| { |
| "paper_id": "W87-0104", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T06:44:30.491941Z" |
| }, |
| "title": "Aarno Lehtola and Timo Honkela KIELIKONE-project, SITRA Foundation P.O.Box 329, SF-00121 Helsinki Finland tel", |
| "authors": [], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "AWARE is a knowledge representation language for specifying NLU inference rules. AWARE-system takes as its input the parse trees of NL utterances and further refines them by using DAG-transformations (Directed Acyclic Graph) and recursive descent translation techniques. AWARE has been used for semantic analysis in our Finnish language database interface. The input dependency tree is transformed first into a predication DAG and then reduced into a conceptual database query.", |
| "pdf_parse": { |
| "paper_id": "W87-0104", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "AWARE is a knowledge representation language for specifying NLU inference rules. AWARE-system takes as its input the parse trees of NL utterances and further refines them by using DAG-transformations (Directed Acyclic Graph) and recursive descent translation techniques. AWARE has been used for semantic analysis in our Finnish language database interface. The input dependency tree is transformed first into a predication DAG and then reduced into a conceptual database query.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "In 1982 SITRA Foundation launched a major project (KIELIKONE) for the study of general computational models for the interpretation of written Finnish. The target application is a Finnish understanding portable database interface. Currently our hierarchical model of language interpretation consists of six processes;", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "INTRODUCTION", |
| "sec_num": "1" |
| }, |
| { |
| "text": "word analysis, lexicalization and disambiguation, sentence parsing, logico-semantic analysis, inference and query adaptation (JAppinen & al. 1988 ).", |
| "cite_spans": [ |
| { |
| "start": 125, |
| "end": 145, |
| "text": "(JAppinen & al. 1988", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "INTRODUCTION", |
| "sec_num": "1" |
| }, |
| { |
| "text": "All intermediate structures until the so called predication DAG represent more and more refined analysis results ( Figure 1 ). The predication DAG is semantically the richest representation in the model. The following semantic processes simplify and modify the representations towards database queries.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 115, |
| "end": 123, |
| "text": "Figure 1", |
| "ref_id": "FIGREF0" |
| } |
| ], |
| "eq_spans": [], |
| "section": "INTRODUCTION", |
| "sec_num": "1" |
| }, |
| { |
| "text": "In our model there is a clean separation between linguistic knowledge and processing mechanisms. The extensive use of specialized knowledge description languages characterizes the different components (Lehtola & al. 1987a) . There is also a hierarchy of representations. When we work in the morphological stratum, the associated knowledge language deals with sets of features.", |
| "cite_spans": [ |
| { |
| "start": 201, |
| "end": 222, |
| "text": "(Lehtola & al. 1987a)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "INTRODUCTION", |
| "sec_num": "1" |
| }, |
| { |
| "text": "In the syntactic stratum trees with feature sets in nodes are the dominating representation. In this paper we outline the com; ational methods used in our logico-semantic stratum which deals with directed acyclic graphs.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "INTRODUCTION", |
| "sec_num": "1" |
| }, |
| { |
| "text": "AWARE -DAG-transformations for Semantic Analysis Aarno Lehtola, Timo Honkela Proceedings of NODALIDA 1987, pages 58-68 DAG-transformations are practical for inferring NL meaning.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "INTRODUCTION", |
| "sec_num": "1" |
| }, |
| { |
| "text": "For instance, they may be used to spesify which NL expressions are near by meaning. Graph rewrite rules may be used to map their syntactic representations into each other. Such rewrite rules would form a meaning-preserving rulebase for a canonizing process.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "INTRODUCTION", |
| "sec_num": "1" |
| }, |
| { |
| "text": "Graph rewrite rules have proved to be convenient also for solving ellipses and anaphoras.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "INTRODUCTION", |
| "sec_num": "1" |
| }, |
| { |
| "text": "-59 - In this paper we present the AWARE DAG-transformation system for modelling logico-semantic analysis. First we outline the architecture of AWARE. Then we demonstrate the power of AWARE-rules by examples which deal with canonization of sentence structures, recognition and marking of semantic predications and solution of certain ellipses. Furthermore we discuss how the resulting predication DAGs can be translated into linear expressions, in our case into queries in an universal relation query language. Next we wiev the knowledge acquisition and rule-base maintenance tools.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "INTRODUCTION", |
| "sec_num": "1" |
| }, |
| { |
| "text": "In the end we evaluate the performance of AWARE.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "KNOWLEDGE PROCESSES REPRESENTATIONS", |
| "sec_num": null |
| }, |
| { |
| "text": "The AWARE-system consists of rule-base maintenance tools and a run-time system ( Figure 2 ). The rule-base is divided into rule packets, which contain rules of An AWARE-description inherits its type definitions from the formal grammar description of the preceding dependency parser (see al. 1985 and a l . 1987) . A user may also define extra types to be used only in transformations and thereafter (Lehtola Sl al. 1987b) . AWAJiE is aware of all information that has been derived by the preceding morphological analysis, the lexicalization and disambiguation process and the dependency parser.", |
| "cite_spans": [ |
| { |
| "start": 287, |
| "end": 299, |
| "text": "al. 1985 and", |
| "ref_id": null |
| }, |
| { |
| "start": 300, |
| "end": 311, |
| "text": "a l . 1987)", |
| "ref_id": null |
| }, |
| { |
| "start": 399, |
| "end": 421, |
| "text": "(Lehtola Sl al. 1987b)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [ |
| { |
| "start": 81, |
| "end": 89, |
| "text": "Figure 2", |
| "ref_id": "FIGREF1" |
| } |
| ], |
| "eq_spans": [], |
| "section": "THE ARCHITECTURE OF AWARE-SYSTEM", |
| "sec_num": "2" |
| }, |
| { |
| "text": "-61 -", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "THE ARCHITECTURE OF AWARE-SYSTEM", |
| "sec_num": "2" |
| }, |
| { |
| "text": "In the following examples we demonstrate the use of transformation rules for different semantic recognition tasks. In many cases the graph transformation reduces into a tree transformation. The first example is very simple, later on we will present more complicated ones.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "HOW TO USE TRANSFORMATION RULES", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Canonization of sentence structure", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "HOW TO USE TRANSFORMATION RULES", |
| "sec_num": "3" |
| }, |
| { |
| "text": "SEKA X 'SEKA \u00ce ---------------------^-----------------------^----------------------\u00ce I I Y 'ETTA Z = > ROOT-ONLY X ' JA Advcp j +-------+---------h I I Y Z", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "HOW TO USE TRANSFORMATION RULES", |
| "sec_num": "3" |
| }, |
| { |
| "text": "In the rule above one defines a trivial situation where the conjunction phrase SEKA-ETTA (ie. BOTH-AND) is turned to an AND phrase. The rule aibove is an example of how ellipses inside sentences can be solved by DAG-transformations.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "HOW TO USE TRANSFORMATION RULES", |
| "sec_num": "3" |
| }, |
| { |
| "text": "The rule is semantically restricted to the case of the form \"the entityl and entity2 of leganpersonA. legalpersonB, ... ,and legalpersonX\".", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "HOW TO USE TRANSFORMATION RULES", |
| "sec_num": "3" |
| }, |
| { |
| "text": "The dependency tree given by the parser has as its root the conjuntion phrase containing the entities. The two (AKO: X 5 ) expressions test that the nouns X2 and X4 coordinate semantically. The conjunction phrase made of the legal persons is syntactically subordinated only to the first entity. The parser does not recognize the ellipsis that also the second entity is in relation with the same legal persons. The meaning of the rule is that when the described situation is recognized the structure XI is made to be shared by both of the entities.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "HOW TO USE TRANSFORMATION RULES", |
| "sec_num": "3" |
| }, |
| { |
| "text": "-64 -One may leave the filling part (r h s ) of a rule partially unspecified. Part of a tree structure is replaced with a call of a wait-rule. Wait-rules are activated afterwards and they loolc for the matching element from the whole tree structure.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Wait-rules for distant bindings", |
| "sec_num": null |
| }, |
| { |
| "text": "Here we demonstrate the use of wait-rules in case of ellipsis. Lets consider the following sentences:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Wait-rules for distant bindings", |
| "sec_num": null |
| }, |
| { |
| "text": "(1) \"Anna yritykset, Joiden liikevaihto on suurempi kuin metsialalla keskimdArin1\" (Give the companies the turnover of which is greater than the averagenal in forestry) (2) \"Anna yritykset, joiden liikevoiton suhde liikevaihtoon on suurempi kuin metsaalalla keskimaa.rinl\" (Give the companies the ratio of profit and turnover of which is greater than the averagenal in forestry)", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Wait-rules for distant bindings", |
| "sec_num": null |
| }, |
| { |
| "text": "Both sentences have an elliptical expression 'the averagenal (turnover/ratio ..) in forestry'. The system cannot locally decide what is the property referred to.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Wait-rules for distant bindings", |
| "sec_num": null |
| }, |
| { |
| "text": "By applying wait-rules the decision can be delayed and the larger context is taken into account.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Wait-rules for distant bindings", |
| "sec_num": null |
| }, |
| { |
| "text": "The following rule matches with the expression 'metsaalalla keskimaarin' (the 64 Proceedings of NODALIDA 1987 See also the label 'AVE' for average which will be attach to the property found.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Wait-rules for distant bindings", |
| "sec_num": null |
| }, |
| { |
| "text": "FIELD X3 (AKO: ala) 'AdjPostAttr Common SG Ad \u00ce -----^I 'KESKIMXXRIN 'Adverbial = > 'HAVE-PROPERTY \u00ce --------------^------\u00ce I 'HAVE-PROPERTY WAIT-PROPERTY 1 AVE I * 1 -------------\u2022 \u00bb - i I 'YRITYS ROOT-ONLY PL X3", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Wait-rules for distant bindings", |
| "sec_num": null |
| }, |
| { |
| "text": "The following WAIT-rules match with our examples (1) and (2). WAIT-rules are contained in a packet of their own. This packet is activated after the packet of the calling transformations has been analyzed. Each call of WAIT-rules may cause only one WAIT-rule to fire. If there is a firing WAIT-rule for each call, the whole dependency structure has been satisfied.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Wait-rules for distant bindings", |
| "sec_num": null |
| }, |
| { |
| "text": "For the production of linear expressions there is an attribute grammar facility.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "RECURSIVE DESCENT TRANSLATION", |
| "sec_num": "4" |
| }, |
| { |
| "text": "Special translation rules specify the way how different DAG constructs are translated into linear expressions and how a collection of such expressions is mapped into a larger one. The idea is that the transfer rules are seen analogously to the cfg-rules in Knuth's attribute grammars (Knuth 1960) . In this Icnowledge acquisition has been very fast by using these tools.", |
| "cite_spans": [ |
| { |
| "start": 284, |
| "end": 296, |
| "text": "(Knuth 1960)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "RECURSIVE DESCENT TRANSLATION", |
| "sec_num": "4" |
| }, |
| { |
| "text": "There is an automatic book-keeping facility that records the input sentences and their analysis results into a corpus file. This recording may be done automatically for all input or it may be invoked by the user. The idea is to collect test material to ensure monotonic improvement of knowledge descriptions.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "RECURSIVE DESCENT TRANSLATION", |
| "sec_num": "4" |
| }, |
| { |
| "text": "After a non trivial change is done in the rulebase, the system runs all test sentences and the results are automatically compared to the previous ones. ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "RECURSIVE DESCENT TRANSLATION", |
| "sec_num": "4" |
| }, |
| { |
| "text": "Compared to certain well known transformation systems (eg. Periphrase of ALPS, MITRE) the AWARE-system offers the following extra properties; ; ^ich type system, (2 processing generalized for directed acyclic graphs, (3) orientation towards dependency structures, (4) powerful tools for knowledge base maintenance, (5) extensive use of graphics to illustrate the operation, (6) attribute grammar facility for translation (7) separate control language (8) lazy evaluation possible using 'wait' rules One of the design objectives in AWARE has been to make it so general that it could be used also in machine translation. Dependency structures have been found a good syntactic representation for machine translation purposes. Our dependency parser al. 1985 and al. 1987) together with AWARE gives interesting prospects for MT.", |
| "cite_spans": [ |
| { |
| "start": 745, |
| "end": 757, |
| "text": "al. 1985 and", |
| "ref_id": null |
| }, |
| { |
| "start": 758, |
| "end": 767, |
| "text": "al. 1987)", |
| "ref_id": "BIBREF11" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "CONCLUSIONS", |
| "sec_num": "7" |
| }, |
| { |
| "text": "Proceedings of NODALIDA 1987", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "The Automatic Transformational Analysis of English Sentences; An I.-npiemen: at io n . Intern", |
| "authors": [ |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Hobbs", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Grishman", |
| "suffix": "" |
| } |
| ], |
| "year": 1976, |
| "venue": "J. Computer Math", |
| "volume": "5", |
| "issue": "", |
| "pages": "267--283", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Hobbs, J . , Grishman, R. (1976): The Automatic Transformational Analysis of English Sentences; An I.-npiemen: at io n . Intern. J. Computer Math. 1976, Section A, Vol. 5, pp. 267-283.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Hierarchical Multilevel Processing Model for Natural Language Database Interface", |
| "authors": [], |
| "year": null, |
| "venue": "Proceedings of the 4th IEEE Conference on Artificial Intelligence A.pplications", |
| "volume": "6", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Hierarchical Multilevel Processing Model for Natural Language Database Interface. Proceedings of the 4th IEEE Conference on Artificial Intelligence A.pplications, San Diego, California, 6 p. (in print).", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "Semantics of Context-Free Languages", |
| "authors": [], |
| "year": null, |
| "venue": "Mathematical Systems Theory", |
| "volume": "2", |
| "issue": "2", |
| "pages": "127--145", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Semantics of Context-Free Languages. Mathematical Systems Theory, vol. 2, no. 2, Springer-Verlag, New York, pp. 127-145.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Language-based Environment for Natural Language Parsing", |
| "authors": [], |
| "year": null, |
| "venue": "Proceedings of the 2nd European Conference of ACL", |
| "volume": "", |
| "issue": "", |
| "pages": "98--106", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Language-based Environment for Natural Language Parsing. Proceedings of the 2nd European Conference of ACL, Geneve, pp. 98-106.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Knowledge Representation Formalisms and Metadescriptions for the Interpretation of Finnish", |
| "authors": [], |
| "year": null, |
| "venue": "Proceedings of the Third Finnish Symposium on Theoretical Computer Science", |
| "volume": "", |
| "issue": "", |
| "pages": "64--87", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Knowledge Representation Formalisms and Metadescriptions for the Interpretation of Finnish. Proceedings of the Third Finnish Symposium on Theoretical Computer Science, pp. 64-87.", |
| "links": null |
| }, |
| "BIBREF11": { |
| "ref_id": "b11", |
| "title": "AWARE -A DAG Production System with Attribute Grammar Facility >*frevised report\u00bb^. Publications of the Kielikone-project, Series B, report 4, Helsinki, 36 p. Valkonen", |
| "authors": [ |
| { |
| "first": "K", |
| "middle": [], |
| "last": "Jappinen", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Lehtola", |
| "suffix": "" |
| }, |
| { |
| "first": "A", |
| "middle": [], |
| "last": "", |
| "suffix": "" |
| } |
| ], |
| "year": 1987, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "AWARE -A DAG Production System with Attribute Grammar Facility >*frevised report\u00bb^. Publications of the Kielikone-project, Series B, report 4, Helsinki, 36 p. Valkonen, K . , JAppinen, H. and Lehtola, A. (1987);", |
| "links": null |
| }, |
| "BIBREF12": { |
| "ref_id": "b12", |
| "title": "Blackboard-based Dependency Parsing", |
| "authors": [], |
| "year": null, |
| "venue": "10th International Joint Conference of Artificial Intelligence", |
| "volume": "", |
| "issue": "", |
| "pages": "700--702", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Blackboard-based Dependency Parsing. 10th International Joint Conference of Artificial Intelligence, Milano, pp. 700-702.", |
| "links": null |
| }, |
| "BIBREF14": { |
| "ref_id": "b14", |
| "title": "Language as a Cognitive Process", |
| "authors": [], |
| "year": null, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Language as a Cognitive Process. Volume I;", |
| "links": null |
| }, |
| "BIBREF15": { |
| "ref_id": "b15", |
| "title": "The MITRE Syntactic Analysis Procedures for Transformational Grammars. Proceedings of the Fall Joint Computer Conference", |
| "authors": [ |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Zwicky", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Friedman", |
| "suffix": "" |
| }, |
| { |
| "first": "B", |
| "middle": [], |
| "last": "Hall", |
| "suffix": "" |
| }, |
| { |
| "first": "D", |
| "middle": [], |
| "last": "Walker", |
| "suffix": "" |
| } |
| ], |
| "year": 1965, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "317--326", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Zwicky, A., Friedman, J., Hall, B., Walker, D. (1965): The MITRE Syntactic Analysis Procedures for Transformational Grammars. Proceedings of the Fall Joint Computer Conference 1965, pp. 317-326.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "num": null, |
| "uris": null, |
| "text": "A stratified model of NL interpretation.", |
| "type_str": "figure" |
| }, |
| "FIGREF1": { |
| "num": null, |
| "uris": null, |
| "text": "The architecture of the AWARE-system.60Proceedings of NODALIDA 1987The rulebase maintenance tools include a rule editor, a rule hypothesis generator and an automatic book keeping system for corpora. The run-time system contains a control language interpreter, a rule tracer, a precompiler and an interpreter for the actual DAG-transformation rules and for the recursive descent compilation rules. AWARE-rules may function locally without paying attention into larger contexts of the processed constructs. They may also cover whole utterances and on global grounds recognize semantic predications which have their parts syntactically distributed. Rules' may amplify themselves by referring to other rules. Also recursive transformations are possible.", |
| "type_str": "figure" |
| }, |
| "FIGREF2": { |
| "num": null, |
| "uris": null, |
| "text": "averagenal in forestry).The rhs of the rule contains a call of a wait-rule.Rule(s) WAIT-PROPEIRTY specifies different ways of expressing a property of something.", |
| "type_str": "figure" |
| }, |
| "FIGREF4": { |
| "num": null, |
| "uris": null, |
| "text": "The list representations are isomorphic with the graphic representations and the user may choose which ones to edit. The graphical editor supports insertion, modification and deletion of rules. The strength of it becomes apparent when one is doing topological changes into rules. Also nodes are easy to manipulate in graphical form.The rule hypothesis generator is integrated with the graphical editor.The idea is that by menus the user chooses one of the already created intermediate results to be the Ihs of his new rule. The hypothesis generator generalizes the match pattern according to certain heuristic rules and automatically forms a rhs pattern. This rhs pattern is a reduced version of the Ihs pattern and the Icnowledge engineer reforms it by the rule editor. The newly created rule is precompiled and ready for use as the user exists the hypothesis generator. The", |
| "type_str": "figure" |
| }, |
| "FIGREF5": { |
| "num": null, |
| "uris": null, |
| "text": "system has proved to be practical in logico-semantic analysis of Finnish and in query synthesis. It is in daily use in our database interface prototype for a Finnish business dataibase. Total processing of a one line long question takes between 5 and 50 seconds of CPU-time on VAX-11/780. The DAG transformations and the conceptual query synthesis consume about 50 percent of this. The size of the rule base is currently almost 400 rules. At the first glance the figures may depress. Taken into account that the complexity of the transformational analysis is very high the time consumption is not surprising. At the moment the rules are precompiled into effective data structures, an inverted index is created out of the match conditions and structure sharing is used to minimize memory consumption. Internal data structures are only partly dynamic for the reason of fast information fetch. In spite of the 67 Proceedings of NODALIDA 1987 preceding measures there are still many ways to improve the performance. The current implementation is in FranzLisp.", |
| "type_str": "figure" |
| }, |
| "TABREF0": { |
| "text": "59Proceedings of NODALIDA 1987 equal priority. Momentarily one or more packets are active. The activation order of packets is specified by a special control language. Each packet has a name and a type. Possible types are 'bottom-up-recursive-scan', 'top-down-recursive-scan , 'wait' and 'transfer'. The type label defines the way the search is carried out.", |
| "content": "<table><tr><td/><td/><td/><td/><td>-60 -</td></tr><tr><td colspan=\"6\">In AWARE-system the DAGs are usually formed from trees by introducing extra</td></tr><tr><td colspan=\"2\">connections.</td><td colspan=\"4\">They have one node as the ancestor of all the other nodes. This</td></tr><tr><td colspan=\"6\">node we call the root node although in general graph terminology that term is not</td></tr><tr><td>used.</td><td colspan=\"5\">Those nodes which have no descendants we call leaves. The edges are</td></tr><tr><td colspan=\"6\">directed out from root nodes towards leaves. The label 'bottom-up-recursive-scan'</td></tr><tr><td colspan=\"6\">makes the system to start the search for possible transformations from the leaves</td></tr><tr><td colspan=\"6\">and to proceed towards the root node. The recursion comes from the fact that</td></tr><tr><td colspan=\"6\">after a succesful transformation the system restarts using the new structure,</td></tr><tr><td colspan=\"6\">'top-down-recursive-scan' works similarly but starts from the root.</td><td>It is</td></tr><tr><td colspan=\"6\">convenient sometimes to let a transformation rule evaluate partially. This is the</td></tr><tr><td colspan=\"6\">case when we try to model distant dependencies eg. certain ellipses and anaphora.</td></tr><tr><td>The</td><td>rules</td><td>in</td><td>'wait'</td><td colspan=\"2\">packets are then used to finnish the incomplete</td></tr><tr><td colspan=\"3\">transformations.</td><td colspan=\"3\">The label 'transfer' marks those packets</td><td>which</td><td>contain</td></tr><tr><td colspan=\"5\">attributed rules for recursive descent compilation.</td><td>These rules are called</td></tr><tr><td colspan=\"3\">transfer-rules.</td><td/><td/></tr></table>", |
| "type_str": "table", |
| "html": null, |
| "num": null |
| }, |
| "TABREF1": { |
| "text": "ie. left-hand-side of the rule, later abbreviated by Ihs) shows the topological and feature conditions which will make the rule to perform a transformation. This rule will be fired if it recognizes a node with the lexeme SEKA and with three subordinates on the right.Thesecond subordinate must have the lexeme ETTA. X and Y are called glue variables and they are used in the construction of a new structure. The right hand side of the rule forms a new tree to substitute the matched one. When the Ihs of the previous rule is matched the glue variable X will have as its value the structure headed by SEKA. The glue variable Y is bound to the first subordinate and the variable Z to the third subordinate. The root node of the new structure is the previous SEKA node (ie. root only of the structure bound to X) with its lexeme changed to JA and with a role label Advcp.The root will have two subordinates.The first is the same as the first subordinate in the matched tree and the second is the same as the third subordinate in the matched tree.In the previous rule we provided the Ihs-nodes with restrictive feature conditions (eg. 'SEKA and 'ETTA) and glue variables. In addition it is possible to provide them with the following directives; ANYNUMBER, ANYORDER, ANYOPTS, ANYDEPS.ANYNUMBER states that a node may have unrestrictedly many subordinates of the specified type. ANYORDEIR lets the subordinates to be located in any mutual order.ANYOPTS states that a node may have unlimited number of optional subordinates.The rhs-nodes may be provided with features to be over-written teg. 'JA and Advcp) and with references to glue variables. The directive R00T-J.^ILY is used id cut out the connections to the subordinates. In the example ru.e it is used to cut out the previous connections of 'SEKA node (referred ty X) so that the new connections introduced in the rhs of the rule would not be overlapping.", |
| "content": "<table><tr><td>KUULUA</td><td>'KUULUA PersForm 3P ANYORDER ANYOPTS REST</td><td/></tr><tr><td colspan=\"3\">Y ' 3 ui) 1 e c t Noun concrete (OR Nom Part) ANYDEPS : > 'POSSESSION Adverbial Y Noun legal_person All ANYDEPS Predication</td></tr><tr><td>Y Actor</td><td>I X</td><td>REST</td></tr><tr><td colspan=\"2\">and means a concrete thing</td><td/></tr><tr><td colspan=\"4\">(2) an adverbial, which is a noun in allative case and means</td></tr><tr><td colspan=\"2\">a legal person</td><td/></tr><tr><td colspan=\"4\">Both of the subordinates may have any subordinates. The root verb KUULUA may have</td></tr><tr><td colspan=\"4\">those subordinates in any order and it may also have unrestricted number of</td></tr><tr><td colspan=\"4\">optional subordinates that are to be bound to the set variable REST.</td></tr><tr><td colspan=\"2\">Solving ellipses</td><td/></tr><tr><td>JA</td><td/><td>X3 ' JA I</td></tr><tr><td/><td colspan=\"2\">X2 (AKO: X5) Noun</td><td>X4 (AKO: X5) Noun</td></tr><tr><td colspan=\"2\">XI ' JA ANYNUMBER</td><td/></tr><tr><td/><td>Gen (AKO: legal_person) Noun</td><td/><td>The rule is composed of a</td></tr><tr><td colspan=\"4\">name and two patterns combined with the rewrite operator ROOT-ONLY X3</td><td>The first pattern</td></tr><tr><td/><td colspan=\"2\">ROOT-ONLY X2</td><td>X4</td></tr><tr><td colspan=\"2\">Proceedings of NODALIDA 1987 Proceedings of NODALIDA 1987 Proceedings of NODALIDA 1987</td><td/><td>61 62 63</td></tr></table>", |
| "type_str": "table", |
| "html": null, |
| "num": null |
| }, |
| "TABREF2": { |
| "text": "65Proceedings of NODALIDA 1987 case we are not recognizing a linear language, but rather a graph. The names of AWARE rules correspond to the nonterminals of Ihs-parts of cfg-rules. The amplifying rule name references in the match patterns of AWARE rules correspond to", |
| "content": "<table><tr><td/><td/><td/><td/><td colspan=\"2\">-66 -</td></tr><tr><td colspan=\"5\">the nonterminals in rhs-parts of cfg-rules.</td><td/><td>In attribute grammar there may be</td></tr><tr><td colspan=\"7\">attribute values associated to nonterminals and passed up and down in the</td></tr><tr><td colspan=\"2\">constituent tree.</td><td colspan=\"5\">In AWARE similar bidirectional attribute value propagation is</td></tr><tr><td colspan=\"7\">possible between rules which together cover the recognized graph. In Knuth's</td></tr><tr><td colspan=\"7\">grammar attributes are properties of nonterminals, in AWARE they are properties of</td></tr><tr><td colspan=\"2\">rule references.</td><td/><td/><td/><td/></tr><tr><td colspan=\"5\">The attribute values may be referred</td><td colspan=\"2\">in</td><td>the</td><td>equations</td><td>associated</td><td>to</td></tr><tr><td colspan=\"7\">transfer-rules. These equations consist of tests, assignments, semantic functions</td></tr><tr><td colspan=\"4\">and structure building functions.</td><td colspan=\"3\">Once a graph is succesfully covered</td><td>with</td></tr><tr><td colspan=\"7\">match-patterns of transfer-rules the equations are instantiated and solved. If</td></tr><tr><td colspan=\"7\">there exists solutions the active transfer rules are satisfied, otherwise the</td></tr><tr><td colspan=\"7\">search proceeds. The solving process brings the wanted token sequences.</td></tr><tr><td colspan=\"4\">5 RULEBASE MAINTENANCE TOOLS</td><td/><td/></tr><tr><td colspan=\"7\">The rule editor lets the user to manipulate his rules in graphical form.</td><td>There</td></tr><tr><td colspan=\"7\">are two user modifiable representations of the rules, a graphic one and a list</td></tr><tr><td colspan=\"7\">expression. The following abstract example demonstrates the parallel use of</td></tr><tr><td colspan=\"4\">graphic and list representations: < rule_name></td><td/><td/></tr><tr><td/><td colspan=\"3\">< tree_variablel>.. <node_propertyl>.. <any_relaxationl>..</td><td/><td/></tr><tr><td/><td colspan=\"2\"><tree_variable2>, <node_property2>, <any_relaxation2>,</td><td colspan=\"3\"><tree_variable4>. <node_property4>. <any_relaxation4>.</td></tr><tr><td/><td>+------I</td><td/><td/><td/><td/></tr><tr><td colspan=\"2\"><ule name 3>.</td><td/><td/><td/><td/></tr><tr><td>= ></td><td colspan=\"2\"><tree_var_ref1>. < insert_j5ropl>. .</td><td/><td/><td/></tr><tr><td colspan=\"4\"><tree_var_ref2>.. <rule_name_ref3> , <insert_prop2>.. (insert_prop3>..</td><td/><td/></tr><tr><td colspan=\"2\">Is the same as:</td><td/><td/><td/><td/></tr><tr><td colspan=\"3\">(< rule_name > ((DEP: (DEP: <rule_name3> <tree_variable2>.. <node_property2>.. <any_relaxation2>. <tree_variablel>.. <node_propertyl>.. <any_relaxationl>.. (DEP: <tree_variable4>.. ( (DEP: <tree__var_ref 2> . . (DEP: (rule name ref3>.. < insert_propl>.. < tree_var_ref1>.. = ></td><td colspan=\"3\">(insert__prop3>. . ) ) ) (insert_prop2>..) (node__property4>,</td><td>(any_relaxation4>, ) )</td></tr><tr><td colspan=\"3\">Proceedings of NODALIDA 1987</td><td/><td/><td/><td>66</td></tr></table>", |
| "type_str": "table", |
| "html": null, |
| "num": null |
| } |
| } |
| } |
| } |
