CN108829666B - Reading comprehension problem solving method based on semantic analysis and SMT (surface mount technology) solution - Google Patents
Reading comprehension problem solving method based on semantic analysis and SMT (surface mount technology) solution Download PDFInfo
- Publication number
- CN108829666B CN108829666B CN201810507874.4A CN201810507874A CN108829666B CN 108829666 B CN108829666 B CN 108829666B CN 201810507874 A CN201810507874 A CN 201810507874A CN 108829666 B CN108829666 B CN 108829666B
- Authority
- CN
- China
- Prior art keywords
- reading
- formula
- sentence
- order logic
- answer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/205—Parsing
- G06F40/211—Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/253—Grammatical analysis; Style critique
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Machine Translation (AREA)
Abstract
The invention provides a reading comprehension question solving method based on semantic analysis and SMT (surface mount technology) solving. According to the method, semantic analysis is firstly carried out on reading comprehension questions, then a first-order logic formula corresponding to the reading comprehension questions is generated, and then four hypotheses, namely a unique name hypothesis, a closed world hypothesis, a closed reason hypothesis and a unique answer hypothesis are introduced to generate an additional first-order logic formula. The two part first order logic formulas form a knowledge base expressing the information contained in the reading comprehension problem. And generating a first-order logic formula corresponding to the candidate answer according to the question sentence in the reading and understanding problem. Finally, the method uses an SMT solver to verify whether the knowledge base contains a first-order logic formula corresponding to the candidate answer, and then the answer is solved. Compared with the existing mode of expressing the text by using a neural network model and word vectors, the method can better establish the relationship among the events described in the reading and understanding problem, thereby endowing the reading and understanding system with stronger expression capability and reasoning capability.
Description
Technical Field
The invention relates to the field of computers, in particular to a reading comprehension question solving method based on semantic parsing and SMT (surface mount technology) solving.
Background
Reading a piece of text and then answering questions related to the text is a challenging task for machines. A machine can only accomplish this task if it has the ability to read, understand and solve inferentially on the text. In order to test the strength of machine reading comprehension and reasoning solving ability, scholars propose various data sets including MCTest, WikiQA, sqad, MS MARCO and the like. The reading comprehension problem in these data sets is composed of a long text and related problems. They are characterized in that: 1. the text is long, and the machine needs to eliminate the interference of information irrelevant to the problem in the text; 2. the answers of most questions directly appear in a certain language in the long text, and the machine can obtain the answers only by analyzing the language; 3. the sentence pattern is rich, the grammar is complex, and the machine needs to analyze different sentence patterns and grammars. These data sets, however, do not emphasize reading the relationships between events in the problem, nor do they require a machine to express and infer these relationships.
Winograd Schema Change (WSC), proposed by Hector Levesque et al, may be used to evaluate the machine's common sense reasoning capabilities. The WSC dataset consists of a set of singleton choice questions. The choice questions can be regarded as reading comprehension questions. When the machine is used for solving the reading and understanding problem in the WSC data set, the machine firstly searches the common knowledge related to the question from the outside of the question text and then combines the common knowledge with the question text so as to deduce the answer of the question. The common sense containing WSC dataset is obtained by adding common sense to the reading comprehension problem in the WSC dataset. When the machine is used for solving the reading and understanding problem in the WSC data set containing the common sense, the machine needs to understand the relationship among the events described by the common sense in the problem and then uses the relationship to carry out reasoning solution on the problem. Thus, a WSC dataset containing common sense can be used to examine the machine's ability to make deep level inferences about reading comprehension problems.
Currently, the following work provides machines with the ability to express reading comprehension problem textual information and to inferentially solve reading comprehension problems. R-NET was proposed by the Nature computing group of Microsoft's Asian institute. The method comprises the steps of firstly obtaining text paragraph information in a reading and understanding problem through a recurrent neural network model, then refining the text paragraph information by using a self-matching mechanism, and finally locating the position of an answer in the text paragraph by using a pointer network. Chuanqi Tan et al proposed S-NET. The method extracts correct answers from an article by using an answer extraction model and an answer synthesis model which are built by a neural network. Yelong Shen et al propose ReasoNet. The method infers the relationship among question sentences, text paragraphs and answers of reading and understanding questions in a multi-round exploration mode. These methods, which use neural network models to obtain information for expressing reading comprehension topic texts, have certain advantages in processing sentence-rich and grammatically complex data sets. However, these methods are not suitable for solving the reading-understanding problem requiring deep-level reasoning. Therefore, these methods cannot solve the reading comprehension problem in the WSC data set containing common sense well, so that a reading comprehension problem solving method with stronger expression ability and reasoning ability is urgently needed.
In the reading comprehension problem solving method based on semantic parsing and SMT solving, two tools are required to be used. The first is sempe, which is a semantic parsing tool developed by stanford university. The second is Z3, which is a Satisfiability Model Theory (SMT) solver developed by Microsoft institute.
Disclosure of Invention
In order to overcome at least one defect (deficiency) in the prior art, the invention provides a reading comprehension question solving method based on semantic parsing and SMT solving.
In order to solve the technical problems, the technical scheme of the invention is as follows:
a reading comprehension question solving method based on semantic analysis and SMT solution comprises the steps of firstly, carrying out semantic analysis on a reading comprehension question by using an SEMPRE semantic analysis tool, then, generating a first-order logic formula phi for expressing text information of the reading comprehension question according to semantic analysis results, and then, generating a first-order logic formula corresponding to candidate answers
Finally, the method is right
Taking the sum of the negation and phi to obtain a formula
And invokes Z3 to solve
Is not required. If it is
Is not satisfiable, then
The corresponding candidate answer is the answer of the reading understanding problem solved by the method.
The reading comprehension problem solving method based on semantic parsing and SMT solving comprises the following steps:
s1, inputting a reading comprehension problem, analyzing the reading comprehension problem by using an SEMPRE semantic analysis tool, and acquiring the lemma and the part of speech of the word in the reading comprehension problem and the dependency relationship between the word and the word.
And S2, identifying predicate symbols and constant elements. For the lemmas and parts of speech of a word, the following decisions are performed to identify predicate symbols and constant elements in a first order logical formula:
1. if the word is a verb, an adjective or a common noun, declaring the word as a predicate symbol in a first-order logic language;
2. a word is said to be a constant in a first-order logical language if the word is a proper noun or a special noun, where a special noun is a noun that is adjective by the definite article "the", the adjective "this", or the adjective "that";
3. if the word is a pronoun, the pronoun is declared to be a constant in a first-order logical language.
And S3, constructing an atomic formula. After predicate symbols and constant elements existing in the reading and understanding problem are obtained, an atomic formula can be constructed according to the dependency relationship between words.
And S4, constructing a complex formula. After the atomic formula is obtained, a complex formula can be constructed according to the connection relation between sentences. Suppose the sentence A corresponds to an atomic formula of
The atomic formula corresponding to sentence B is psi, Pair
And ψ uses the following construction rules:
1. the structure contains the formula: when the sentence A and the sentence B appear in the sentence pattern "If A the B", then pair
And the psi configuration implication,
2. structure combination formula: when the sentence A and the sentence B are connected by the conjunction "and" or there is no conjunction between the sentence A and the sentence B, the pair
Carrying out conjunction with psi;
3. structure extraction formula: when the sentence A and the sentence B are connected by the conjunction "or", then the pair
And psi.
And S5, adding quantifier words in the complex formula. The rule for adding quantifier is as follows:
1. when the words are adjective by "all", "every", adding a full-scale word to quantize the words;
2. when a word is adjective by "some", "a", "not all", a quantifier is added to quantify the word.
Obtaining final first-order logic formula by adding quantifier
Steps S2, S3, S4, S5 are only basic methods of translating natural language text into a first order logical formula. Due to the richness of natural language, when the first order logic formula generated according to the rules in the above steps cannot correctly express the natural language text, additional rules need to be introduced to assist in generating the first order logic formula. The specific situation may depend on the embodiment.
S6, steps S2, S3, S4 and S5 generate a first order logical formula for expressing reading comprehension topic text. These formulas do not fully express the information contained in the topic text. Thus, four assumptions are introduced to generate additional first order logic formulas, respectively
First, unique name assumptions are introduced to generate
The content is as follows: suppose that the reading-understanding problem text contains a set of constant elements C ═ C1,...,ckAnd k is larger than or equal to 1, generating a formula:
this assumption ensures that different constant elements represent different entities in the real world.
Second, introducing closed-world assumptions to generate
The content is as follows: suppose that the reading-understanding problem text contains a set of constant elements C ═ C1,...,ckAnd k is larger than or equal to 1, generating a formula:
this assumption ensures that the constant elements in the reading-understanding problem text constitute all elements in the domain.
Third, the introduction of the closed cause hypothesis to generate
This assumption applies only to natural language text sentences having a sentence pattern "If A then B", which means that If event A occurs, then event B also occurs. Then event a is the only reason for event B in the world formed by the sentence, i.e. event a must occur when event B occurs. In the following, the content of this assumption is given: assuming that event a is represented by the formula λ and event B is represented by the formula ψ, the formula is generated:
ψ→λ
the formula shows that when event B occurs, event a also occurs.
Fourth, when reading comprehension questions as single choice questions, unique answer hypotheses are introduced to generate
The content is as follows: assume a candidate answer set a ═ a1,a2,...,akK is more than or equal to 2, and the first order corresponding to each candidate answerThe logical formula is phi1,φ2,...,φkThen, the formula is generated:
V1≤i≤kφi
the two formulas are combined to obtain
This assumption ensures that Z3 returns only one answer per question.
S7, the first-order logic formula is generated in the steps
These equations are then conjuncted to obtain the new equation phi.
And S8, performing semantic analysis on the question sentences in the reading and understanding problem, and generating a first-order logic formula gamma corresponding to the candidate answers. Then negating gamma to obtain a first-order logic formula
Then will be
Is combined with phi to obtain a first order logic formula
And invokes Z3 to solve
Is not required. If it is
If the answer is unsatisfiable, phi implies gamma, namely the first-order logic formula corresponding to the reading and understanding problem text implies the first-order logic formula corresponding to the candidate answer, so that the candidate answer is the solved answer. Otherwise, the first order corresponding to the next candidate answerThe logic formula performs the same verification until the solution to the formula by Z3 is unsatisfiable or all candidate answers are verified.
Compared with the prior art, the technical scheme of the invention has the beneficial effects that:
(1) the invention better analyzes the semantic information contained in the reading and understanding problem by adopting a semantic analysis mode, and generates a first-order logic formula corresponding to the text according to the analysis result to express the text, and the first-order logic formula can express the relationship among the events described by the words in the text in a display mode.
(2) The present invention introduces four assumptions to generate additional first order logical formulas related to reading comprehension problems. Through the additional first-order logic formulas, the invention can more completely express the information contained in the reading-understanding problem, thereby strengthening the reasoning and solving capability of the invention on the reading-understanding problem.
(3) The method has stronger expression capability and reasoning capability when processing the reading-understanding problem solving data set which needs deep reasoning.
Drawings
FIG. 1 is a flow chart of a reading comprehension problem solving method based on semantic parsing and SMT solving.
FIG. 2 is a diagram of semantic parsing results of an embodiment.
FIG. 3 is a graph comparing experimental data.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the patent;
for the purpose of better illustrating the embodiments, certain features of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product;
it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
In the description of the present invention, it is to be understood that the terms "first" and "second" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implying any number of technical features indicated. Thus, a defined feature of "first", "second", may explicitly or implicitly include one or more of that feature.
The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.
Example 1
A reading-understanding problem solving method based on semantic parsing and SMT solving is shown in a step flow chart of the reading-understanding problem solving method based on semantic parsing and SMT solving, and the reading-understanding problem solving method based on the semantic parsing and SMT solving is only suitable for reading-understanding problems composed of English texts.
In the following, a reading comprehension question is taken as an example, which is composed of a statement sentence, an interrogative sentence, and a candidate answer. Due to the particularity of this example, the text "thing B" and "thing C" in the example may be considered arguments and need to be quantized using full-scale words. Further, the text "because it's too small" in the statement sentence is related to the question sentence "What is too small". Thus, the text "because it's too small" is redundant information, so in this example, the present invention does not need to generate a relevant first order logical formula to express the text.
Displaying sentences: the graphics do't fit into The suitcase, beacon it's to small, if this is big B is big, and this is small C, The n this is B do't fit into this C.
Question sentence: what is a too small?
Candidate answers: the suitcase/The graphics.
The invention takes the reading and understanding problem solving text as input and outputs the solved answer.
The process of solving the reading comprehension problem of the invention is as follows:
s1, inputting a reading comprehension question text, and performing semantic analysis on a first sentence in a statement sentence of the reading comprehension question by using SEMPRE to obtain a result shown in figure 2. The parsing result in the figure includes six items, which are token, Lemmatized token, POS Tags, NER Values, and Dependency. The Lemmatized Tokens contain the lemmas of all words. And performing semantic analysis on the remaining text.
S2, according to the result of semantic analysis on the text in the step S1, the lemma information of the word is contained in the content of the word-shaped participle (Lemmatized Tokens) in the result, so that the lemma and the part of speech (POS Tags) of the word are judged, and the predicate symbol and the constant element can be obtained. Therefore, a binary predicate symbol fitInto, a unary predicate symbol large, and an unary predicate symbol small, a normal predicate, and a cup can be obtained. Because both graphics and cup are terms. According to the nature of the reading comprehension problem, the word phrases thing B and thing C are regarded as arguments, and need to be quantized using full-scale words.
And S3, constructing an atomic formula according to the Dependency relationship (Dependency child) in the semantic analysis result, constructing a complex formula, and finally adding quantifier words into the first-order logic formula. Thus, a first order logic formula can be obtained:
fitInto(trophy,suitcase) (1)
s4, introducing four assumptions to generate an additional first-order logic formula. First, unique name assumptions are introduced to generate the corresponding first order logical formulas for orthonormal control and suitcase:
secondly, introducing a closed-world assumption to generate a first-order logic formula:
the formula indicates that any argument can only take the value in the argument domain D that represents the argument suitcase or argument trophy.
Then, a closed cause assumption is introduced to generate a first order logic formula:
finally, unique answer hypotheses are introduced to generate a first order logical formula:
this formula indicates that there is one and only one correct answer in suitcase and role.
S5, the obtained formulas are combined to obtain the formulas
Information contained in the reading-understanding problem text in embodiment 1 is shown.
S6, semantic parsing is carried out on the question sentence, first-order logic formulas small (trend) and small (suitcase) corresponding to the candidate answers are generated, and then whether the two formulas are verified one by one is judged
Implications. Firstly, negating the formula small (trophy) to obtain
Then the formula is compared with
Obtaining the product by mixing
Recall Z3 pair formula
And (6) solving. Finally, the resulting solution is "sat", which means
Is satisfiable, so "trophy" is not the solution answer. Then, theSimilar operations are performed for the formula small (suitcase). Then, the resulting solution is "unsat", which means
Is not satisfactory, i.e.
Implications are small (suitcase). Thus, "suitcase" is the answer to the solution. The present invention outputs the string "suitcase".
FIG. 3 is a graph comparing experimental data. The data set adopted in the experiment is a WSC data set containing common sense. One of which includes 149 pairs of reading problems, each pair of reading problems including two reading problems that are dual to each other. The term "reading problem" is simply referred to as "problem". As can be seen from the figure, the present invention can correctly solve both of the 50-pair problems and one of the 5-pair problems, so the present invention can correctly solve 105 problems in total. And the R-NET can only solve 2 pairs of problems and 56 problems correctly, i.e. 60 problems in total. If a method can solve both problems of a pair of problems correctly, then the method is said to solve the problem correctly by inference. If a method can only solve one of a pair of questions, the method is considered likely to guess the answer to the question without reasoning correctly on the question. Therefore, compared with R-NET, the method can more accurately solve the problem in the WSC data set containing common sense, can more completely express the relation between the events in the problem text, and can use the relation to solve the problem.
Claims (6)
1. A reading-understanding problem solving method based on semantic analysis and SMT solving is characterized in that a semantic analysis tool is used for carrying out semantic analysis on a reading-understanding problem, and then a first-order logic formula phi representing text information of the reading-understanding problem is generated according to a semantic analysis result; then generating a first-order logic formula corresponding to the candidate answer
Finally, the method is right
Taking the sum of the negation and phi to obtain a formula
And calling SMT solver to solve
Satisfiability of (a); if it is
Is not satisfiable, then
The corresponding candidate answer is the answer of the reading understanding problem solved by the method,
the method comprises the following steps:
s1, inputting a reading and understanding problem solving text, analyzing the reading and understanding problem solving text by using an SEMPRE semantic analysis tool, and acquiring the word elements and the parts of speech of words and the dependency relationship among the words in the reading and understanding problem solving text;
s2, judging the word element and the part of speech of the word to obtain a predicate symbol and a constant element;
s3, constructing an atomic formula according to the predicate symbols and the relevant parameters thereof;
s4, constructing a complex formula according to the connection relation among sentences in the text;
s5, adding quantifier words in the complex formula to obtain a complete first-order logic formula
S6, introducing four assumptions to generate an additional first-order logic formula
S7. pair
Performing a combination to obtain a new formula phi;
and S8, performing semantic analysis on the question sentences in the reading and understanding problem and solving the question sentences to obtain answers.
2. The reading comprehension problem solving method based on semantic parsing and SMT solution according to claim 1, wherein the determining of the lemmas and parts of speech of the words comprises:
if the word is a verb, an adjective or a common noun, declaring the word as a predicate symbol in a first-order logic language;
a word is said to be a constant in a first-order logical language if the word is a proper noun or a special noun, where a special noun is a noun that is adjective by the definite article "the", the adjective "this", or the adjective "that";
if the word is a pronoun, the pronoun is declared to be a constant in a first-order logical language.
3. The reading-understanding problem solving method based on semantic parsing and SMT solving of claim 1, wherein the complex formula is constructed by assuming that the atomic formula corresponding to sentence A is
The atomic formula corresponding to sentence B is ψ, and the complex formula rule is constructed as follows:
the structure contains the formula: when the sentence A and the sentence B appear in the sentence pattern "If A the B", then pair
And the psi configuration implication,
structure combination formula: when the sentence A and the sentence B are connected by the conjunction "and" or there is no conjunction between the sentence A and the sentence B, the pair
Carrying out conjunction with psi;
structure extraction formula: when the sentence A and the sentence B are connected by the conjunction "or", then the pair
And psi.
4. The reading understanding problem solving method based on semantic parsing and SMT solving of claim 1, wherein the adding quantifier comprises:
when words in the sentence are adjective by "all" and "every", adding a full-scale word to quantize the words;
when words in a sentence are adjective by "some", "a", "not all", presence quantifier is added to quantify the words.
5. The reading understanding problem solving method based on semantic parsing and SMT solution according to claim 1, wherein the four hypotheses comprise a unique name hypothesis, a closed world hypothesis, a closed cause hypothesis and a unique answer hypothesis.
6. The reading-understanding problem solving method based on semantic parsing and SMT solving of claim 1, wherein the process of performing semantic parsing on the question in the reading-understanding problem and solving the answer is to perform semantic parsing on the question and generate a first-order logic formula γ corresponding to the candidate answer; then negating gamma to obtain a first-order logic formula
Then the obtained data is combined with phi to obtain a first-order logic formula
Finally calling Z3 to solve
Whether it is satisfiable; if it is
If the answer is not satisfied, then the candidate answer corresponding to the gamma is the solved answer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810507874.4A CN108829666B (en) | 2018-05-24 | 2018-05-24 | Reading comprehension problem solving method based on semantic analysis and SMT (surface mount technology) solution |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810507874.4A CN108829666B (en) | 2018-05-24 | 2018-05-24 | Reading comprehension problem solving method based on semantic analysis and SMT (surface mount technology) solution |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108829666A CN108829666A (en) | 2018-11-16 |
| CN108829666B true CN108829666B (en) | 2021-11-26 |
Family
ID=64147289
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810507874.4A Active CN108829666B (en) | 2018-05-24 | 2018-05-24 | Reading comprehension problem solving method based on semantic analysis and SMT (surface mount technology) solution |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108829666B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109949637B (en) * | 2019-03-13 | 2021-07-16 | 广东小天才科技有限公司 | Automatic answering method and device for objective questions |
| CN110457710B (en) * | 2019-08-19 | 2022-08-02 | 电子科技大学 | Method and method for establishing machine reading understanding network model based on dynamic routing mechanism, storage medium and terminal |
| CN111859990B (en) * | 2020-07-30 | 2023-11-21 | 韩朝晖 | Method and system for processing question-answer data based on semantic logic uniqueness judgment |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5794050A (en) * | 1995-01-04 | 1998-08-11 | Intelligent Text Processing, Inc. | Natural language understanding system |
| CN100524294C (en) * | 1997-07-22 | 2009-08-05 | 微软公司 | System for processing textual inputs natural language processing techniques |
| CN101853314A (en) * | 2010-07-02 | 2010-10-06 | 上海交通大学 | Automatic generating system for semantic Web service |
| US8533142B2 (en) * | 2011-02-22 | 2013-09-10 | Siemens Product Lifecycle Management Software Inc. | Product lifecycle management system using partial solve |
| CN104090958A (en) * | 2014-07-04 | 2014-10-08 | 许昌学院 | Semantic information retrieval system and method based on domain ontology |
| CN105528349B (en) * | 2014-09-29 | 2019-02-01 | 华为技术有限公司 | The method and apparatus that question sentence parses in knowledge base |
| US10504025B2 (en) * | 2015-03-13 | 2019-12-10 | Cisco Technology, Inc. | Parallel processing of data by multiple semantic reasoning engines |
| US10078514B2 (en) * | 2016-05-11 | 2018-09-18 | International Business Machines Corporation | Techniques for dynamic sequential instruction prefetching |
| CN107463553B (en) * | 2017-09-12 | 2021-03-30 | 复旦大学 | Text semantic extraction, representation and modeling method and system for elementary mathematic problems |
-
2018
- 2018-05-24 CN CN201810507874.4A patent/CN108829666B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN108829666A (en) | 2018-11-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ullmann | Automated analysis of reflection in writing: Validating machine learning approaches | |
| Leacock et al. | Automated grammatical error detection for language learners | |
| Crossley et al. | Understanding expert ratings of essay quality: Coh-Metrix analyses of first and second language writing | |
| Friedrich et al. | Situation entity types: automatic classification of clause-level aspect | |
| Ormandjieva et al. | Toward a text classification system for the quality assessment of software requirements written in natural language | |
| CN108829666B (en) | Reading comprehension problem solving method based on semantic analysis and SMT (surface mount technology) solution | |
| Günther et al. | Semantic transparency is not invisibility: A computational model of perceptually-grounded conceptual combination in word processing | |
| Bäumer et al. | Flexible ambiguity resolution and incompleteness detection in requirements descriptions via an indicator-based configuration of text analysis pipelines | |
| Quispesaravia et al. | Coh-Metrix-Esp: A complexity analysis tool for documents written in Spanish | |
| Dessai et al. | Depression detection on social media using text mining | |
| Clark et al. | Computational learning theory and language acquisition | |
| Qudar et al. | A survey on language models | |
| Barbella et al. | Analogical word sense disambiguation | |
| Prabhu et al. | A hybrid approach towards automated essay evaluation based on Bert and feature engineering | |
| Saber et al. | Arabic ontology extraction model from unstructured text | |
| Kogkalidis et al. | Discontinuous constituency and BERT: A case study of Dutch | |
| Nassiri et al. | Approaches, methods, and resources for assessing the readability of arabic texts | |
| Mahajan et al. | Machine learning based paraphrase identification system using lexical syntactic features | |
| McShane et al. | The Ontological Semantic treatment of multiword expressions | |
| CN115906818A (en) | Grammar knowledge prediction method, grammar knowledge prediction device, electronic equipment and storage medium | |
| Kurosawa et al. | Logical inference for counting on semi-structured tables | |
| CN106844448B (en) | Chinese event factual identification method and system | |
| Groza et al. | Enacting textual entailment and ontologies for automated essay grading in chemical domain | |
| Hameed et al. | Automatic Short Answer Grading System Based on Semantic Networks and Support Vector Machine | |
| Dascălu et al. | Readerbench (2)-individual assessment through reading strategies and textual complexity |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |