Patent Grant
9626353
This application claims the benefit of priority under 35 USC 119 to Russian Patent Application No. 2014101124, filed Jan. 15, 2014; the disclosure of which is incorporated herein by reference.
The present invention is related to computer-implemented analysis of syntactic structure of a text.
There exist multiple systems for natural language processing of written texts in human languages. Since there is substantial ambiguity in structure of a human language, such texts are not easy to analyze. The number of potential syntactic links in a sentence may be very large.
As a result there are a large number of potential syntactic constructions that can be considered when analyzing a text. The more expanded the analysis system is and the more possibilities it considers, the more variants have to be analyzed. An attempt to analyze all the links will lead to a combinatorial explosion. An attempt to look at only the most probable combinations may lead to a loss of meaning. As a result, the problem of optimizing the process of syntactic analysis to minimize required time while maintaining integrity of the result is crucial. The present invention is offering a method of evaluating all possible syntactic combination quickly and without losing true meaning of the text.
This document discloses methods, techniques, and systems for analyzing a source sentence by identifying a sentence, identifying a graph of generalized constituents for the sentence based on rough syntactic analysis of the lexical-morphological structure of the sentence, where the graph of generalized constituents comprises arcs and nodes; filtering the arcs of the graph of generalized constituents using a combination classifier. In some implementations the combination classifier comprises a tree classifier and at least one linear classifier; and identifying a syntactic structure of the source sentence by performing precise syntactic analysis based on the graph of generalized constituents with the filtered arcs.
In some implementations, the tree classifier divides the arcs into clusters based on a predetermined set of features. In further implementations, the predetermined set of featured is based on parallel analysis of a two-language text corpus. In further implementations, the order of the features from the predetermined set of features is determined based on entropy measures of the features. In some implementations, the tree classifier is based on Iterative Dichotomiser 3 (ID3) algorithm. In some implementations, weights for the linear classifier are based on parallel analysis of a two-language untagged text corpus.
This document also discloses methods, techniques, and systems for classifier training including performing parallel analysis of a two-language text corpus, identifying two parallel sentences from the untagged two-language text corpus; preparing two corresponding graphs of generalized constituents for each of the two parallel sentences; generating at least one syntactic tree based on the graphs of generalized constituents for the two parallel sentences, where an arc of the graph of generalized constituents is included in the syntactic tree based on whether the second corresponding graph of generalized constituents comprises a matching arc; and training a combination classifier based on the generated syntactic tree.
Additional purposes, characteristics, and advantages of this invention will be disclosed in the following description of the embodiment of the invention with reference to the attached drawings, in which:
In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the invention. It will be apparent, however, to one skilled in the art that the invention can be practiced without these specific details. In other instances, structures and devices are shown only in block diagram form in order to avoid obscuring the invention.
Reference in this specification to “one implementation” or “an implementation” means that a particular feature, structure, or characteristic described in connection with the implementation is included in at least one implementation of the invention. The appearances of the phrase “in one implementation” in various places in the specification are not necessarily all referring to the same implementation, nor are separate or alternative implementations mutually exclusive of other implementations. Moreover, various features are described which may be exhibited by some implementations and not by others. Similarly, various requirements are described which may be requirements for some implementations but not other implementations.
A graph of generalized constituents of a sentence is a representation of all possible relationships between words in the sentence. Nodes of the graph represent constituents of the sentence. A constituent is a word or a group of words that functions as a single unit within a hierarchical structure. Each constituent in the sentence is represented in the graph of generalized constituents by a node. Generalized, for example, by a part of speech, nodes may represent many variants of lexical and grammatical values of words they represent. The arcs between the nodes represent surface (syntactic) slots expressing various types of relationships between the lexical values. Constituent has a core represented by word with its morphological meaning and semantic ancestor (semantic class), and can have child constituent of a lower level.
The graph of generalized constituents 109 is build from the lexical-morphological structure 105 of the sentence 101. All the possible surface syntactic models for each element of lexical-morphological structure 105 are applied, and all the possible constituents are built and generalized. A rough syntactic analyzer or its equivalents is adapted to generate the graph of the generalized constituents 109 from the lexical-morphological structure 105 using surface models, deep models, and a lexical-semantic dictionary.
In one implementation, all the possible syntactic descriptions and syntactic structures for the sentence 101 are considered and generalized. As a result, the graph of the generalized constituents 109 is built, having each constituent generalized from all the possible constituents for each element of the sentence 101, and building generalized constituents are performed for all the elements of the sentence 101. The graph of generalized constituents 109 reflects, on a surface model level, all hypothetical possible syntactic relationships between the words of the sentence 101.
To build all possible constituents, every element of the source sentence 101 which is not a space or a punctuator is viewed as a potential core of a constituent. The building of the graph of generalized constituents 109 starts with building those constituents which have only the core word form and further expands to build constituents of the next level by including neighboring constituents. For each pair of “lexical meaning-grammatical value” which corresponds to a non-trivial arc of lexical-morphological structure, its surface model is initialized, attempting to attach other constituents in the surface slots of the syntforms of its surface model to the right and the left neighboring constituents. If an appropriate syntform is found in the surface model of the corresponding lexical meaning, the selected lexical meaning may be the core of a new constituent.
The graph of generalized constituents 109 is first built as a tree, from the leaves to the root (bottom up). Building of additional constituents is performed bottom-up by attaching child constituents to parent constituents via filling the surface slots of parent constituents to cover all the initial lexical units of the sentence 101.
The root of the tree is the main clause, represented by a special constituent corresponding to various types of maximal units of a text analysis (complete sentences, enumerations, titles, etc.). The core of the main clause is generally a predicate. During this process, the tree actually becomes a graph, because lower-level constituents (the leaves) can be included into different upper-level constituents (the root).
Some of the constituents which are built for the same element of the lexical-morphological structure may be generalized to obtain generalized constituents. Constituents are generalized by the lexical meanings, by the grammatical values, for example, by parts of speech, by their boundaries, among others. Constituents are generalized by the boundaries, since there may be very different syntactic relationships in the sentence, and the same word may be included in different constituents. As a result of the rough syntactic analysis 107, the graph 109 of generalized constituents is built which represents the sentence 101.
The preliminary assembly 201 of the constituents during the rough syntactic analysis 107 is performed on the lexical-morphological structure 105 of the sentence to be analyzed, including certain word groups, the words in brackets, inverted commas, etc. Only one word in the group (the constituent core) can attach or be attached to constituents outside the group. The preliminary assembly 201 is performed early during the rough syntactic analysis 107 before building of generalized constituents 203 and building of the generalized constituent models 207 to cover all the boundaries of the whole sentence.
Building of generalized constituents 203 generally require that all possible pairs of the lexical meaning and the grammatical value are found or assigned for each of the constituents and attach the surface slots of the child constituents thereof to each of the constituents. Lexical units of the sentence 101 can form into core constituents at bottom levels. Each constituent can be attached to a constituent at a higher level if the surface slots of the constituent at the higher level can be filled. Thus, the constituents are further expanded to include the neighboring constituents built at previous constituent building process until all of the possible constituents have been built to cover the entire sentence.
During rough syntactic analysis 107, the number of the different constituents which may be built and the syntactic relationships among them are considerably large, some of the surface models of the constituents are chosen to be filtered through the process of filtering 205 prior to and after the building the constituents in order to reduce the number of the different constituents to be considered.
The filtering 205 during the rough syntactic analysis 107 include filtering of a set of syntforms 215 performed prior to and during the building of generalized constituents 203. The syntforms 215 and the surface slots are filtered a priori, and constituents are filtered after they are built. The process of the filtering 205 distills out a number of syntforms 215 including, but not limited to, those syntforms that do not correspond to the grammatical values of the constituent, those syntforms where none of the core slots can be filled, those syntforms with special slots which describe grammatical movement, among others. A special slot, such as relativization and question, presupposing a special lexeme (relative or interrogative pronoun), is filtered out if the special lexeme is not present in the sentence.
In general, the syntax forms (syntforms 215) which do not have fillers for at least one surface slot can be filtered and discarded. In addition, those lexical meanings which do not have syntforms 215 with filled surface slots are filtered and discarded. The rough syntactic analysis 107 is impossible to succeed if there is no syntform and no filled surface slot, and as such the filtering 205 is performed.
Once all possible constituents are built, the generalization procedure is performed for building of the generalized constituents 207. All possible homonyms and all possible meanings for elements of the source sentence which are capable of being present in the same part of a speech are condensed and generalized, and all possible constituents built in this fashion are condensed into generalized constituents 217.
A generalized constituent 217 describes all the constituents with all the possible boundaries in a given source sentence which have a word form as the core constituent and various lexical meanings of this word form. Since the constituents are generalized, a single constituent for each lexical meaning corresponding to each entity of a sentence, including homonyms, is built, and their syntactic forms may be analyzed simultaneously.
The building of generalized constituent models 207 is performed and a set of models of generalized constituents 219 having generalized models of all generalized lexemes are built. A generalized constituent model of a lexeme contains a generalized deep model and a generalized surface model. A generalized deep model of a lexeme includes the list of all of the deep slots which have the same lexical meaning for a lexeme, together with the descriptions of all the requirements for the fillers of the deep slots. A generalized surface model contains information about the syntforms 215, where the lexeme may occur, about the surface slots, about the diatheses correspondences between surface slots and deep slots, and about the linear order description.
The syntforms 215 and the surface slots that are significant for this lexeme are selected with the help of the bit-mask. In addition, models of the generalized constituents are built because a constituent is generalized not only by lexical meanings and syntactic forms of its core, but also by the fragments it fills. The use of the models of the generalized constituents reduces the number of wrong relationships and helps to optimize the process to extract a syntactic tree so that all possible boundaries are considered.
As shown in
The building of the graph of the generalized constituents 209 is performed. The graph of generalized constituents 109 which describes all possible syntactic structures of the entire sentence is built by linking and assembling the generalized constituents 217 to each other. The building of the graph of the generalized constituents 209 is organized via generating and processing of the queue of requests to attach one constituent to another constituent. In general, contact pairs of constituents representing contact groups of words in the sentence can be included in the request queue.
Links in the graph 109 represent filled surface slots of constituent cores. Slot names are displayed on the arrows of the graph. Any constituent is formed by a lexeme-core which may have outgoing named arrows which denotes surface slots filled by child constituents. An incoming arrow means attaching this constituent to a surface slot of another constituent. The graph 109 is so complicated and has so many arrows, because it shows all relationships which can be established between constituents of the sentence “This boy is smart, he'll do well in life.” Among them there are many relationships in the graph 109 which, however, will be discarded. A value of said rough rating is saved by each arrow denoting a filled surface slot. Surface slot and relationships with high rating scores are selected hierarchically during the next step of syntactic analysis.
Often several arrows may connect the same pairs of constituents. It means that there are different acceptable surface models for this pair of constituents, and several surface slots of the parent constituent may be independently filled by this child constituent. So, three surface slots named Idiomatic_Adverbial 301, Modifier_Adverbial 303 and AdjunctTime 305 of the parent constituent “do<Verb>” 307 may be independently filled by the child constituent “well<Adverb>” 309 in accordance with surface model of the constituent “do<Verb>”. Thus, roughly speaking “do<Verb>” 307+“well<Adverb>” 309 form a new constituent with the core “do<Verb>” which is attached to another parent constituent, for example, to #NormalSentence<Clause> 311 in the surface slot Verb 313, and to “child<Noun&Pronoun>” 315 in the surface slot RelativClause_DirectFinite 317 The marked element #NormalSentence<Clause> 311, being the “root”, corresponds to the whole sentence.
The process of rough syntactic analysis and lexical-morphological analysis are described, for example, in U.S. Pat. No. 8,078,450, <<Method and system for analyzing various languages and constructing language—independent semantic structures>> filed Oct. 10, 2006.
Returning to
The arc filtering 111 of the graph of generalized constituents 109 is performed by a combination classifier 117.
A classifier is an algorithm performing classification of a set of items. Classification is a process of dividing a set of items into classes in some way.
In some implementations, the combination classifier 117 is a combination of a linear classifier and a tree classifier. The combination classifier 117 is organized as a decision tree with linear classifiers 405 at the nodes. The combination classifier 117 filters arcs using two types of features: symbolic and numerical.
Symbolic features are identifiers corresponding to elements of the description. In some implementations, the following list and combinations of features are used as symbolic features:
a surface slot, which is a type of syntactic description of a deep slot. For example, a surface slot could be an indication of the part of the sentence that is a word such as: subject, object, predicate, etc. or can play role of attributive, locative etc.
a length of a link (with a character count), which is a number of steps that one can use to get through the syntactic tree from one word to another, taking the punctuation into account;
syntparadigm which is a combination of syntactic forms that describe the surface implementation of a deep model for a given core.
In some implementations, the numerical features may include but are not limited to various a priori and statistical ratings, such as ratings of lexical meanings, ratings of fillers, punctuation ratings, or ratings of semantic descriptions.
As illustrated in
When filtration process is launched, the tree classifier 403 initially divides arcs into clusters using symbolic features 409 selected as primary features during a classifier training 411. The number of steps for clustering is determined by the number of symbolic features 409 selected during the classifier training 411 and the amount of data that satisfies them.
The tree classifier 403 uses a decision tree to classify the arcs of the graph of generalized constituents 401. Each node of the tree corresponds to one of the features of the set of symbolic features 409.
In some implementations, the order in which the symbolic features 409 are applied to the set of arcs is determined by an entropy (or information gain) of the corresponding features. During each iteration the classifier calculates entropy H(S) for every unused feature in the set of symbolic features 409, where S is the set of arcs of the graph of generalized constituents 401. H(S) should be understood as a measure of uncertainty of the information that we obtain from the set S based on each feature, and it selects the feature with the smallest entropy Hmin(S). This measure corresponds to the maximum information gain for this feature, that is, the difference in entropy before and after the set S is split in accordance with this feature. In other words, information gain shows how many uncertainties in S we remove, after splitting S in accordance with this feature. The feature with the smallest entropy (or largest information gain) is selected, and the set of arcs of the graph of generalized constituents 401 is split into two subsets such that the elements of one subset have the feature with Hmin(S) and those of the other set don't.
The algorithm of the tree classifier 403 continues recursively on each subset, only looking at those elements not previously chosen. Recursion on a subset may stop in one of the following cases:
each element of the subset belongs to one of the clusters (+ or −);
there are no more features to be selected, but the examples still do not belong to the same class (some are + and some are −), then the node is turned into a leaf and labeled with the most common class of the examples in the subset;
there are no examples in the subset corresponding to the feature chosen at this step. Then a leaf is created, and labeled with the most common class of the examples in the parent set.
The algorithm used by the tree classifier includes the following main steps:
computing the entropy for each feature of the data set S;
splitting set S into subsets based on the feature with the minimum entropy (or, in other words, with the maximum information gain);
creating nodes for the decision tree that contain each feature;
recursion by subsets using the remaining features.
The algorithm can return an excessive number of features. In order to avoid that, a limitation is set on the size of the tree. It is more effective to create smaller trees than larger ones. When this is done, the method frequently leads to construction of small trees, but not the smallest possible ones.
Entropy H(S) is a measurement of the uncertainty in a set S. H(S)=−ΣxεX p(x)log2 p(x) where S is the current set for which entropy is computed at each iteration of the algorithm; x is the selection of classes in set S, and p(x) is the ratio between the number of elements in class x and the number of elements in set S.
When H(S)=0, set S is ideally classified in the sense that all elements of S belong to a single class.
The information gain IG(i) is a measure of the reduction of uncertainty in set S after it split according to the feature i.
IG(i)=H(S)−ΣtεTp(t)H(t), where H(S) is the entropy of selection S, T is the subset produced by breaking down S based on feature i, S=UtεTt; p(t) is the ratio of the number of elements in t to the number of elements in S, and H(t) is the entropy of subset t.
In one implementation, the features in the tree classifier 403 are ordered. In some implementations the tree classifier 403 is based on an Iterative Dichotomiser 3 (ID3) algorithm, CART algorithm, or a C4.5 algorithm.
In some implementations, the combination classifier 117 comprises a linear classifier 405. In other implementations the combination classifier 117 comprises two or more linear classifiers 405.
A linear classifier is a classifier that makes a classifying decision based on a linear combination of characteristics of the items being classified. The characteristics of an item are usually represented as a vector. If a characteristic vector {right arrow over (x)} is entered into the classifier, the output rating looks like y=f({right arrow over (w)},{right arrow over (x)})=f(Σjwjxj) where {right arrow over (w)} is the weighting vector and f is the function that transfers the scalar multiplication of vectors to the desired output value. The weighting vector {right arrow over (w)} or a set of weights are computed while the classifier is being trained using marked training examples. Usually f is a simple function that puts all of the values larger than some threshold in the first class and all the rest in the second. More complex functions might, for example, provide the probability that an element belongs to a specific class. For tasks with a distribution between two classes, one can interpret the work of a linear classifier as a breakdown of a multi-dimensional space of input data using a hyperplane. All of the points on one side of the plane are classified as “yes” and the others as “no”.
In some implementations, the linear classifier 405 receives as an input a set of arcs in a graph of generalized constituents 401 of the sentence 101, which are first clustered using the tree classifier 403. The linear classifier 405 makes a decision for each cluster about whether the cluster of arcs is “good” or “bad.” The linear classier 405 makes this decision about a cluster by calculating an output rating for the cluster based on individual characteristics of the cluster being analyzed and a vector of weights 413 of the classifier, which is provided to the combination classifier 117 and the linear classifier 405 through classifier training 411.
After decisions for all clusters have been reached, the linear classifier 405 combines the decisions made for each cluster and finally applies a Bayesian formula to the decisions. The probabilities that arcs will fall into clusters of “good” and “bad” arcs are compared. These probabilities are computed based on an assumption of a normal distribution of projections of arcs in the space of numerical features onto a surface normal to a dividing plane. The projection onto a surface normal is computed as
x=X*W/|W|,
where X is the vector for numerical features of links, and
The probability that links fall into the clusters of “good” and “bad” links can be computed using the following formula:
Pgood=p(x)good/(p(x)good+p(x)bad),
For each combination of symbolic features of arcs, we know the percentage of “good” arcs. So the full probability that an arc is “good” is calculated as
Pgood=Pgood(symb)*Pgood(x).
The symbolic features 409 for the tree classifier 403 and the weights 413 for the linear classifier 405 are determined by classifier training 411. In some implementations, the classifier training is performed by parallel analysis 415 of a two-language text corpus.
The two-language text corpus is an untagged text corpus in two languages, such as Russian and English, where each sentence in one language has a corresponding sentence is the second language, with one sentence being an accurate translation of the other.
During parallel training 415, two corresponding sentences in both languages are first parsed, i.e. for each sentence graphs of generalized constituents are constructed. Then the system compares the two graphs of generalized constituents of corresponding sentences, and generates a tree with arcs from the two graphs. Only the arcs having the best matches in terms of semantic classes between the two graphs are selected for the tree. For example, if one graph has a link “” in which the lexeme “
” belongs to the semantic class “TO SHOW” and the lexeme “
” belongs to the class “FILE”, while in the other graph there is an equivalent link “show file” for which “show” belongs to the class “TO SHOW” and “file” is from the class “FILE”, this link is considered “good”. If, for example, for “show file” from the second graph, the lexeme “show” is a noun and belongs to the class “PERFORMANCE” and the lexeme “file” is a verb from the class “TO FILE”, this link is “bad”. Since the corpus used in the analysis is untagged, i.e., the translator did not indicate “good” or “bad” arcs manually, in the general case we can construct more than one such tree structures for each pair of graphs corresponding to a pair of sentences. As a result of the parallel analysis 411, based on the pairs of graphs of generalized constituents, the system generates tree structures containing only “good” arcs.
Based on the parallel analysis 415 of the two-language text corpus, the system performs classifier training 411 to teach the classifier 117 to determine whether an arc from the set of arcs of the graph of generalized constituents 401 is “good” or “bad”. Training is based on the principle of comparing a graph of generalized constituents with at least one tree structure produced using parallel analysis 415. Input graph has all possible syntactic arcs. Then all the arcs on this graph that are also present in at least one syntactic tree, created by parallel analysis 415, are marked “good” and the arcs that are absent from the syntactic trees are marked “bad”. The output graph is a graph with marked arcs. During training, the classifier not only learns to distinguish “good” arcs from “bad” ones, it also forms a set of symbolic features 409 which will be used by the tree classifier 403.
The linear classifiers 405 on nodes of a decision tree of the combination classifier 117 can be trained by a variety of methods. In some implementations, these methods include a method of minimizing an error function or an SVM method.
Processing Grammatical Homonymy
In some implementation the combination classifier 117 is used to alleviate a homonyms problem. If there are multiple grammatical homonyms at one of the nodes of the graph of generalized constituents 109, their probabilities are calculated based on the proportional frequency in N-grams, which in this context must be understood as a sequence of words or a context:
p1=exp(Q1)/[exp(Q1)+exp(Q2)]
p2=exp(Q2)/[exp(Q1)+exp(Q2)]
The selection of a path in the syntactic tree is complicated by grammatical homonymy and the ambiguity of ellipsis templates. Therefore the final decision is made as follows:
The programs used to accomplish the methods corresponding to this disclosure may be a part of an operating system or may be a specialized application, component, program, dynamic library, module, script or a combination thereof.
This description shows the basic inventive concept of the inventors, which cannot be limited by the hardware mentioned earlier. It should be noted that hardware is primarily intended to solve a narrow problem. As time goes by and as technology develops, this type of task becomes more complex or evolves. New tools arise that are capable of meeting new demands. In this sense, it is appropriate to look at this hardware from the point of view of the class of technical tasks it can solve, not simply a technical implementation on some set of hardware components.
The computer system 500 also usually includes input and output ports to transfer information out and receive information. For interaction with a user, the computer system 500 may contain one or more input devices 506 (such as a keyboard, a mouse, a scanner, or other) and output devices 508 (such as displays or special indicators). The computer system 500 may also have one or more permanent storage devices 510 such as an optical disk drive (CD, DVD, or other), a hard disk, or a tape drive. In addition, the computer system 500 may have an interface with one or more networks 512 that provide connection with other networks and computer equipment. In particular, this may be a local area network (LAN) or a Wi-Fi network, and may or may not be connected to the World Wide Web (Internet). It is understood that the computer system 500 may include analog and/or digital interfaces between the processor 502 and each of the components 504, 506, 508, 510 and 512.
The computer system 500 is managed by an operating system 514 and includes various applications, components, programs, objects, modules and other, designated by the consolidated number 516.
The programs used to accomplish the methods corresponding to this invention may be a part of an operating system or may be a specialized peripheral, component, program, dynamic library, module, script, or a combination thereof.
This description shows the basic inventive concept of the inventors, which cannot be limited by the hardware mentioned earlier. It should be noted that hardware is primarily intended to solve a narrow problem. As time goes by and as technology develops, such a task becomes more complex or it evolves. New tools arise that are capable of meeting new demands. In this sense, it is appropriate to look at this hardware from the point of view of the class of technical tasks they solve, not simply as a technical implementation on some elementary framework.
As will be appreciated by one skilled in the art, aspects of the present invention may be embodied as a system, method or computer program product. Accordingly, aspects of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, micro-code, etc.) or an embodiment combining software and hardware aspects that may all generally be referred to herein as a “circuit,” “module” or “system.” Furthermore, aspects of the present invention may take the form of a computer program product embodied in one or more computer readable medium(s) having computer readable program code embodied thereon.
Any combination of one or more computer readable medium(s) may be utilized. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that may contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical fiber cable, RF, etc., or any suitable combination of the foregoing. Computer program code for carrying out operations for aspects of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the “C” programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider).
Aspects of the present invention have been described above with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer readable medium that may direct a computer, other programmable data processing apparatus, or other devices to function in a particular manner, such that the instructions stored in the computer readable medium produce an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks. The computer program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks.
The flowchart and block diagrams in the above figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, may be implemented by special purpose hardware-based systems that perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2014101124 | Jan 2014 | RU | national |
| Number | Name | Date | Kind |
|---|---|---|---|
| 4706212 | Toma | Nov 1987 | A |
| 5068789 | Van Vliembergen | Nov 1991 | A |
| 5128865 | Sadler | Jul 1992 | A |
| 5146405 | Church | Sep 1992 | A |
| 5175684 | Chong | Dec 1992 | A |
| 5268839 | Kaji | Dec 1993 | A |
| 5301109 | Landauer et al. | Apr 1994 | A |
| 5386556 | Courts et al. | Jan 1995 | A |
| 5418717 | Su et al. | May 1995 | A |
| 5426583 | Uribe-Echebarria Diaz De Mendibi | Jun 1995 | A |
| 5475587 | Anick et al. | Dec 1995 | A |
| 5477451 | Brown et al. | Dec 1995 | A |
| 5490061 | Tolin et al. | Feb 1996 | A |
| 5497319 | Chong et al. | Mar 1996 | A |
| 5510981 | Berger et al. | Apr 1996 | A |
| 5550934 | Van Vliembergen et al. | Aug 1996 | A |
| 5559693 | Anick et al. | Sep 1996 | A |
| 5669007 | Tateishi | Sep 1997 | A |
| 5677835 | Carbonell et al. | Oct 1997 | A |
| 5678051 | Aoyama | Oct 1997 | A |
| 5687383 | Nakayama et al. | Nov 1997 | A |
| 5696980 | Brew | Dec 1997 | A |
| 5715468 | Budzinski | Feb 1998 | A |
| 5721938 | Stuckey | Feb 1998 | A |
| 5724593 | Hargrave et al. | Mar 1998 | A |
| 5729741 | Liaguno et al. | Mar 1998 | A |
| 5737617 | Bernth et al. | Apr 1998 | A |
| 5752051 | Cohen | May 1998 | A |
| 5768603 | Brown et al. | Jun 1998 | A |
| 5784489 | Van Vliembergen et al. | Jul 1998 | A |
| 5787410 | McMahon | Jul 1998 | A |
| 5794050 | Dahlgren et al. | Aug 1998 | A |
| 5794177 | Carus et al. | Aug 1998 | A |
| 5826219 | Kutsumi | Oct 1998 | A |
| 5826220 | Takeda et al. | Oct 1998 | A |
| 5848385 | Poznanski et al. | Dec 1998 | A |
| 5867811 | O'Donoghue | Feb 1999 | A |
| 5873056 | Liddy et al. | Feb 1999 | A |
| 5884247 | Christy | Mar 1999 | A |
| 5995920 | Carbonell et al. | Nov 1999 | A |
| 6006221 | Liddy et al. | Dec 1999 | A |
| 6055528 | Evans | Apr 2000 | A |
| 6076051 | Messerly et al. | Jun 2000 | A |
| 6081774 | De Hita et al. | Jun 2000 | A |
| 6161083 | Franz et al. | Dec 2000 | A |
| 6182028 | Karaali et al. | Jan 2001 | B1 |
| 6223150 | Duan et al. | Apr 2001 | B1 |
| 6233544 | Alshawi | May 2001 | B1 |
| 6233546 | Datig | May 2001 | B1 |
| 6243669 | Horiguchi | Jun 2001 | B1 |
| 6243670 | Bessho et al. | Jun 2001 | B1 |
| 6243689 | Norton | Jun 2001 | B1 |
| 6243723 | Ikeda et al. | Jun 2001 | B1 |
| 6246977 | Messerly et al. | Jun 2001 | B1 |
| 6260008 | Sanfilippo | Jul 2001 | B1 |
| 6266642 | Franz et al. | Jul 2001 | B1 |
| 6275789 | Moser et al. | Aug 2001 | B1 |
| 6278967 | Akers et al. | Aug 2001 | B1 |
| 6282507 | Horiguchi et al. | Aug 2001 | B1 |
| 6285978 | Bernth et al. | Sep 2001 | B1 |
| 6330530 | Horiguchi et al. | Dec 2001 | B1 |
| 6345245 | Sugiyama et al. | Feb 2002 | B1 |
| 6349276 | McCarley et al. | Feb 2002 | B1 |
| 6356864 | Foltz et al. | Mar 2002 | B1 |
| 6356865 | Franz et al. | Mar 2002 | B1 |
| 6381598 | Williamowski et al. | Apr 2002 | B1 |
| 6393389 | Chanod et al. | May 2002 | B1 |
| 6442524 | Ecker et al. | Aug 2002 | B1 |
| 6463404 | Appleby | Oct 2002 | B1 |
| 6470306 | Pringle et al. | Oct 2002 | B1 |
| 6523026 | Gillis et al. | Feb 2003 | B1 |
| 6529865 | Duan et al. | Mar 2003 | B1 |
| 6601026 | Appelt et al. | Jul 2003 | B2 |
| 6604101 | Chan et al. | Aug 2003 | B1 |
| 6622123 | Chanod et al. | Sep 2003 | B1 |
| 6658627 | Gallup et al. | Dec 2003 | B1 |
| 6721697 | Duan | Apr 2004 | B1 |
| 6760695 | Kuno et al. | Jul 2004 | B1 |
| 6778949 | Duan et al. | Aug 2004 | B2 |
| 6871174 | Dolan et al. | Mar 2005 | B1 |
| 6871199 | Binnig et al. | Mar 2005 | B1 |
| 6901399 | Corston et al. | May 2005 | B1 |
| 6901402 | Corston-Oliver et al. | May 2005 | B1 |
| 6928407 | Ponceleon et al. | Aug 2005 | B2 |
| 6928448 | Franz et al. | Aug 2005 | B1 |
| 6937974 | D'Agostini | Aug 2005 | B1 |
| 6947923 | Cha et al. | Sep 2005 | B2 |
| 6965857 | Decary | Nov 2005 | B1 |
| 6983240 | Ait-Mokhtar et al. | Jan 2006 | B2 |
| 6986104 | Green et al. | Jan 2006 | B2 |
| 7013264 | Dolan et al. | Mar 2006 | B2 |
| 7020601 | Hummel et al. | Mar 2006 | B1 |
| 7027974 | Busch et al. | Apr 2006 | B1 |
| 7050964 | Menzes et al. | May 2006 | B2 |
| 7085708 | Manson et al. | Aug 2006 | B2 |
| 7132445 | Taveras et al. | Nov 2006 | B2 |
| 7146358 | Gravano et al. | Dec 2006 | B1 |
| 7167824 | Kallulli | Jan 2007 | B2 |
| 7191115 | Moore | Mar 2007 | B2 |
| 7200550 | Menezes et al. | Apr 2007 | B2 |
| 7263488 | Chu et al. | Aug 2007 | B2 |
| 7269594 | Corston-Oliver et al. | Sep 2007 | B2 |
| 7346493 | Ringger et al. | Mar 2008 | B2 |
| 7356457 | Pinkham et al. | Apr 2008 | B2 |
| 7409404 | Gates | Aug 2008 | B2 |
| 7461056 | Cao et al. | Dec 2008 | B2 |
| 7466334 | Baba et al. | Dec 2008 | B1 |
| 7475015 | Epstein et al. | Jan 2009 | B2 |
| 7577683 | Cho et al. | Aug 2009 | B2 |
| 7619656 | Ben-Ezra et al. | Nov 2009 | B2 |
| 7668791 | Azzam et al. | Feb 2010 | B2 |
| 7672830 | Goutte et al. | Mar 2010 | B2 |
| 7672831 | Todhunter et al. | Mar 2010 | B2 |
| 7739102 | Bender | Jun 2010 | B2 |
| 7769579 | Zhao et al. | Aug 2010 | B2 |
| 8065290 | Hogue et al. | Nov 2011 | B2 |
| 8073865 | Davis et al. | Dec 2011 | B2 |
| 8078450 | Anisimovich et al. | Dec 2011 | B2 |
| 8145473 | Anisimovich et al. | Mar 2012 | B2 |
| 8176048 | Morgan et al. | May 2012 | B2 |
| 8214199 | Anismovich et al. | Jul 2012 | B2 |
| 8229730 | Van Den Berg et al. | Jul 2012 | B2 |
| 8229944 | Latzina et al. | Jul 2012 | B2 |
| 8234106 | Marcu | Jul 2012 | B2 |
| 8260049 | Deryagin et al. | Sep 2012 | B2 |
| 8266077 | Handley | Sep 2012 | B2 |
| 8271453 | Pasca et al. | Sep 2012 | B1 |
| 8285728 | Rubin | Oct 2012 | B1 |
| 8300949 | Xu | Oct 2012 | B2 |
| 8301633 | Cheslow | Oct 2012 | B2 |
| 8370128 | Brun et al. | Feb 2013 | B2 |
| 8402036 | Blair-Goldensohn et al. | Mar 2013 | B2 |
| 8423495 | Komissarchik et al. | Apr 2013 | B1 |
| 8468153 | Ahlberg et al. | Jun 2013 | B2 |
| 8495042 | Symington et al. | Jul 2013 | B2 |
| 8533188 | Yan et al. | Sep 2013 | B2 |
| 8548951 | Solmer et al. | Oct 2013 | B2 |
| 8554558 | McCarley et al. | Oct 2013 | B2 |
| 8577907 | Singhal et al. | Nov 2013 | B1 |
| 8856096 | Marchisio et al. | Oct 2014 | B2 |
| 20010014902 | Hu et al. | Aug 2001 | A1 |
| 20010029442 | Shiotsu et al. | Oct 2001 | A1 |
| 20010029455 | Chin et al. | Oct 2001 | A1 |
| 20020040292 | Marcu | Apr 2002 | A1 |
| 20020152246 | Critchlow | Oct 2002 | A1 |
| 20030145285 | Miyahira et al. | Jul 2003 | A1 |
| 20030158723 | Masuichi et al. | Aug 2003 | A1 |
| 20030176999 | Calcagno et al. | Sep 2003 | A1 |
| 20030182102 | Corston-Oliver et al. | Sep 2003 | A1 |
| 20030204392 | Finnigan et al. | Oct 2003 | A1 |
| 20040034520 | Langkilde-Geary et al. | Feb 2004 | A1 |
| 20040064438 | Kostoff | Apr 2004 | A1 |
| 20040098247 | Moore | May 2004 | A1 |
| 20040122656 | Abir | Jun 2004 | A1 |
| 20040172235 | Pinkham et al. | Sep 2004 | A1 |
| 20040193401 | Ringger et al. | Sep 2004 | A1 |
| 20040254781 | Appleby | Dec 2004 | A1 |
| 20040261016 | Glass et al. | Dec 2004 | A1 |
| 20050010421 | Watanabe et al. | Jan 2005 | A1 |
| 20050015240 | Appleby | Jan 2005 | A1 |
| 20050080613 | Colledge et al. | Apr 2005 | A1 |
| 20050086047 | Uchimoto et al. | Apr 2005 | A1 |
| 20050108630 | Wasson et al. | May 2005 | A1 |
| 20050137853 | Appleby et al. | Jun 2005 | A1 |
| 20050155017 | Berstis et al. | Jul 2005 | A1 |
| 20050171757 | Appleby | Aug 2005 | A1 |
| 20050209844 | Wu et al. | Sep 2005 | A1 |
| 20050240392 | Munro, Jr. et al. | Oct 2005 | A1 |
| 20060004563 | Campbell et al. | Jan 2006 | A1 |
| 20060004653 | Strongin | Jan 2006 | A1 |
| 20060080079 | Yamabana | Apr 2006 | A1 |
| 20060095248 | Menezes | May 2006 | A1 |
| 20060095250 | Chen et al. | May 2006 | A1 |
| 20060217964 | Kamatani et al. | Sep 2006 | A1 |
| 20060224378 | Chino et al. | Oct 2006 | A1 |
| 20060293876 | Kamatani et al. | Dec 2006 | A1 |
| 20070010990 | Woo | Jan 2007 | A1 |
| 20070016398 | Buchholz | Jan 2007 | A1 |
| 20070083359 | Bender | Apr 2007 | A1 |
| 20070083505 | Ferrari et al. | Apr 2007 | A1 |
| 20070094006 | Todhunter et al. | Apr 2007 | A1 |
| 20070100601 | Kimura | May 2007 | A1 |
| 20070150800 | Betz et al. | Jun 2007 | A1 |
| 20070203688 | Fuji et al. | Aug 2007 | A1 |
| 20070250305 | Maxwell | Oct 2007 | A1 |
| 20080086300 | Anismovich | Apr 2008 | A1 |
| 20080091405 | Anisimovich | Apr 2008 | A1 |
| 20080133218 | Zhou et al. | Jun 2008 | A1 |
| 20080228464 | Al-Onaizan et al. | Sep 2008 | A1 |
| 20080319947 | Latzina et al. | Dec 2008 | A1 |
| 20090063472 | Pell et al. | Mar 2009 | A1 |
| 20090070094 | Best et al. | Mar 2009 | A1 |
| 20090182549 | Anisimovich | Jul 2009 | A1 |
| 20100082324 | Itagaki et al. | Apr 2010 | A1 |
| 20110055188 | Gras | Mar 2011 | A1 |
| 20110072021 | Lu et al. | Mar 2011 | A1 |
| 20110191286 | Cho et al. | Aug 2011 | A1 |
| 20110258181 | Brdiczka et al. | Oct 2011 | A1 |
| 20110295864 | Betz et al. | Dec 2011 | A1 |
| 20110301941 | De Vocht | Dec 2011 | A1 |
| 20110307435 | Overell et al. | Dec 2011 | A1 |
| 20120023104 | Johnson et al. | Jan 2012 | A1 |
| 20120030226 | Holt et al. | Feb 2012 | A1 |
| 20120123766 | Anisimovich | May 2012 | A1 |
| 20120131060 | Heidasch et al. | May 2012 | A1 |
| 20120197628 | Best et al. | Aug 2012 | A1 |
| 20120197885 | Patterson | Aug 2012 | A1 |
| 20120203777 | Laroco, Jr. et al. | Aug 2012 | A1 |
| 20120221553 | Wittmer et al. | Aug 2012 | A1 |
| 20120246153 | Pehle | Sep 2012 | A1 |
| 20120296897 | Xin-Jing et al. | Nov 2012 | A1 |
| 20120310627 | Qi et al. | Dec 2012 | A1 |
| 20130013291 | Bullock et al. | Jan 2013 | A1 |
| 20130054589 | Cheslow | Feb 2013 | A1 |
| 20130091113 | Gras | Apr 2013 | A1 |
| 20130132383 | Ahlberg et al. | May 2013 | A1 |
| 20130138696 | Turdakov et al. | May 2013 | A1 |
| 20130144592 | Och et al. | Jun 2013 | A1 |
| 20130144594 | Bangalore et al. | Jun 2013 | A1 |
| 20130185307 | El-Yaniv et al. | Jul 2013 | A1 |
| 20130254209 | Kang et al. | Sep 2013 | A1 |
| 20130282703 | Puterman-Sobe et al. | Oct 2013 | A1 |
| 20130311487 | Moore et al. | Nov 2013 | A1 |
| 20130318095 | Harold | Nov 2013 | A1 |
| 20140012842 | Yan et al. | Jan 2014 | A1 |
| Number | Date | Country |
|---|---|---|
| 2400400 | Dec 2001 | EP |
| 1365329 | Oct 2009 | EP |
| 2011160204 | Dec 2011 | WO |
| Entry |
|---|
| Bolshakov, “Co-Ordinative Ellipsis in Russian Texts: Problems of Description and Restoration”, Published in: Proceeding COLING '88 Proceedings of the 12th conference on Computational linguistics—vol. 1 doi>10.3115/991635.991649, 1988, 65-67. |
| Hutchins, “Machine Translation: past, present, future”, (Ellis Horwood Series in Computers and their Applications) Ellis Horwood: Chichester, 1986, 382 pp. ISBN 0-85312-788-3, $49.95 (hb). |
| Mitamura, et al., “An Efficient Interlingua Translation System for Multi-Lingual Document Production”, http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.44.5702, Jul. 1, 1991. |
| Number | Date | Country | |
|---|---|---|---|
| 20150199331 A1 | Jul 2015 | US |
