| { |
| "paper_id": "C88-1008", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T12:15:24.172487Z" |
| }, |
| "title": "The E-Framework: A Formalism for Natural Language Processing", |
| "authors": [ |
| { |
| "first": "Annelise", |
| "middle": [], |
| "last": "Bech", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "University of Copenhagen", |
| "location": { |
| "addrLine": "Njalsgade 80", |
| "postCode": "DK-2300", |
| "settlement": "Kbh. S", |
| "country": "Denmark\u00b0" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Anders", |
| "middle": [], |
| "last": "Nygaard", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "University of Copenhagen", |
| "location": { |
| "addrLine": "Njalsgade 80", |
| "postCode": "DK-2300", |
| "settlement": "Kbh. S", |
| "country": "Denmark\u00b0" |
| } |
| }, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "This paper presents the most important characteristic of the new formalism used in Eurotra, the E-Framework. It is a formalism for natural language processing within a stratificational model.", |
| "pdf_parse": { |
| "paper_id": "C88-1008", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "This paper presents the most important characteristic of the new formalism used in Eurotra, the E-Framework. It is a formalism for natural language processing within a stratificational model.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "In the E-Framework, mapping between levels of representation is performed on the basis of transitions between trees and partial descriptions of objects, called descriptors. These descriptors are completed using the definition of the target level.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "The tree to descriptor strategy simplifies the expression of complex two-way relations between text and abstract representations when major structural changes have to be performed on the input. This is illustrated by way of a detailed example showing the interaction of the two formal devices of the E-Framework, the translator and the generator, the basic ideas of which are also briefly described.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "The E-Framework has been implemented and forms the basis of the development of Euretra's pre-industrial prototype of a transfer-based, multi-lingual machine translation system. The E-Framework was developed because of problems in expressing this relation in a simple and perspicuous way in the earlier formalisms used in Eurotra: When gra~m~ars grew to have substantial linguistic coverage, the rules describing the mapping between levels became highly complex and numerous due to interdependence between the linguistic phenomena triggering structural changes in a representational tree.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "The proliferation in the number of rules was mainly due to the fact that mapping rules identified specific target level rules to evaluate the new tree structures, and the complexity of rules was due to the tree to tree transducing strategy used which strictly required specification of full ~ fledged target level tree structures.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Therefore this kind of mapping strategy was abandoned, and a different one requiring only partial target level tree descriptions, has been adopted as the basic method for performing transitions within the E-Framework.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "The E-Framework is the formalism used in Eurotra for implementing a pre-industrial prototype of a stratificational transferbased, multi-lingual machine translation system.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "i. Introduction.", |
| "sec_num": null |
| }, |
| { |
| "text": "The E-Framework has been developed on the basis of experiences with earlier Eurotra formalisms, e.g. <C,A>,T /Arnold, et al. 1986/. However, in at least two respects, it differs significantly from its predecessors and other formalisms which transform tree structures by mapping procedures, e.g. Rosetta (cf. /Appelo, et al. 1987/) , GETA (cf. /Vauquois, et al. 1985/) or Mu (cf. /Nagao, et al. 1985 A grammar defines a level of representation, and a t-module states the relation between source level trees and their corresponding partial descriptions to be completed by the generator at the target level. 3.1. Representational object and descriptor. Before we describe the translator and generator devices in greater detail, it is useful to have a look at the nature of the material they produce. This property of g-~rule application is the key reason why a t-module need only specify very little; the E-Framework generator device has autonomous power to finish a partial description, and the descriptions in the grm~lars are not only used for checking structures, but also for constructing them. In this section, we illustrate how the devices described above interact to perform major structural changes with only a very simple t-module. For the sake of clarity, the example given is somewhat simplified and contains no more than the essential details.", |
| "cite_spans": [ |
| { |
| "start": 295, |
| "end": 330, |
| "text": "Rosetta (cf. /Appelo, et al. 1987/)", |
| "ref_id": null |
| }, |
| { |
| "start": 333, |
| "end": 367, |
| "text": "GETA (cf. /Vauquois, et al. 1985/)", |
| "ref_id": null |
| }, |
| { |
| "start": 371, |
| "end": 398, |
| "text": "Mu (cf. /Nagao, et al. 1985", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "i. Introduction.", |
| "sec_num": null |
| }, |
| { |
| "text": "Consider the source level representational object (i), representing the sentence \"the woman works\", and its target level representation (2). Note that in the source object, the determiner \"the\" is represented only as a feature of the np-node, whereas in the target object, it has a structural representation. Note also that the constituents in the source objects appear in a canonicalized order. These are just some of the abstractions that are made in Eurotra in order to neutralize language specific behaviour, and which have to be undone in Synthesis. The inserted node for the determiner contains the information which enables the generator to find and add the feature for the lexical unit \"the\".", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "i. Introduction.", |
| "sec_num": null |
| }, |
| { |
| "text": "A structural representation of the definiteness feature has been created, and as the original information in (4) has also been validated, the generator has finished its construction of the np.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "i. Introduction.", |
| "sec_num": null |
| }, |
| { |
| "text": "As the dominance relation between the s-node and the v-node given by the descriptor in (4) does not necessarily imply immediate dominance in the finished objectt the generator can create a vp-node by application of rule (6). The original descriptor is completed by rule (5), which also estab -~ lishes the immediate precedence relation between the np-node and the vp-nodeo This gives us the target object depicted in (2), which has a much richer structure than the one provided by the descriptor. As the structure was autonomously created by the generator applying g-rules, the example illustrates how the generator can add lineal precedence information and introduce new nodes to complete the dominance and precedence relations given in a descriptor\u00b0 5o Conclusion\u00b0 The E-Framework itself does not put any restrictions on the number of representational levels, and it does not prescribe any specific distribution of linguistic phenomena over levels0", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "i. Introduction.", |
| "sec_num": null |
| }, |
| { |
| "text": "Expressing linguistic knowledge irl a simple and modular way is obviously crucial for the perspicnity, extensibility and repairability of any large-scale natural language processing system. The tree to descriptor strategy used in the E-Framework makes it possible to express the complex two-way relation between a text and its abstract representation as a set of simple, econ~nical and non-procedural descriptions\u00b0 This has been achieved by giving the generator device the power to use the grammar rules defining a level of representation for other than just checking purposes. As a consequence, the task of the translator has been diminished in that full structural specifications need no longer be stated in the t-modules as well as in the grammars In our application, the advantage of the tree to descriptor strategy is that even when aiming at simple bi-lingual t-modules in a stratifieational translation model, which tends to demand a n~aber of structural changes to be performed within monolingual components, the description of these changes is rather simple.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "i. Introduction.", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [ |
| { |
| "text": " Appelo, L., Fellinger, C., Landsbergen, J\u00b0 (1987) Multilingua; vol. 5, no. 3, 1985 .", |
| "cite_spans": [ |
| { |
| "start": 1, |
| "end": 50, |
| "text": "Appelo, L., Fellinger, C., Landsbergen, J\u00b0 (1987)", |
| "ref_id": null |
| }, |
| { |
| "start": 51, |
| "end": 63, |
| "text": "Multilingua;", |
| "ref_id": null |
| }, |
| { |
| "start": 64, |
| "end": 83, |
| "text": "vol. 5, no. 3, 1985", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "References~", |
| "sec_num": null |
| }, |
| { |
| "text": "The E-F[amework in its present form is the resulL of the joint effort of Eurotra' s framework group, of which the authors are members.I f, however, there are any misrepresentations in this presentation, only we ~re to be blamed~We are particularly indebted to Sergei Perschke, who put forth the basic ideas of the E-Framework.We are g~ateful to Hanne Ruus, Uniw~rsity of Copenhagen, for having scrutinized and criticized the draft versions of our paper. Also, we want to thank our colleague Bente Maegaard for some useful cormments.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Acknowledgements :", |
| "sec_num": null |
| } |
| ], |
| "bib_entries": {}, |
| "ref_entries": { |
| "FIGREF0": { |
| "text": "below gives a schematic overview.", |
| "type_str": "figure", |
| "uris": null, |
| "num": null |
| }, |
| "FIGREF1": { |
| "text": "Figure Io", |
| "type_str": "figure", |
| "uris": null, |
| "num": null |
| }, |
| "TABREF0": { |
| "text": "", |
| "html": null, |
| "type_str": "table", |
| "num": null, |
| "content": "<table><tr><td/><td/><td/><td/><td/><td/><td/><td>3. Overview of the E-Framework.</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td>The E-Framework consists of just two formal</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td>devices,</td><td>namely</td><td>a</td><td>generator</td><td>and</td><td>a</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td>translator. The generator and the translator</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td>are</td><td>abstract</td><td>devices</td><td>which</td><td>interpret</td></tr><tr><td/><td/><td/><td/><td colspan=\"4\">/): In the E-Framework,</td></tr><tr><td colspan=\"8\">there are just two formal devices, and the</td></tr><tr><td colspan=\"8\">mapping from one level to the next is not</td></tr><tr><td colspan=\"3\">performed</td><td>by</td><td colspan=\"3\">traditional</td><td>tree</td><td>to</td><td>tree</td></tr><tr><td colspan=\"8\">transducers. Instead, the mapping strategy</td></tr><tr><td>is</td><td colspan=\"2\">based</td><td>on</td><td colspan=\"3\">transitions</td><td>from</td><td>trees</td><td>to</td></tr><tr><td colspan=\"5\">partial descriptions.</td><td/><td/></tr><tr><td colspan=\"8\">The large number of transfer components in a</td></tr><tr><td colspan=\"8\">multi-lingual translation system obviously</td></tr><tr><td colspan=\"8\">makes it desirable to keep them as small and</td></tr><tr><td colspan=\"2\">simple</td><td>as</td><td colspan=\"3\">possible.</td><td>To</td><td>achieve</td><td>this,</td></tr><tr><td colspan=\"8\">language specific phenomena are neutralized</td></tr><tr><td colspan=\"8\">in Eurotra's interface representation of a</td></tr><tr><td colspan=\"8\">text, and the main burden of work is shifted</td></tr><tr><td colspan=\"8\">to the analysis/synthesis components. This</td></tr><tr><td colspan=\"2\">means</td><td>that</td><td/><td>every</td><td colspan=\"3\">monolingual</td><td>component</td></tr><tr><td colspan=\"8\">expresses a rather complex two-way relation</td></tr><tr><td colspan=\"8\">between abstract representation and actual</td></tr><tr><td colspan=\"2\">text.</td><td/><td/><td/><td/><td/></tr></table>" |
| }, |
| "TABREF1": { |
| "text": "-.-module to p~:-oduce a descripto< for the targut[[.~Velo A t-module is dec]arative; 9t:: consists of t~ ~.~].e~ which describe the relal:ion between source Level representational objects and t~.rget ]eve]. descriptors\u00b0 A t-rule consists of a l~d!t ha~d ~id.e which describes ~t tree at: the source level~ and a ~'ight ha~d sJ.,d~] which states the corresponding desc~cipto)~ at the target ].cveloThe left hand side of a t<cule is, a tree description where the feature bundles to be -rule only defines the descriptor for the part of a source level tree matching the description on its left hand side. The set of t-.rLlles to be used in the production of the descriptor for the full source level tree, J,~ selected by the translator in a top-odown driven match of the tree against the ].eft hand sides\u00b0 ]~f some part of the tree doe~.: not match the left hand side of", |
| "html": null, |
| "type_str": "table", |
| "num": null, |
| "content": "<table><tr><td colspan=\"6\">3 o 2 o Th(~ N::t'ans],al;o~ ,,</td><td/><td/><td/><td/><td/><td/></tr><tr><td>TIll?</td><td colspan=\"3\">tr anslatox</td><td/><td>uses</td><td>a</td><td/><td colspan=\"2\">source</td><td colspan=\"2\">level</td><td>The</td><td>generator</td><td>uses</td><td>the</td><td>level</td><td>definition</td></tr><tr><td colspan=\"4\">representational</td><td colspan=\"8\">object and the specifica-</td><td>contained in a grammar and the descriptors</td></tr><tr><td colspan=\"12\">tions i:0 a tincluded in the target descYiptor are marked</td><td>provided representational objects. by the translator The granmlar itself is declarative; to create it consists of a set of g-rules, each of which is a description of legal parts of represen-tational objects. The generator completes a descriptor output from the translator by repeatedly applying g-ruleso The application of rules is non-d(~terministie; it continues until no more information can be added and all information from the original descriptor has been validated by at least one g-ruleo This</td></tr><tr><td colspan=\"12\">w#[th Jd\u00a2.ntifierso 'J.'he right hal~d side is a</td><td>results</td><td>Jn</td><td>one</td><td>or</td><td>more</td><td>representational</td></tr><tr><td colspan=\"3\">definition</td><td colspan=\"9\">of a descriptor J.n ~he f!orm of</td><td>objects\u00b0</td></tr><tr><td colspan=\"7\">identJ.f;k~_rs,, and domJ.nance</td><td/><td colspan=\"4\">and prucedenee</td></tr><tr><td colspan=\"6\">relations between th(~mo</td><td/><td/><td/><td/><td/><td/><td>If some information</td><td>is not validated,</td><td>the</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>generator cannot produce a legal represen-</td></tr><tr><td colspan=\"8\">Feature bundles represented</td><td colspan=\"4\">by identifiers</td><td>tational object.</td><td>Consequently,</td><td>the attempt</td></tr><tr><td colspan=\"12\">on the r{_ght hand ~'~ide ar[; by default copies</td><td>at completion fails\u00b0</td></tr><tr><td colspan=\"12\">of the source [Level feature bundles,, but t-.</td></tr><tr><td colspan=\"9\">:!:tt].es may be staked to specify</td><td colspan=\"3\">addition~</td><td>The application of a g-rule is a customized</td></tr><tr><td colspan=\"9\">change oc deletion of features\u00b0</td><td/><td/><td/><td>version</td><td>of</td><td>unification</td><td>which</td><td>not</td><td>only</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>permits addition of features,</td><td>but also of</td></tr><tr><td colspan=\"3\">A single tany t-rule</td><td colspan=\"4\">in a t-qnodule,</td><td/><td colspan=\"4\">the translator</td><td>structural bundles.</td><td>information</td><td>and</td><td>entire</td><td>feature</td></tr><tr><td colspan=\"2\">copies</td><td colspan=\"2\">the</td><td colspan=\"3\">dominance</td><td colspan=\"2\">and</td><td colspan=\"3\">precedence</td></tr><tr><td colspan=\"10\">informati.on fxom the source object\u00b0</td><td/><td/></tr><tr><td>Since</td><td colspan=\"5\">tile descriptors</td><td colspan=\"2\">are</td><td colspan=\"2\">only</td><td colspan=\"2\">partial</td></tr><tr><td colspan=\"2\">target</td><td colspan=\"2\">object</td><td colspan=\"4\">descriptions,</td><td colspan=\"2\">they</td><td>must</td><td>be</td></tr><tr><td colspan=\"12\">completed; this is done autonomously by the</td></tr><tr><td colspan=\"5\">generato:,: device\u00b0</td><td>So,</td><td colspan=\"3\">compared</td><td colspan=\"3\">to tree to</td></tr><tr><td colspan=\"5\">tree transducers~</td><td colspan=\"4\">the translator</td><td colspan=\"3\">device of</td></tr><tr><td>the</td><td colspan=\"4\">E-F:,~mnework</td><td>can</td><td>be</td><td colspan=\"4\">characterized</td><td>as</td></tr><tr><td colspan=\"12\">quite weak: The translator can only provide</td></tr><tr><td colspan=\"2\">guidance</td><td colspan=\"2\">for</td><td>the</td><td colspan=\"4\">construction</td><td>of</td><td colspan=\"2\">target</td></tr><tr><td colspan=\"11\">objects; it cannot buJ, ld theln itself\u00b0</td><td/></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>The</td><td>generator</td><td>produces</td><td>representational</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>objects. A representational object is a tree</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>in which the nodes are feat%ire bt~dleso A</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>feature bundle is a set of simple features</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>of the attribute-value type.</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>A representational object is fully described</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>by the feature bundles and the dominance and</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>precedence</td><td>relations</td><td>between</td><td>them\u00b0</td><td>By</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>leaving</td><td>out</td><td>features</td><td>and/or</td><td>information</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>about the relations between feature bundles~</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>we obviously have only a partial description</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>of the object. These partial descriptions we</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>call descriptors,</td><td>and they are what the</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>translator feeds to the generator\u00b0</td></tr></table>" |
| } |
| } |
| } |
| } |
