WO2004084156A1 - テンプレート−テンプレート構造に基づく対話式学習システム - Google Patents
テンプレート−テンプレート構造に基づく対話式学習システム Download PDFInfo
- Publication number
- WO2004084156A1 WO2004084156A1 PCT/JP2004/003838 JP2004003838W WO2004084156A1 WO 2004084156 A1 WO2004084156 A1 WO 2004084156A1 JP 2004003838 W JP2004003838 W JP 2004003838W WO 2004084156 A1 WO2004084156 A1 WO 2004084156A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- template
- learning system
- interactive learning
- templates
- nodes
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
- G09B—EDUCATIONAL OR DEMONSTRATION APPLIANCES; APPLIANCES FOR TEACHING, OR COMMUNICATING WITH, THE BLIND, DEAF OR MUTE; MODELS; PLANETARIA; GLOBES; MAPS; DIAGRAMS
- G09B19/00—Teaching not covered by other main groups of this subclass
- G09B19/06—Foreign languages
Definitions
- Template Template-interactive learning system based on template structure
- the present invention relates to a new interactive learning system that utilizes extraction rule-based templates—the template structure and the scalable functionality of buggy rules.
- the motivation for the invention is the desire to automate and simplify the time-consuming authoring 'tasks used in language-oriented intelligent learning systems. Even if the number of possible model responses is reasonably limited, when developing an ideal learning system, the number of errors that a learner may actually make is theoretically close to an infinite number. In many cases. As far as the present inventors judge, at least in the foreseeable future, even with state-of-the-art natural language processing technology, it is possible to provide automatic correction of complete free-form and error-prone sentences immediately. Not reached. The system, like many talented human teachers, would be able to do this only if it could introduce a so-called common sense world knowledge base into the system.
- the learning system (Azalea) of the present invention introduces the concept of a template 'automaton', which collects many “expected examples of various learners” consisting of “correct” answers and “wrong” answers.
- a typical NLP technology called HCS (Longest Common String) or LCS (Longest Common String) algorithm plays a decisive role as an efficient error diagnosis engine used in language learning systems, and is embedded in the template. These examples are used for diagnostic analysis of learner responses. The diagnosis is made by selecting the path with the highest similarity to the learner's input sentence from the vast number of candidate paths in the template 'database.
- a template consisting of a well-formed model translation and a badly-formed incorrect sentence
- the authoring task which builds a single pass, consumes a lot of time and wastes manpower.
- the new system of the present invention includes not only the task of simplifying, or reducing, the task of generating a template that would otherwise be time-consuming (see, for example, Nagoyuki Tokuda, Ryo Chen, Hiroyuki Sasai et al. It is also effective in improving system performance.
- the first reason that the introduced template-template architecture provides system simplicity and improved performance is that it applies the extraction rules assigned to some of the transition nodes in a single template-template.
- the new extraction rule-based and buggy 'Norail'-based template-template structure is a text-based, interactive learning system, a voice-based technology, a center or a voice portal, a system, or a system.
- a more important human 'computer' interface that implements a more natural human-computer interaction between humans and humans plays an important role in many applications, including any system. It is expected to carry. According to the present invention, the following functions are provided.
- a single template can represent a variety of different types of existing templates.
- the HCS Matching 'algorithm matches the input sentence directly to a simpler template-template, thus eliminating the need to actually expand the template-template from all the expected paths of all extracted templates. It can be developed to reduce the space and time calculation amount of the matching process when searching for the optimal path.
- templates-template is defined as follows. Templates Templates are marked with a symbol in which some of the nodes are related to the extraction rules, and that template is considered as many templates or a set of unbound templates as one template. Is defined as a special template that can be expanded to a so-called large template. With such a set of truncated templates, the various possible translations of a single L1 sentence form a large single template-template composed of a group of translated L2 sentences. It becomes possible. Since it is an unfolded template, the template-template 'scheme allows one or more templates to be extracted as it were.
- an extraction rule is always associated with a set of symbols, eg, ⁇ S l , s 2 , s J, each of which is assigned one or more nodes of the template. These associated symbols are assigned one or more values, and their function is to represent the style of the nodes that appear in one or more templates extracted from the template. In the present invention, these symbols are Le 'symbol'. Symbols associated with a single rule are called “related symbols”. Relationship symbols should have certain restrictions. As a general constraint, for a given srl, s k often needs to be constrained to 2, or to some positive integer other than 1.
- AP-NAP rules of type A rule are such that when deployed, the newly deployed template can include nodes marked with Ap / or nodes marked with ⁇ , S, imposes the condition that both of these cannot be included at the same time.
- 0 is used to indicate the case where the node marked with # 1 does not appear in the template.
- the template one template 'rules one set of appearing in the template Shikabane / 7 /? / In the marked node and Shikabane 3? Shikabane / in marked other nodes (/ is Each integer imposes the condition that it must be in the form of a personal pronoun and a pronoun possessive case, as required by the natural language grammar of the pronoun.
- the required values for PPRP, (or for PP, PPR, (or society)) must be defined by the natural language grammar of the pronoun
- Type C rules AN (arbitrary number) rules
- Type C rules impose the condition that nodes marked with AI ⁇ can be assigned any positive real number. "If I have 5 books on ZenJ is true, this rule AN; can be assigned to error 'node 5 because any number other than 5 is incorrect.
- a buggy rule is defined here as a production of general syntactical error representations characterized by expected deviations from the correct syntax rules. Specifically, assume the following form of buggy rule.
- R ⁇ ..R M is a set of nodes representing typical error representation is exact form. It is immediately understood that errors are identified by deviations from the exact path of the template-template. An example is shown.
- VB P is the first and second person present tense verb
- VBZ is the third person singular present tense verb.
- the syntactically correct expression “/ wre are 5 books” incorrectly matches the subject and verb attributes. Has been misused by a student who understands this, which in this example means that the expression " ⁇ ee 5 booksj.” Has occurred.
- FIG. 1 is a diagram showing the structure of a template-template of the present invention.
- FIG. 2 is a diagram showing a template-template structure developed according to a development rule.
- Figure 3 is a diagram showing Template 1 expanded with an example of a sentence meaning “Japan is dotted with beautiful parks nationwidej”.
- Figure 4 is a diagram showing a template expanded with an example of a sentence meaning “Japan is dotted with beautiful parks nationwidej”.
- NNS Noun, singular or set
- NNP Proper noun
- singular RB ⁇ ij
- VBN verb, past participle
- VBP verb, non-third-person singular, present tense
- Fig. 1 is a diagram showing the structure of an original template template
- Fig. 2 is a diagram showing a template-template developed according to the above-described development rules
- Fig. 3 is a diagram showing template 1
- Fig. 4 is a diagram showing template 1.
- FIG. 6 is a diagram showing a template 2; Template templates, template expansion with extraction rules, and buggy rule
- a template template for a translation of a Japanese sentence meaning "Japan is dotted with beautiful gardens nationwide.”
- the numbers shown in FIG. 1 and the like indicate the weight of each word that emphasizes the relative importance of each word in the sentence.
- the default weight of words in the template is set to 1 and these need to be assigned in relation to the importance of the word as judged by experts in the field.
- JP-A-2002-49617 by Naoyuki Tokuda, Ryo Chen and Hiroyuki Sasai for a detailed description.
- the symbol between "[" and "]” is a part of speech tag.
- the node shown at the left end in the figure is the start node.
- templates can be extracted from a single template.
- each extracted from the template template /? Kufunshi ⁇ s "Pl, with s 2, p 2 AJ.
- A. is a suitable allocation for Shimpo Le.
- A. may be either a number or a word, depending on the extraction rules used.
- the heaviest common strings of two sentences are the next most common strings ai , a 2 ,... Of ordered words that appear in both sentences in the following order:, then ..., then. .., it is defined to be a m.
- the definition of a common character string is described in AV Aho and J. D. Ullman (Computer science Press, 1992, pp. 321—327), “ ⁇ ⁇ 3 ⁇ 4 / /? ⁇ 5 ⁇ of Computer Science fw. The group ⁇ /].
- the heaviest common string between the path and the input sentence in the template is all predictable with the highest total weight. Defined as the heaviest common string in the common string.
- the search unit searches all possible valid paths of the template for the word or Z of the input sentence or the most common character string of the phrase.
- the most common string of the template and the input sentence is the highest word of the heaviest total weight in the most common string obtained from one pass of the template and the input sentence, respectively. Defined as a double common character string.
- the next step is to match the input sentence with each and every predictable template and then select the closest path .
- a detailed description of a DP (dynamic programming) -based matching procedure of a template for a sentence is described in Japanese Patent Application Laid-Open No. 2002-49617 by Naoyuki Tokuda, Ryo Chen, Hiroyuki Sasai et al.
- the method of the present invention does not physically extract all templates from a single template, but instead extracts all valid templates that can be directly extracted by the extraction rules (but not by buggy's Norail). Search for the closest path from the best paths.
- the first step required by the algorithm is to simply represent each node of the template as one or more arcs in the graph by adding weight to the arc labeled mouth for each applicable empty node
- the template-to-template is converted to a double value of an acyclic weighted finite directed graph (directed graph). Since a directed graph is transformed from a template-to-template, the function contains many arcs associated with labels and symbols whose functions depend critically on the values assigned to that symbol.
- a completely different template can be extracted if a different set of labels' symbols are assigned to the arcs. That is, assuming that there is such a directed graph, it is possible to obtain many directed graphs corresponding to templates that can be extracted from a template-template.
- the directed graph extracted from the template-template is hereinafter referred to as a template directed graph.
- the present inventors then calculate the maximum shared value from the path of all directed graphs and the common character string of the input sentence. The procedure for searching for a character string is defined below.
- the heaviest common character string of the path and input sentence ending at any specific node N of the directed graph is the heaviest of all the heaviest common character strings obtained from one path and input sentence ending in the directed graph / Defined as a sequence of words having a total weight.
- N /,..., ⁇ e represent the paths of all the directed graphs extracted from the template directed graph and ending at the marker node. Where the symbol is the value Pi
- n-tuple, ⁇ is called the label of node ⁇ .
- ⁇ the label of node ⁇ .
- the labeled node and the most common string of the input sentence have inconsistent labels, ⁇ , A? Marked nodes
- the weighted common character string of the word having the heaviest sum weight among all the weighted common character strings obtained as the weighted common character string of one directed graph extracted from the directed graph 'template Defined. Note that some nodes, such as the node labeled AP2 and the node labeled NAP2 in one directed graph, may not appear simultaneously in one directed graph extracted from the directed graph 'template.
- a rule violation label such as Ni (..., ⁇ 2, 1,..., ⁇ 2, 1 7) can be included in any computation plan of the common string between the node of the directed graph 'template and the input sentence. You should not be allowed to.
- the following algorithm describes the procedure for calculating the heaviest common character string between a template and an input sentence. In the following calculations, " ⁇ " is used as a very special value of the label 'symbol, so that its value remains unspecified until a certain stage of the calculation is reached.
- C (Ni ⁇ s 1 , p 1 , s 2 , p 2 , ..., s n , p n ⁇ , ⁇ is defined as the maximum CM (Ni, /?..., Xj, pxj, pxj, Mj) , All already defined
- CM (N .J, Mj) If the arc does not have the NiN K force S label, CM (N .J, Mj), CM (N, extended, Mj, CM (N k ⁇ ... ⁇ , Mj), CM (N k extended, M are all checked, and CM (M ⁇ s 1 ⁇ pj, s 2 , p 2 , s n , pj, ⁇ , 2 , p 2 , ..
- CM V Nk (si, p i s 2 , p 2 " .., s n , pj, Mj), s 2 , p 2 , s n , p, M J + 1 ) If one is already defined, define CM V; ⁇ / ⁇ as the maximum of the following data.
- CWN;. ⁇ ⁇ Mj is already defined and matches the arc NiN k , W k s 2 ..., s n , P J, Mj) + W (N, N k ).
- CM (Nk ⁇ sp is s 2 , p 2 .., s n , pj, Mj.
- CM (Nk ⁇ Sl , ps 2 , p 2 , s n! P n , s, pj, M If JW is already defined.
- CM Ni ⁇ s llPl , s 2 , p 2 , ... , s n , Pn , s, p, ⁇ , M are already defined
- CM (Nk ⁇ s Pl , s 2 , p 2 , ... ! Sn , Pn ,, s > P ⁇ , M J ) is already defined •
- CM (N k ⁇ s l! Pl , s 2 , p 2 , ..., s n , p n , s, p ⁇ , Mj is defined as the maximum data of the data defined above and the following data .
- CM Ni ⁇ s PL , s 2 , P 2 ".”
- S n , P J, Mj is defined and matches with ⁇ ⁇ ⁇ .
- M j + i after is set to CM (Ni ⁇ s or p s 2 , P 2 , .. "s n ' P MJ) + W (MJ
- s CM Ni ⁇ Sl, Pl , s 2, p 2, s n, p n, s, p ⁇ , Mj
- W W
- the present invention has been described in the specification with respect to the technical field of natural language translation, which is a typical application, the present invention is not limited to a natural language learning system. It can be used for any language learning system, or any system that requires a more natural, extensible interface that allows, for example, human-computer interaction.
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Physics & Mathematics (AREA)
- Educational Administration (AREA)
- Educational Technology (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Machine Translation (AREA)
Abstract
Description
Claims
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2005503774A JPWO2004084156A1 (ja) | 2003-03-22 | 2004-03-22 | テンプレート−テンプレート構造に基づく対話式学習システム |
US10/550,090 US7509296B2 (en) | 2003-03-22 | 2004-03-22 | Interactive learning system based on template-template structure |
EP04722383A EP1607925A4 (en) | 2003-03-22 | 2004-03-22 | INTERACTIVE LEARNING SYSTEM BASED ON A TEMPLATE TEMPLATE STRUCTURE |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2003-120733 | 2003-03-22 | ||
JP2003120733 | 2003-03-22 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2004084156A1 true WO2004084156A1 (ja) | 2004-09-30 |
Family
ID=33028293
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2004/003838 WO2004084156A1 (ja) | 2003-03-22 | 2004-03-22 | テンプレート−テンプレート構造に基づく対話式学習システム |
Country Status (4)
Country | Link |
---|---|
US (1) | US7509296B2 (ja) |
EP (1) | EP1607925A4 (ja) |
JP (1) | JPWO2004084156A1 (ja) |
WO (1) | WO2004084156A1 (ja) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8417513B2 (en) * | 2008-06-06 | 2013-04-09 | Radiant Logic Inc. | Representation of objects and relationships in databases, directories, web services, and applications as sentences as a method to represent context in structured data |
WO2012170053A1 (en) * | 2011-06-09 | 2012-12-13 | Rosetta Stone, Ltd. | Producing controlled variations in automated teaching system interactions |
US20140052659A1 (en) * | 2012-08-14 | 2014-02-20 | Accenture Global Services Limited | Learning management |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20010044098A1 (en) * | 2000-01-26 | 2001-11-22 | Johnson Benny G | Intelligent tutoring methodologh using consistency rules to improve meaningful response |
JP2003150584A (ja) * | 2001-11-16 | 2003-05-23 | Nec Corp | マルチマルチテンプレート管理システムおよびその方法とマルチマルチテンプレート管理プログラム |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3778785B2 (ja) | 2000-08-01 | 2006-05-24 | 株式会社サン・フレア | 最適翻訳文選定方法,選定装置および記録媒体 |
-
2004
- 2004-03-22 WO PCT/JP2004/003838 patent/WO2004084156A1/ja active Application Filing
- 2004-03-22 US US10/550,090 patent/US7509296B2/en not_active Expired - Lifetime
- 2004-03-22 EP EP04722383A patent/EP1607925A4/en not_active Withdrawn
- 2004-03-22 JP JP2005503774A patent/JPWO2004084156A1/ja active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20010044098A1 (en) * | 2000-01-26 | 2001-11-22 | Johnson Benny G | Intelligent tutoring methodologh using consistency rules to improve meaningful response |
JP2003150584A (ja) * | 2001-11-16 | 2003-05-23 | Nec Corp | マルチマルチテンプレート管理システムおよびその方法とマルチマルチテンプレート管理プログラム |
Non-Patent Citations (2)
Title |
---|
NAOYUKI TOKUDA ET AL.: "Template automaton ni yoru on line chiteki eisakubun kyoiku shien system", INFORMATION AND COMMUNICATION ENGINEERS RONBUNSHI, vol. J-84-D-I, no. 7, 1 July 2001 (2001-07-01), pages 1089 - 1101, XP002983033 * |
See also references of EP1607925A4 * |
Also Published As
Publication number | Publication date |
---|---|
EP1607925A1 (en) | 2005-12-21 |
EP1607925A4 (en) | 2011-01-26 |
JPWO2004084156A1 (ja) | 2006-06-22 |
US20060154218A1 (en) | 2006-07-13 |
US7509296B2 (en) | 2009-03-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Dagan et al. | Recognizing textual entailment: Models and applications | |
Khan et al. | A novel natural language processing (NLP)–based machine translation model for English to Pakistan sign language translation | |
Friedrich et al. | Process model generation from natural language text | |
Bates | The theory and practice of augmented transition network grammars | |
JP4451435B2 (ja) | 言語理解装置、言語理解方法、及び、コンピュータプログラム | |
CN106919434A (zh) | 一种代码生成方法及装置 | |
Sowa | Conceptual graphs: Draft proposed american national standard | |
Santana et al. | A survey on narrative extraction from textual data | |
CN113343694A (zh) | 一种医疗命名实体识别方法及系统 | |
Hana et al. | Building a learner corpus | |
Mille et al. | Underspecified universal dependency structures as inputs for multilingual surface realisation | |
Dolatian et al. | Redtyp: A database of reduplication with computational models | |
Wang | Abstract Meaning Representation Parsing | |
Pust et al. | Using syntax-based machine translation to parse english into abstract meaning representation | |
Lin et al. | Designing a tag-based statistical math word problem solver with reasoning and explanation | |
Kempson et al. | Natural-language syntax as procedures for interpretation: the dynamics of ellipsis construal | |
Abdel Monem et al. | Generating Arabic text in multilingual speech-to-speech machine translation framework | |
Zanetti | NLP methods for the automatic generation of exercises for second language learning from parallel corpus data | |
WO2004084156A1 (ja) | テンプレート−テンプレート構造に基づく対話式学習システム | |
Carenini et al. | Improving communication in E-democracy using natural language processing | |
Amtrup | Incremental speech translation | |
Kate et al. | Semantic parsing. The task, the state of the art and the future | |
Lapalme | Realizing Universal Dependencies structures using a symbolic approach | |
Tokuda et al. | A new KE-free online ICALL system featuring error contingent feedback | |
Bajwa | A natural language processing approach to generate sbvr and ocl |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
AK | Designated states |
Kind code of ref document: A1 Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BW BY BZ CA CH CN CO CR CU CZ DE DK DM DZ EC EE EG ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NA NI NO NZ OM PG PH PL PT RO RU SC SD SE SG SK SL SY TJ TM TN TR TT TZ UA UG US UZ VC VN YU ZA ZM ZW |
|
AL | Designated countries for regional patents |
Kind code of ref document: A1 Designated state(s): BW GH GM KE LS MW MZ SD SL SZ TZ UG ZM ZW AM AZ BY KG KZ MD RU TJ TM AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HU IE IT LU MC NL PL PT RO SE SI SK TR BF BJ CF CG CI CM GA GN GQ GW ML MR NE SN TD TG |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application | ||
DPEN | Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed from 20040101) | ||
WWE | Wipo information: entry into national phase |
Ref document number: 2005503774 Country of ref document: JP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2004722383 Country of ref document: EP |
|
ENP | Entry into the national phase |
Ref document number: 2006154218 Country of ref document: US Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 10550090 Country of ref document: US |
|
WWP | Wipo information: published in national office |
Ref document number: 2004722383 Country of ref document: EP |
|
WWP | Wipo information: published in national office |
Ref document number: 10550090 Country of ref document: US |