules, to determine a set of observations from the set of rules and the concept identifier (s) associated with the characteristic to which the rule must be applied. structural construct to be applied, and to apply the non-monotonic probabilistic logical inference engine with the rule set and observation set to determine the structural construct to be applied to the transient and the list of options, - the rectifier defines frequency rules from lexical characteristics of the transient, rules of e neighborhood from the concept identifiers of the lexical characteristics of the transient and the concept identifiers associated with them in the concept database at a chosen distance, and structural rules drawn from semantic attributes 5 of non-lexical characteristics of the transient linking two lexical characteristics together, the combiner is arranged to store only the copy of the transient with the list of options associated with the structural construction, and in that, when the pairer does not return any structural construction to be applied to the transient , to replace the current transient with the most recent copy of the transient, and to run the combiner with the first construct in the options list as a structural construct to apply to the transient, then repeating the application of the rectifier, from the pairer and combiner with the resulting transient, 10 - the combiner is arranged for st Store the remaining structural constructions to be applied to the transient at the same time as the copy of the transient with the list of options associated with the structural construction, and in that when the matcher does not return any structural construction to be applied to the transient, to replace the current transient by copying the most recent transient, and to run the combiner with the first construct from the option list as a structural construct to apply to the transient along with the remaining structural constructs, then repeating the application of the rectifier, of the pairer and of the combiner with the resulting transient, - the device is arranged to analyze the returned tree and to produce a semantic graph whose nodes are formed by the lexical characteristics and their concept identifier, and therefore the links are defined by the semantic attributes of non-lexical characteristics linking between el the two lexical characteristics, and the inference engine is designed to apply a multivalued logic operator chosen from the group comprising the Lukasiewicz t-norm, the minimal t-norm and the Hamacher product.

The invention also relates to a computer-implemented automatic text processing method, comprising the following operations: a) receiving text data to be analyzed in the form of tokens each comprising a character string and a unique token identifier , a concept database associating character strings and concept identifiers, at least some of the concept identifiers being associated with each other, model lexical construct data 6 and model structural construct data, each comprising one or more conditions of application to a characteristic and one or more conclusions constituting elements to be applied to a characteristic, and a database of observations associating at least two concept identifiers, a type of relation and an observation value indicating a probability of veracity of the type of relation between the at least two identifiers of concept, b) initiali be a transient capable of comprising lexical features and structural features produced by applying model lexical constructions and model structural constructions, with lexical features comprising for each token a concept identifier with the highest frequency in the database concepts and which is associated with the character string of the token, c) work repeatedly on the transient by repeating the following successive operations: c) determine for each lexical characteristic of a transient a list of associated concept identifiers to the concept identifier of this lexical characteristic, to determine a set of observations corresponding to the concept identifiers of the lists thus determined, and to apply a non-monotonic probabilistic logical inference engine to determine the concept identifier of each list such as observation values associated with these concept identifiers minimize a defined cost function by applying a multivalued logic operator to one or more rules taken from the content of the characteristics of the transient and instantiated with corresponding observation values, and to replace the concept identifiers of the transient. lexical characteristics of the transient by the concept identifiers thus determined, 25 c2) determine among the model structural constructions those whose condition (s) apply to one or more of the characteristics of the transient, and to return the list of structural constructions with the or the characteristics to which their conditions apply, c3) classify the structural constructions associated with each characteristic by frequency of use, the first structural construction to be applied to the transient and the others forming a list of options, and 7 c4) to execute sequentially the selecti on of a structural construction to be applied to the transient, storing a copy of the transient with the list of options associated with the structural construction, and applying the structural construction to be applied to the transient at the feature (s) of the transient to which this structural construction has been associated by the pairer, and to repeat this sequential execution on the transient thus modified with the structural construction to be applied to the following transient, c5) determine, after the execution of the combiner, if the characteristics of the produced transient define a tree of which all the nodes are linked to each other and without a cycle, 10 to return this tree if it is the case, and to repeat the operations cl) to c5) on the current transient otherwise, c6) if operation c2) does not return any structural construction to be applied to the transient, replace the current transient by the copy of the transient l a more recent, perform operation c5) with the first construct from the list of options as the structural construct to be applied to the transient.

In various variants, the method may have one or more of the following characteristics: - the application of the non-monotonic probabilistic logic inference engine comprises the application of an optimizer using the algorithm of multipliers with alternating directions, - the method further comprises, between operation c3) and operation c4) c7) determining, for each structural construction to be applied and the associated list of options a set of rules, determining a set of observations from the set 25 rules and the concept identifier (s) associated with the characteristic to which the structural construction to be applied must be applied, and apply the non-monotonic probabilistic logical inference engine with the rule set and the observation set in order to determine the structural construction to be applied to the transient and the list of options, 30 - operation c) includes the definition of frequency rules from d e lexical characteristics of the transient, of neighborhood rules from the concept identifiers of the lexical characteristics of the transient and of the concept identifiers associated with them 8 in the concept database at a chosen distance, and of structural rules drawn from semantic attributes of non-lexical characteristics of the transient linking together two lexical characteristics, - operation c4) store only the copy of the transient with the list of options associated with the structural construction, and in which, after the execution of l 'operation c7), operations cl) to c7) are repeated on the current transient, - operation c4) stores the remaining structural constructions to be applied to the transient at the same time as the copy of the transient with the associated list of options to the structural construction, and in which, after the execution of the operation c7), the operation 10 c5) is applied with the constructions stru cturelles, then by repeating the application of the rectifier, then, operations c1) to c7) are repeated on the current transient, and - the method further comprises the following operation: d) analyze the tree returned by the operation c6) and produce a semantic graph 15 whose nodes are formed by the lexical characteristics and their concept identifier, and therefore the links are defined by the semantic attributes of non-lexical characteristics linking together two lexical characteristics.

Other characteristics and advantages of the invention will emerge better on reading the description which follows, taken from examples given by way of illustration and not limiting, taken from the drawings in which: FIG. 1 represents a schematic view of a device according to the invention, FIG. 2 represents an example of implementation of an automatic text processing function by the device of FIG. 1, FIGS. 3 to 7 represent examples of implementation of 'operations of Figure 2, - Figures 8 to 17 represent representations of processing steps of the function of Figure 2 in a simplified example, - Figure 18 shows a shaft returned by the device according to the invention, and - figure 19 represents a semantic graph taken from the tree of figure 18.

The drawings and the description which follow essentially contain elements of a definite nature.

They can therefore not only serve to better understand the present invention, but also contribute to its definition, if necessary.

This description is likely to involve elements susceptible of protection by copyright and / or copyright.

The rights holder has no objection to any identical reproduction by any person of this patent document or its description as it appears in official records.

For the rest, he reserves his rights in full.

Figure 1 shows a schematic view of a device according to the invention.

The device 2 comprises a memory 4, a rectifier 6, a pairer 8, a filter 10, a combiner 12 and a validator 14.

In the context of the invention, the memory 4 can be any type of data storage suitable for receiving digital data: hard disk, hard disk with flash memory (SSD in English), flash memory in any form, random access memory, magnetic disk, storage distributed locally or in the oud key, etc.

The data calculated by the device can be stored on any type of memory similar to memory 4, or on the latter.

This data can be erased after the device has performed its tasks or kept.

In the example described here, the memory 4 receives permanent data which is used to implement the device 2 and can be enriched as it is executed.

In the example described here, the memory 4 also receives temporary data or work data, which are generated for the needs of a given execution of the device 2, and which are not kept after this execution.

The permanent data and the working data can be stored on the same memory 4 or on separate memories.

30 The master data includes a database of semantic concepts based at least in part on the Wordnet database (see https://wordnet.princeton.edu/) for the English language (other databases 10 may be used for other languages).

The permanent data also contain, and in a non-exhaustive manner, the following data, which will be defined in detail below: - model lexical constructions for a given language, that is to say a condition 5 based on the lexical category d '' a word (its Part of speech in English), and a conclusion allowing to characterize the lexical characteristic by its character string, its part of speech, its unique identifier of token and a concept identifier, - model structural constructions for a given language, i.e. one or more conditions which connect a lexical or structural characteristic and another lexical or structural characteristic and reflects the syntactic structuring of the given language, such as whether an adjective is prepended or postposed , or the structure of a subordinate proposition of cause, etc., - filtering rules, which make it possible to formulate each a rule associated with a structural construct it, in order to remove a lexical ambiguity thanks to the semantics, and 15 - a database of observations, which are quadruplets associating a type of relation, two identifiers of concepts, and an observation value between 0 and 1 (0 means that the type of relation suggested is false, 1 that it is true, and the other values a probability that it is true or false).

For example, for the relation type Is One (A, B), the value of the observation of Is One (Einstein, man) would be 1, while 20 the value of the observation of Is One (Einstein, dog) would be 0, and that of Is One (Einstein, genius) would be 0.95.

Thus, according to a concept of "closed world" (the world is closed), if two concepts are not connected, then the observations connecting them are initialized to 0.

This observation database can be supplemented from the contents of the concept database to deduce logical relationships between the 25 concepts.

Observations could link more than two concept identifiers.

Nevertheless, this type of relation can be reformulated by a set of several observations linking the concept identifiers in pairs.

In the context of the invention, the rectifier 6, the pairer 8, the filter 10, the combiner 12 and the validator 14 are elements which directly or indirectly access the memory 4.

They can be implemented in the form of appropriate computer code executed on one or more processors.

By processors, it must be understood any processor suitable for the calculations described below.

Such a processor can be produced in any known manner, in the form of a microprocessor for a personal computer, of a dedicated chip of FPGA or SoC (“system on chip”) type, of a computing resource on a computer. grid, a microcontroller, or any other form suitable for providing the computing power necessary for the embodiment described below.

One or more of these elements can also be made in the form of specialized electronic circuits such as an ASIC.

A combination of processor and electronic circuits can also be envisaged.

FIG. 2 represents an example of the implementation of an automatic word processing function by the device of FIG. 1 The role of the function of FIG. 2 is fundamental.

Indeed, it is a loop which calls the rectifier 6, the pairer 8, the filter 10, the combiner 12 and the validator 14 in order to progressively create a semantic graph which represents the meaning of the sentence received as input.

Thus, device 2 offers with this function a fundamental building block to NLP because it is almost heuristic-free, and therefore offers a repeatable solution for creating a language sublayer which allows machines to access the language. semantic understanding of texts.

The device 2 finds a particularly effective application in the fields of “Question answering”.

Indeed, it allows first to analyze a text to establish its semantics, then to analyze a question in the same way to bring it closer to this semantics and provide the answer.

More generally, the device 2 is based on an existing database of semantic concepts, but is capable of enriching it thanks to its operation, unlike the methods based on automatic training.

The function of Figure 2 begins with an operation 200 with an Init () function.

An example of implementation of the function Init (is explained with FIG. 3.

The role of Init () is to initialize the loop of FIG. 2, and in particular to initialize the main object which is modified by the loops in order to obtain the semantic graph which is the result.

This object is called here "transient", because it is called to evolve many times, very quickly, in order to generate the semantic graph.

As we will see below, a transient is made up of characteristics.

Each characteristic can relate either to a particular token, in which case it is called a lexical characteristic, or to a relation between two lexical characteristics, in which case it is called a structural characteristic.

A structural feature may itself relate to the relationship between other structural features, which allows for the creation of complex semantic meanings.

10 This way of working also makes it possible to construct the semantic graph in a way that is both syntactic and semantic, which respects the nature of the language.

Finally, this will also make it possible to deal with the case of plurilingual sentences.

Indeed, if a part of a sentence is not resolved in a given language, it can be analyzed with the characteristics of another language in order to identify a subset of sentence in a second language which gives a semantic meaning to a sentence which did not have one in the first language.

This is completely new in the field of NLP, and is completely inaccessible to solutions based on automatic training.

The characteristics, whether they are lexical or structural, are the result of the application of linguistic objects called constructions.

These objects, well known in the field of linguistic analysis, have never found an effective computer application until today.

25 The constructions are based on a condition (s) / conclusion (s).

In other words, a construction is an object which behaves as follows: if lexical or structural characteristics meet the condition (s), then we will apply the conclusion (s) of the construction to them.

Thus, whenever a loop finds constructs that apply to the characteristics of the transient, it will complement them or create additional characteristic (s) in the transient if these do not already exist.

The semantic graph will thus be built step by step, starting with the words and assigning them the meaning they have in the sentence, then grouping them into 13 nominal groups, verbal groups, then nominal and verbal sentences, etc. ., until the sentence is fully defined.

As will be seen, the function of FIG. 2 is particularly powerful because it nevertheless makes it possible to re-evaluate during its execution the meaning associated with a word and to propagate the consequence of this change throughout the semantic graph.

The conditions and conclusions of constructions consist of fields whose values are fixed or variable.

10 The conditions typically relate to one or more of the fields of one or more of the following types: - Boundaries, which uses predicates of the same name to define the bounds of a group of elements, - SubUnits, which receives a list of elements designated by the element, 15 - LexicalCategory, which defines a lexical category (for example, noun, verb, common noun, proper noun, adverb, article, adjective; etc., - LexicalClass, which defines a lexical class ( for example, transitive verb, intransitive verb, auxiliary, determinant, etc.), - CategoryPhrasal, which defines groups of words among themselves (for example nominal sentence, verbal sentence, etc.), and - CategoryClausal, which defines a category clause and allows groups of words to be grouped together, for example nominal sentences, verbal sentences, etc.

The conclusions typically relate to one or more of the fields of one or more of the following types: - Referent, which defines a variable identifier which unites several elements at the same level, - Args, which defines an argument constituting links between elements, - Parent, which defines the parent element of the current element in the structure, 30 - Meaning, which defines a relational meaning (for example a temporal relation - simultaneous, temporal reference point, etc.), and - SemanticClass (), which defines a semantic class.

14 Conditions and conclusions can have many other fields, such as Form (relating to an attribute form), String (a character string), Precedes (a predicate linking two elements), Follow (a predicate linking two elements) , Interrogation (indicates that a group of items includes a verb, pronoun, or interrogative adverb), Preposition, PhraseType (for example, noun sentence, etc.), Number, Date, Person, SemanticValence (for example, actor and its element identifier, etc.), SyntacticValence (for example, subject or object and its element identifier), SyntaxicFunction (for example adjective, auxiliary, verb, etc.), Tense, Voice (active or passive), PassiveForm (yes or no), etc.

10 For example, the construct of type "Compound" can be defined as follows: {"id": "compound-names", "score": 0.0, 15 "type": "Phrasai", "description": "rules of compound names "," category ":" en "," group "-" stagel "," constructionClass ":" Classic ", 20" locks ": [{name 1" "name": "? nounl", "comprehension ": {25" name ":"? Nounl "," map ": {" Categori eLexi cale "30" name ":"? Noun2 ", 15" comprehension ":" name ":"? Noun2 "," map " : {"LexicalCategory": "[name]"} {"name": "? Compound", "comprehension": 10 "name": "? Compound", "map": {"Shape": ("name": "? v2", "map": {15 "Attached": "[[? noun 1 noun2]]" 20], "conclusions": [{"name": "? compound", "map": {25 " Agreement ":"? V5 "," Args ":" [? V9,? V7] "," Referrer ":"? Ref ', "SemanticClass": "identifier", "Parent": "? V10", 30 " Meaning ":" [[attribute,? V9,? V6]] "," SubUnits ":" [? Noun 1,? Noun2] // union "," Classical ":" [name, compound] "16 In this construct , the "?" indicates that it is a variable, which can be found at the fo is in the conditions and in the conclusions, serve to define the application of the construction, etc. It should also be noted that the function of Figure 2 has a quasi-recursive nature.

This makes its understanding less easy than a conventional sequential function and must be taken into account when reading what follows.

For this reason, Figures 8 to 17 are provided.

They do not in themselves teach the technique of device 2, but they make it possible to better understand how the latter explores all the possibilities to establish the semantic graph.

The [nit () function starts with an operation 300 in which an array Tk [] is initialized by a Lex () function.

The Lex () function performs the lexical analysis of a sentence received as input by the device 2, and provides an array Tk [] in which the sentence is split into standardized tokens.

The Tk [] array stores for each token the corresponding character string and a unique identifier of this token in the sentence.

This result is also stored in a Tst Stack [] array which will be described below.

The Lex () function implements a lexical analyzer to produce a sequence of standardized tokens.

The notion of normalization is in reference to the fact that certain words can be written in several forms (for example contractions in English language), or that certain characters must be deleted or grouped together.

Thus the lexical analyzer performs one or more of the following functions: - cleaning up annoying characters (footnote index, special characters, etc.), - splitting a text into sentences (thanks to delimiters such as the period, exclamation point, etc.), 17 - grouping of special expressions together (dates, etc.), - expansion of contract words (for example "don't" becomes "do not"), - splitting of the sentence in standardized tokens for processing by the device 2.

The lexical analyzer is not the subject of the invention and a person skilled in the art knows several solutions in the state of the art for implementing it.

Then, in an operation 310, shape predicates are initialized by an SFPO function which receives the array Tk [] as a variable.

The SFPO function takes the array Tk [] 10 and will produce predicates relating to the positions of the tokens with respect to each other.

Thus, for two [balloon] [red] tokens, the SFPO function creates a predicate of type Attached ([ball], [red]) and a predicate of type Precede ([ball], [red]).

These predicates therefore indicate that the token [balloon] is attached to the token [red] and that it directly precedes it.

In the example described here, the Predicate Precedes () is generated for all 15 tokens downstream of an upstream token.

The SFPO function is arranged to also generate predicates of type Borders (), which indicate start and end indices of a chain of several tokens.

The set of predicates thus produced is stored in memory 4, and is accessed to determine the application of conditions, as described below.

A loop is then launched to analyze each lexical characteristic of the transient in order to initialize the concept identifiers.

At this point, and this is the last time in the loop, we can identify characteristic of the transient and concept Thus, in an operation 320, the transient Tst is popped, and in an operation 330, a Find () function determines, for the current token, the syntactic identifier of the database of 25 concepts in memory 4 which has the highest frequency, and stores it in an Ltk [] array.

For example, if the token is "red", the Find () function will return the syntactic identifier associated with the adjective "red" rather than the one associated with the color "red", because the word is used more often as a. adjective as a noun.

Simultaneously, the Find () function assigns the most frequent concept identifier among the concept identifiers associated with this syntactic identifier.

Finally, the other syntactic identifiers are stored as options in an OC [] array. These options will be stored in the Tst Stack [] array in an operation 350 described below.

18 When all the tokens have been processed in this way, the Ltk [] array is supplied as an argument to a LexConstr () function in an operation 340.

The LexConstr () function returns a list of lexical constructions Cstr [] which will allow the lexical characteristics of the transient to be initialized.

This is done in an operation 350 in which a Merge () function receives the list of lexical constructs Cstr [] from operation 340 and the transient Tst, and combines them.

Here again, since this is the first operation, the combination is guaranteed, i.e. the condition of each lexical construction of the list Cstr [] is necessarily fulfilled by a characteristic of the transient Tst, since 'they were chosen specifically for this.

As indicated above, optional lexical features are generated and stored in a Tst Stack [] array.

These options can be explored when a problem is identified.

This is notably ensured by a Bck () function in an operation 290.

This will be described in more detail with the description of the combiner 12 in connection with FIG. 7.

At output, the transient Tst is therefore initialized with the lexical characteristics corresponding to each token, with the unique token identifier and the concept identifier which has been determined as being the most probable.

The Next () and 20 Precedes () predicates are also stored for the rest, and the function ends in an operation 399.

After the initialization operation 200, the loop of Figure 2 begins with an operation 205 in which a Max () function determines whether an exit condition related to an excessive number of loop executions is met.

This makes it possible to avoid getting stuck in a too long calculation loop (for example beyond 1000 iterations).

When this condition is met, the function of Figure 2 ends in an operation 299 with an error.

Alternatively, operation 205 can be omitted.

30 Then a new loop begins.

This loop begins in an operation 210 by the execution of a function Sem ().

In the example described here, the function Sem () is implemented by the rectifier 6, and FIG. 4 gives an exemplary embodiment.

19 From a general point of view, the goal pursued by the function Sem () is to analyze the current transient, which has just been enriched by the previous loop, and to see if it would not be appropriate to change a or more of the concept identifiers of the lexical characteristics taking into account the structural characteristics of the transient.

In other words, the function Sem () “shakes” the bag of concept identifiers available for each lexical characteristic, in order to determine if there is not a new concept which would give more meaning to the sentence described by the transient at this stage, from a semantic point of view.

Thus, in an operation 400, the rectifier 6 creates a Concept table [] which collects all the concept identifiers of the lexical characteristics of the current transient Tst.

Then, in an operation 410, an Extrap () function determines for each of these concept identifiers the list of concept identifiers which are linked to it in the concept database 15 and stores each list in an entry of a Candid table. [].

On the basis of the Candid [] array, an Obsety () function collects all the observations related to each of the concepts of each list of the Candid [] array and groups them together in an Obs [] array in an operation 420.

Finally, in an operation 430, an Infer () function uses the Candid [] array and the Obs [] array to modify the current transient, and the function ends with an operation 499.

More precisely, the function Infer () uses a non-monotonic probabilistic logical inference engine which calculates a multitude of cost functions according to the observations of the table Obs [] for each combination of a concept identifier by list of the table. Candid [].

In other words, the combination of the lists creates a combinatorial of concept identifiers, and the observations associated with each concept identifier are used to calculate a cost function from them.

The cost function is performed by determining a plurality of rules from the transient.

Then, these rules are evaluated on the basis of the observations by applying a multivalued logic operator which allows the problem to be linearized.

In the preferred version of the invention, it is the Lukasiewicz t-standard.

Alternatively, the operator could be the minimum t-norm or the Harnacher product.

These rules fall into three categories.

The first category of rules includes so-called frequency rules.

They are based on the frequency of the concept identifier which is associated with the lexical characteristics present in the transient.

These rules are expressed in the form CaractLex 10 (Group (Char)) => Concept (Group (Char)).

There is one rule per group of concept identifiers resulting from the ExtrapQ function.

The second category of rules includes so-called neighborhood rules, which are based on the links between the concept identifiers of the lexical features in the concept database.

For these rules, the concept database is explored from each group of concept identifiers, and searches for the “neighboring” concept identifiers in the other groups of concept identifiers, at a selected distance.

For example, if a first group contains the concept "Fred", and a second group contains the concept "hold", then these concepts are two concepts apart in the concept database.

Indeed, “Fred” is a proper name, associated with the concept “human being”, and the concept “human being” is itself related to the capacity “hold” since the human beings hold objects.

Thus, when a link is found between two concept identifiers of two distinct groups at a distance chosen in the concept database, a neighborhood rule is created.

These rules 25 are expressed in the form: CaractLex (Group (Carl)) & Caract Lex (Group e (C ar2)) & Li en (Group (Carl), Group (Char 2)) => Concept (Group (Car2) )) Finally, the third category of rules includes so-called structural rules, because they result from the semantic links established between the lexical characteristics within the transient.

These rules are therefore taken from the semantic attributes of the characteristics resulting from the structural constructions which link two lexical characteristics between them.

21 For example, if it has been identified that the lexical characteristic associated with the string "Fred" is linked to the lexical characteristic associated with the string "holds" by a semantic attribute of the "actor" type, then a corresponding rule replaces the second category rule which linked these two lexical characteristics.

These rules are therefore expressed in the form: CaractLex (Group (Carl)) & Caract Lex (Group e (C ar2)) 8.4.

AttSem (Group (Carl), Group (Car2)) Concept (Group (Car2)) The cost function instantiates these rules with the observations chosen according to the combination of concept identifiers of each group resulting from the ExtrapO function .

The optimization of this cost function makes it possible to determine the combination of each concept identifier of each list which offers the best semantics for the current transient, whose lexical characteristics are thus updated with the new concept identifiers which are considered as more relevant. 15 It therefore appears that, when the transient contains only lexical constructions, there are only rules of the first category and of the second category, and the cost function is based on the co-occurrence of the concept identifiers in the database of concepts.

Then, as the structural constructions add semantic links between the lexical characteristics in the transient, third category rules, much more discriminating, are introduced into the cost function and will strongly constrain it.

The Applicant has discovered that the use of a non-monotonic probabilistic logic inference engine makes it possible for the first time to offer a satisfactory result for implementing a method based on linguistic constructions.

Indeed, the function Sem (), thanks to the semantic adjustment that it offers at each execution of the loop, is fundamental in obtaining a favorable result.

The Applicant has also discovered that it was particularly advantageous to use an inference engine including an optimizer using the algorithm of multipliers with alternating directions (or ADMIM for “Alternating Direction Method of 22 Multipliers”).

Indeed, the use of such an optimizer makes it possible to reduce the costs of computation time by linearizing the problem, whereas the basic problem is of NP type, that is to say a combinatorial of all the variants of each list between them, multiplied by the quantity of observations for each member of each list.

Once the semantic adjustment has been made, the loop continues with a loop which will test each of the model structural constructions on all the characteristics of the transient and determine those which are likely to apply.

For this, in an operation 220 an array ConStr [] of model structural constructions is popped, and in an operation 230, the structural construction c from operation 220 is tested with all the characteristics of the current transient in a Match ° function. .

The Match () function is executed by the matcher 8 and FIG. 5 represents an example of implementation of this function.

Generally speaking, the Match () function analyzes each of the conditions of the structural construct c and constructs the tuples of conditions which satisfy the conditions of the structural construct.

Thus, in an operation 500 the list L [] of the conditions of the construction c is retrieved by means of a function Locks [].

Next, a loop is started in which this list is popped into an operation 510 and the characteristics of the current transient are each compared to the current condition.

For this, the transient is unstacked in an operation 520 and the corresponding characteristic f compared with the current condition 1 in an operation 530.

If characteristic f satisfies condition 1, then the next characteristic is tested by repeating operation 520.

If the characteristic f satisfies the condition, then an AddFt () function is executed in an operation 540.

The AddFt () function adds 25 in an array m [] all the groups of characteristics which satisfy a condition of the structural construction.

Thus, when a characteristic satisfies the condition 1, the function AddFt () determines those of the groups of the array m [] which are compatible with this characteristic taking into account all the conditions of the structural construction, and adds the characteristic f to all groups of characteristics 30 compatible.

After operation 540, or if operation 530 is negative, then the loop resumes with the next characteristic in operation 520.

23 When all the characteristics have been tested for the current condition 1, the loop is repeated with the next condition by repeating operation 510.

When all the conditions have been tested, in an operation 550, a function Rem () reduces the array m [] to check the groups of characteristics produced and to keep only those which are complete, that is to say which meet all of them. the conditions of structural construction.

These groups therefore form n-tuples of characteristics which are likely to have the structural construction applied, then the function ends in an operation 599.

Note that the function of Fig. 5 can be performed in many ways.

For example, the loop could be carried out in such a way as to exclude the testing of a characteristic as soon as it is detected that it does not satisfy a condition, for example by testing the value associated with it in the table m [] at the start of the loop.

Other variants could be considered.

The characteristics which correspond to the construction c are then stored in an array Con [] in an operation 235, then the loop resumes with the operation 210.

After all the model structural constructions have been tested, the Con [] array is tested in an operation 240.

If the array Con [] contains constructs, then operations 250 to 270 will analyze these constructs and choose the most relevant and establish a list of options for the case where the chosen constructs lead to a dead end in the following loops.

If this table is empty, then no structural construction can be applied to the characteristics of the transient.

As this test is inside a loop, it means that the semantic graph has not been fully resolved.

The fact that the array Con [] is empty indicates that it is not possible to complete the semantic graph.

It will therefore be necessary to explore the options established in the previous loop (s).

This will be done in a 290 operation.

In operation 250, an Ord () function processes the array Con [] and produces two arrays C2M [] and OC [].

Table C2M [] contains the list of most probable structural constructs, while table OC [] contains the list of options.

More precisely, the Ord () function initializes a first list by removing the first tuple from the Con [] array.

Then, it iterates through all the other tuple of the array Con [], and, each time a tuple concerns a characteristic of the first tuple, it inserts it in the first list and removes it from the array. Con [].

Once all the tuples have been scanned, the operation is repeated with the rest of the Con [] array, until it is empty.

This results in a C2M [] array containing the tuples which were used to generate the lists, and an OC [] array which contains the tuples which were gradually added to each list, as options to the n- corresponding tuple of table C21 \ 4].

In operation 260, arrays C2M [] and OC [] are processed in a Filt () function by filter 10 in order to determine if there is a reason to believe that the frequency choice 10 was not the correct one.

More precisely, in the case of tables C2M [] and OC [], we find ourselves in a situation where several structural constructions correspond to the same characteristic.

In other words, there is a lexical ambiguity, and the Filt () function will try to resolve it by semantic analysis.

FIG. 6 represents an example of implementation of the Filt () function.

The Filt () function is a loop which parses each conclusion of the array C2M [] by unstacking it in an operation 610.

Then, in an operation 620, an Opt () function generates an array N [] which receives the current conclusion c and all the options corresponding to it in the array OC [].

Then, in an operation 620, a Rules [] function 20 generates an array R [] of analysis rules.

For example in the case where two nominal groups are separated by a comma, it is necessary to determine whether it is a list or if it is an apposition.

For this, the structural constructions corresponding to the list on the one hand and to the apposition on the other hand are translated into two rules which are stored in the table R [] with the concept identifiers attached to the characteristic concerned.

These rules are in the example described here taken from the filtering rules in memory 4.

More precisely, the rules define predicates between the characteristics.

However, if for all the constructions of the array N [] there is no rule, then nothing is introduced in the array R [] for the construction c.

Then, in an operation 630, an observation table Obs [] is generated by a function Observ () in order to determine the observations related to the concept identifier of the characteristic concerned by the construction c and each of the rules of the table R [].

Finally, in an operation 640, the inference engine is again applied to the table R [] and the table Obs [].

If the array R [] is empty, then nothing is done and the order established with the arrays C2M [] and OC [] is maintained.

Otherwise, the inference engine makes it possible to semantically determine which of the constructs is the most semantically relevant.

The result is an array Con2 [] of the constructions to be applied to the current transient and an array OC [] of options.

When all the constructs in array C2M [] have been processed, the function ends in an operation 699.

Alternatively, the Filt [] function could be omitted.

After the Filt () function has been executed, the constructions chosen from the table Con2 [] are applied to the current transient in an operation 270 by the combiner 12 in a function Merge0.

FIG. 7 represents an example of implementation of the MergeQ function.

Again, the array Con2 [] is popped in a 700 operation.

Then, in an operation 710, the OC array [] is also popped in order to store the corresponding options.

Then in an operation 720, the current transient is stored with the options of operation 710 in an operation 720.

This operation is crucial because it is this which will make it possible to travel in the most complete and efficient way in operation 290.

There is therefore a transient coupled to a list of construction options which is generated before each application of a construction, and as many fallback solutions in the event of failure.

In other words, the transients introduced into the table Tst Stack [] in operation 720 are all different from each other since between two applications of operation 720, the transient is modified in operation 730.

Thus, the table Tst Stack [] contains the detail of all the constructions applied to the transient, one by one, and classified temporally by the very nature of the loop.

Finally, in an operation 730, the conclusions of the construction unstacked in operation 700 are applied to the current transient.

To do this, if these conclusions apply to an existing characteristic, then this characteristic is updated.

Otherwise, a new characteristic is created in the transient.

When all the constructs have been applied to the transient, the Merge () function ends in a 799 operation.

26 Once the Merge () function is complete, an operation 280 determines with a Goal () function whether the sentence has been fully resolved and whether the semantic graph is finished.

For this, the Goal () function determines if all the tokens have been filled with a concept identifier, if all the characteristics define between them a tree of which all the branches are connected to each other, that is to say if it It is possible to reach all the characteristics of the tree from each characteristic, and finally if there is no cycle in the generated structure.

If so, then the function of Figure 2 ends in operation 299.

Otherwise, the loop resumes with operation 205 to process the current transient.

Operation 290 consists of unstacking the Tst Stack [] array fed by the execution of operation 720, and resuming the loop at operation 270 using the first option.

In the foregoing, it therefore appears that the function of FIG. 2 is a systematic algorithm based on known data and almost devoid of heuristics (the filtering rules could be qualified as heuristic, but they are optional).

This demonstrates the repeatability of the treatment of device 2.

In addition, the processing of device 2 is both syntactic by applying structural constructs and semantic by using rectifier 6 and filter 10.

It is this completely new approach, made possible through the use of a non-monotonic probabilistic logic inference engine which makes possible the application of a construct-based model and which produces a semantic understanding.

Figures 8 to 17 are given by way of example in order to help better understand the operation of the loop of Figure 2 and in particular operation 290 on the sentence "Freds holds a small match".

In a first step represented in FIG. 8, the device 2 initializes the transient with the lexical constructions in the order of their frequency in the database of 30 concepts: “Holds, verb” “small, adjective”, “a, construction article ”,“ match, verb ”and“ Fred, proper noun ”.

Alternatively, the lexical constructs could be ordered by their order within the sentence, or in an order chosen so as to determine a degree of confidence in the choice of the first lexical constructs, and to place the lexical constructs towards the end of the transient. which have the lowest confidence.

At the next loop, shown in Figure 9, an adjectival sentence has been created above "small", a nominal sentence above "Fred", and two verbal sentences respectively above "holds" and "Match".

In the loop shown in Figure 10, a clause links the nominal phrase of "Fred" and the verbal phrase of "holds", then subsequent loops fail.

This failure goes back through the transients of the Tst Stack [] array until it determines that the error relates to the lexical construction of "match".

As "Fred" ranks lower than "match", the transient is reduced to figure 11 "match, common name" and "Fred, proper name" are then tested as an option, then figures 12 to 15 show the development. constructions on this basis.

Finally, with figure 16 the verbal sentence is concluded and the structure is closed with an affirmative clause shown in figure 17.

In order to better understand the rules of the first, second and third category, they will be explained for example 20 of FIGS. 16 and 17.

In FIG. 16, a single semantic link is established between the lexical characteristics, the attribute link between “small” and “match”.

Thus, the ruleset produced will be: CaractLex (Group (Fred)) => Concept (Group (Fred)) 25 CaractLex (Group (hold)) => Concept (Group (hold)) CaractLex (Group (a)) = > Concept (Group (a)) CaractLex (Group (small)) => Concept (Group (small)) CaractLex (Group (match)) => Concept (Group (match)) CaractLex (Group (Fred)) & taractLex (Group (hol d)) & Li en (Group (Fred), Group (hol d 30)) Concept (Group (hold)) CaractLex (Group (hold)) & CaractLex (Group (match)) & Link (Group (hold), Group ( ma tch)) => Concept (Group (match)) 28 CaractLex (Group (small)) & CaractLex (Group (match)) & Attrib (Group (small), Group (match)) => Concept (Group (match)) On FIG. 17, two other semantic links are established between the lexical characteristics: the actor link from "Fred" to "hold" and the theme link between "hold" and "match".

Thus, two rules of the third category are added and replace the rules of the second category in figure 16: 10 CaractLex (Group (Fred)) & CaractLex (Group (hold)) & Actor (Group (Fred), Group (ho 1d)) = > Concept (Group (hold)) CaractLex (Group (hold)) & CaractLex (Group (match)) & Theme (Group (hold), Group (match)) => Concept (Group (match)) 15 Figure 18 represents the transient complete associated with the example of FIG. 17, and FIG. 19 represents the semantic graph which corresponds to it.

Thus, it appears that the lexical characteristics are linked together by semantic attributes in non-lexical characteristics.

This is shown in Figure 18 by the connection of elements of type? ArgXX and? RefYY or of type? ArgXX and? ArgYY.

20 Thus, the following links appear: Actor? Arg27? Ref28, Theme? Arg27? Ref28, Attr? Arg15? Arg17, NonldentifiableReferent? Arg 2? Arg17.

FIG. 19 translates these semantic links, and makes it possible to represent the semantic graph which describes the meaning of the sentence, as produced by the device 2.

Device 2 therefore produces a final transient which has a tree structure which contains all the links, both syntactic and semantic, of the sentence produced as a whole.

This tree makes it possible to produce a semantic graph which gives the meaning of the sentence.

This is really fundamental because it is possible to automatically create, without human intervention, a semantic description layer of a text, which therefore becomes searchable.

Moreover, by the very nature of the device 2, this semantic description layer can be enriched incrementally, by providing new sentences, without having to redo all the training.

In addition, this layer is searchable, and makes it possible to analyze what the device 2 comprises,

Claims (15)

CLAIMS 1 Device for automatic computer word processing, comprising a memory (4) arranged to receive text data to be analyzed in the form of tokens each comprising a character string and a unique token identifier, a database of concepts associating character strings and concept identifiers, at least some of the concept identifiers being associated with one another, model lexical construct data and model structural construct data, each comprising one or more conditions of application to a characteristic and a or several conclusions constituting elements to be applied to a characteristic, and a database of observations associating at least two concept identifiers, a type of relation and an observation value indicating a probability of veracity of the type of relation between the au at least two concept identifiers, the device (2) being arranged to work d e repetitively on a transient comprising lexical characteristics and structural characteristics produced by applying model lexical constructions and model structural constructions, the transient being initialized with lexical characteristics comprising for each token a concept identifier whose frequency is the most important in the database of concepts and which is associated with the character string of the token, the device (2) further comprising a rectifier (6) arranged to determine for each lexical characteristic of a transient a list of concept identifiers associated with the concept identifier of this lexical characteristic, to determine a set of observations corresponding to the concept identifiers of the lists thus determined, and to apply a non-monotonic probabilistic logical inference engine to determine the concept identifier of each list such as observation values s associated with these concept identifiers minimize a cost function defined by applying a multivalued logic operator to one or more rules drawn from the content of the characteristics of the transient and instantiated with corresponding observation values, and to replace the identifiers of concept of the lexical characteristics of the transient by the concept identifiers thus determined, a matcher (8) arranged to determine among the model structural constructions those whose condition (s) apply to one or more of the characteristics of the transient, and to return the list of structural constructions with the characteristic (s) to which their conditions apply, the device (2) being further arranged to classify the structural constructions associated with each characteristic by frequency of use, the first structural construction having to be applied to the 10 transient and the others forming a list of options, and a combiner (12) arranged to perform sequentially the selection of a structural construct to be applied to the transient, storing a copy of the transient with the list of options associated with the structural construct , and applying the structural construction to be applied to the transient to the characteristic (s) of the transient with which this structural construction has been associated by the matcher (8), and to repeat this sequential execution on the transient thus modified with the structural construction to be applied to the next transient, the device (2) being further arranged to determine, after the execution of the combiner (12), whether the characteristics of the produced transient define a tree of which all the nodes are linked together and without cycle, to return this shaft if it is the case, and to repeat the execution of the rectifier (6), of the pairer (8) and of the combi nator (12) on the last transient produced otherwise, the device (2) being further arranged, when the pairer (8) does not return any structural construction to be applied to the transient, to replace the current transient by the copy of the transient. more recent, and to execute the combiner (12) with the first construct of the list of options as a structural construct to be applied to the transient. 30 2. Device according to claim 1, in which the non-monotonic probabilistic logic inference engine comprises an optimizer using the algorithm of multipliers with alternating directions. 32 3. Device according to claim 1 or 2, further comprising a filter (10) arranged to determine, for each structural construction to be applied and the associated list of options a set of rules, to determine a set of observations from of the set of rules and of the concept identifier (s) associated with the characteristic to which the structural construction to be applied must be applied, and to apply the non-monotonic probabilistic logical inference engine with the set of rules and the set of observation to determine the structural construction to be applied to the transient and the list of options. 10 4. Device according to one of the preceding claims, wherein the rectifier (6) defines frequency rules from lexical characteristics of the transient, neighborhood rules from the concept identifiers of the lexical characteristics of the transient and the concept identifiers which are associated with them in the concept database at a chosen distance, and structural rules drawn from semantic attributes of non-lexical characteristics of the transient linking two lexical characteristics together. 5. Device according to one of claims 1 to 4, wherein the combiner (12) is arranged to store only the copy of the transient with the list of options 20 associated with the structural construction, and in that, when the pairer (8) does not return any structural construct to apply to the transient, to replace the current transient with the most recent copy of the transient, and to execute the combiner (12) with the first construct of the option list as a construct structural to be applied to the transient, then repeating the application of the rectifier (6), the matcher (8) and the combiner (12) with the resulting transient. Device according to one of claims 1 to 4, wherein the combiner (12) is arranged to store the remaining structural constructs to be applied to the transient together with the copy of the transient with the list of options associated with. the structural construct, and in that when the matcher (8) does not return any structural construct to apply to the transient, to replace the current transient with the most recent copy of the transient, and to execute the combiner (12) with the first build the option list as a structural construct to apply to the transient along with the remaining structural constructs, then repeating the application of the rectifier (6), matcher (8) and combiner (12) with the resulting transient. 7. Device according to one of the preceding claims, arranged to analyze the returned tree and to produce a semantic graph whose nodes are formed by the lexical characteristics and their concept identifier, and therefore the links are defined by the semantic attributes of non-lexical features linking two lexical features together. 8. Device according to one of the preceding claims, in which the inference engine is arranged to apply a multivalued logic operator chosen from the group comprising the Lukasiewicz t-norm, the minimum t-norm and the Harnacher product. . 9. A computer-implemented automatic text processing method, comprising the following operations: a) receiving text data to be analyzed in the form of tokens each comprising a character string and a unique token identifier, a database. concept data associating character strings and concept identifiers, at least some of the concept identifiers being associated with one another, model lexical construct data and model structural construct data, each comprising one or more application conditions to a characteristic and one or more conclusions constituting elements to be applied to a characteristic, and a database of observations associating at least two concept identifiers, a type of relation and an observation value indicating a probability of the veracity of the type of relation between the at least two identifiers of concept, b) to initialize a transient being able to co understand lexical characteristics and structural characteristics produced by applying model lexical constructions and model structural constructions, with 34 lexical characteristics comprising for each token a concept identifier whose frequency is the most important in the concept database and which is associated with the character string of the token, c) work repeatedly on the transient by repeating the following successive operations: cl) determining for each lexical characteristic of a transient a list of concept identifiers associated with the identifier concept of this lexical characteristic, to determine a set of observations corresponding to the concept identifiers of the lists 10 thus determined, and to apply a non-monotonic probabilistic logical inference engine to determine the concept identifier of each list such as the observation values associated with these identifia Concept nts minimize a defined cost function by applying a multivalued logic operator to one or more rules taken from the content of the characteristics of the transient and instantiated with corresponding observation values, and to replace the concept identifiers of the lexical characteristics of the transient by the concept identifiers thus determined, c2) determine among the model structural constructions those whose 20 or conditions apply to one or more of the characteristics of the transient, and to return the list of structural constructions with the characteristic or characteristics to which their conditions apply, c3) classify the structural constructs associated with each characteristic by frequency of use, the first structural construct to be applied to the transient and the others forming a list of options, and c4) sequentially perform the selecting a struct construct urel to be applied to the transient, storing a copy of the transient with the list of options associated with the structural construction, and applying the structural construction to be applied to the transient to the characteristic (s) of the transient at which this structural construction has been associated by the pairer (8), and to repeat this sequential execution on the transient thus modified with the structural construction to be applied to the following transient, c5) determine, after the execution of the combiner (12), if the characteristics of the produced transient define a tree of which all the nodes are linked to each other and without a cycle, to return this tree if this is the case, and to repeat operations cl) to c5) on the current transient otherwise, 10 c6) if operation c2) does not return any structural construction to apply to the transient, replace the current transient by the copy of the most recent transient, execute operation c5) with the first construction of the list of options as a structural construction to be applied to the transient. 15 10. The method of claim 9, wherein the application of the non-monotonic probabilistic logic inference engine comprises the application of an optimizer using the algorithm of multipliers with alternating directions. 20 11. The method of claim 9 or 10, further comprising, between operation c3) and operation c4): c7) determining, for each structural construction to be applied and the associated list of options, a set of rules, determine a set of observations from the rule set and the concept identifier (s) associated with the characteristic to which the structural construct to be applied is to be applied, and apply the non-monotonic probabilistic logic inference engine with the rule set and the observation set to determine the structural construction to be applied to the transient and the list of options. 30 12. Method according to one of claims 9 to 11, wherein the operation c) comprises the definition of frequency rules from lexical characteristics 36 of the transient, of neighborhood rules from the concept identifiers of the lexical characteristics of the transient. and concept identifiers associated with them in the concept database at a selected distance, and structural rules derived from semantic attributes of non-lexical characteristics of the transient linking two lexical characteristics to one another. 13. Method according to one of claims 9 to 12, wherein the operation c4) store only the copy of the transient with the list of options associated with the structural construction, and wherein, after the execution of the operation c7), operations 10 c1) to c7) are repeated on the current transient. 14. Method according to one of claims 9 to 12, wherein operation c4) stores the remaining structural constructs to be applied to the transient along with the copy of the transient with the list of options associated with the structural construct. , and in which, after the execution of the operation c7), the operation c5) is applied with the remaining structural constructions, then repeating the application of the rectifier (6), then, the operations c1) to c7) are repeated on the current transient. 20 15. Method according to one of claims 9 to 14, further comprising the following operation: d) analyzing the tree returned by operation c6) and producing a semantic graph whose nodes are formed by the lexical characteristics and their. concept identifier, and therefore the links are defined by the semantic attributes of 25 non-lexical characteristics linking together two lexical characteristics.