LMF Lexical Markup Framework E-Book
139,99 €
Mehr erfahren.
- Herausgeber: John Wiley & Sons
- Kategorie: Wissenschaft und neue Technologien
- Sprache: Englisch
The community responsible for developing lexicons for Natural Language Processing (NLP) and Machine Readable Dictionaries (MRDs) started their ISO standardization activities in 2003. These activities resulted in the ISO standard - Lexical Markup Framework (LMF). After selecting and defining a common terminology, the LMF team had to identify the common notions shared by all lexicons in order to specify a common skeleton (called the core model) and understand the various requirements coming from different groups of users. The goals of LMF are to provide a common model for the creation and use of lexical resources, to manage the exchange of data between and among these resources, and to enable the merging of a large number of individual electronic resources to form extensive global electronic resources. The various types of individual instantiations of LMF can include monolingual, bilingual or multilingual lexical resources. The same specifications can be used for small and large lexicons, both simple and complex, as well as for both written and spoken lexical representations. The descriptions range from morphology, syntax and computational semantics to computer-assisted translation. The languages covered are not restricted to European languages, but apply to all natural languages. The LMF specification is now a success and numerous lexicon managers currently use LMF in different languages and contexts. This book starts with the historical context of LMF, before providing an overview of the LMF model and the Data Category Registry, which provides a flexible means for applying constants like /grammatical gender/ in a variety of different settings. It then presents concrete applications and experiments on real data, which are important for developers who want to learn about the use of LMF. Contents 1. LMF - Historical Context and Perspectives, Nicoletta Calzolari, Monica Monachini and Claudia Soria. 2. Model Description, Gil Francopoulo and Monte George. 3. LMF and the Data Category Registry: Principles and Application, Menzo Windhouwer and Sue Ellen Wright. 4. Wordnet-LMF: A Standard Representation for Multilingual Wordnets, Piek Vossen, Claudia Soria and Monica Monachini. 5. Prolmf: A Multilingual Dictionary of Proper Names and their Relations, Denis Maurel, Béatrice Bouchou-Markhoff. 6. LMF for Arabic, Aida Khemakhem, Bilel Gargouri, Kais Haddar and Abdelmajid Ben Hamadou. 7. LMF for a Selection of African Languages, Chantal Enguehard and Mathieu Mangeot. 8. LMF and its Implementation in Some Asian Languages, Takenobu Tokunaga, Sophia Y.M. Lee, Virach Sornlertlamvanich, Kiyoaki Shirai, Shu-Kai Hsieh and Chu-Ren Huang. 9. DUELME: Dutch Electronic Lexicon of Multiword Expressions, Jan Odijk. 10. UBY-LMF - Exploring the Boundaries of Language-Independent Lexicon Models, Judith Eckle-Kohler, Iryna Gurevych, Silvana Hartmann, Michael Matuschek and Christian M. Meyer. 11. Conversion of Lexicon-Grammar Tables to LMF: Application to French, Éric Laporte, Elsa Tolone and Matthieu Constant. 12. Collaborative Tools: From Wiktionary to LMF, for Synchronic and Diachronic Language Data, Thierry Declerck, Pirsoka Lendvai and Karlheinz Mörth. 13. LMF Experiments on Format Conversions for Resource Merging: Converters and Problems, Marta Villegas, Muntsa Padró and Núria Bel. 14. LMF as a Foundation for Servicized Lexical Resources, Yoshihiko Hayashi, Monica Monachini, Bora Savas, Claudia Soria and Nicoletta Calzolari. 15. Creating a Serialization of LMF: The Experience of the RELISH Project, Menzo Windhouwer, Justin Petro, Irina Nevskaya, Sebastian Drude, Helen Aristar-Dry and Jost Gippert. 16. Global Atlas: Proper Nouns, From Wikipedia to LMF, Gil Francopoulo, Frédéric Marcoul, David Causse and Grégory Piparo. 17. LMF in U.S. Government Language Resource Management, Monte George. About the Authors Gil Francopoulo works for Tagmatica (www.tagmatica.com), a company specializing in software development in the field of linguistics and documentation in the semantic web, in Paris, France, as well as for Spotter (www.spotter.com), a company specializing in media and social media analytics.
Sie lesen das E-Book in den Legimi-Apps auf:
Seitenzahl: 348
Veröffentlichungsjahr: 2013
Ähnliche
First published 2013 in Great Britain and the United States by ISTE Ltd and John Wiley & Sons, Inc.
Apart from any fair dealing for the purposes of research or private study, or criticism or review, as permitted under the Copyright, Designs and Patents Act 1988, this publication may only be reproduced, stored or transmitted, in any form or by any means, with the prior permission in writing of the publishers, or in the case of reprographic reproduction in accordance with the terms and licenses issued by the CLA. Enquiries concerning reproduction outside these terms should be sent to the publishers at the undermentioned address:
ISTE Ltd
27-37 St George’s Road
London SW19 4EU
UK
www.iste.co.uk
John Wiley & Sons, Inc.
111 River Street
Hoboken, NJ 07030
USA
www.wiley.com
© ISTE Ltd 2013
The rights of Gil Francopoulo to be identified as the author of this work have been asserted by him in accordance with the Copyright, Designs and Patents Act 1988.
Library of Congress Control Number: 2012955535
British Library Cataloguing-in-Publication Data
A CIP record for this book is available from the British Library
ISBN: 978-1-84821-430-9
Table of Contents
Preface
Chapter 1. LMF – Historical Context and Perspectives
1.1. Introduction
1.2. The context
1.3. The foundations: the Grosseto Workshop and the “X-Lex” projects
1.4. EAGLES and ISLE
1.5. Setting up methodologies and principles for standards
1.6. EAGLES/ISLE legacy
1.7. Interoperability: the keystone of the field
1.8. Bibliography
Chapter 2. Model Description
2.1. Objectives
2.2. The ISO specification
2.3. Means of description
2.4. Core model
2.5. Core model and extension packages
2.6. Morphology extension
2.7. Machine-Readable Dictionary extension
2.8. NLP syntax extension
2.9. NLP semantic extension
2.10. Multilingual notation extension
2.11. NLP morphological pattern extension
2.12. NLP multiword expression pattern extension
2.13. Constraint expression extension
2.14. Conclusion
2.15. Bibliography
Chapter 3. LMF and the Data Category Registry: Principles and Application
3.1. Introduction
3.2. Data category specifications
3.3. The ISOcat Data Category Registry
3.4. LMF and data categories
3.5. Conclusions and future work
3.6. Bibliography
Chapter 4. Wordnet-LMF: A Standard Representation for Multilingual Wordnets
4.1. Introduction
4.2. The KYOTO project
4.3. LMF and Wordnet representation
4.4. Wordnet-LMF
4.5. Conclusions
4.6. Bibliography
Chapter 5. Prolmf: A Multilingual Dictionary of Proper Names and their Relations
5.1. Motivation
5.2. Prolmf basis
5.3. More on lexica and relations in Prolmf
5.4. Conclusion
5.5. Bibliography
5.6. Appendix
Chapter 6. LMF for Arabic
6.1. Introduction
6.2. Modeling of the basic properties
6.3. Modeling of the morphologic extension
6.4. Modeling of the morphologic pattern extension
6.5. Modeling of the syntactic extension
6.6. Modeling of the semantic extension
6.7. Arabic LMF applications
6.8. Implementation
6.9. Conclusion
6.10. Bibliography
Chapter 7. LMF for a Selection of African Languages
7.1. Introduction
7.2. Less-resourced languages
7.3. From published dictionaries to LMF
7.4. Illustrations
7.5. Difficulties and proposals
7.6. Conclusion
7.7. Acknowledgments
7.8. Bibliography
Chapter 8. LMF and its Implementation in Some Asian Languages
8.1. Introduction
8.2. Lexical specification and data categories
8.3. Upper-layer ontology
8.4. Evaluation platform
8.5. Discussion
8.6. Conclusion
8.7. Acknowledgments
8.8. Bibliography
Chapter 9. DUELME: Dutch Electronic Lexicon of Multiword Expressions
9.1. Introduction
9.2. DUELME
9.3. LMF
9.4. The DUELME class model
9.5. Comparison with the LMF Core Package
9.6. Comparison with the LMF NLP multiword expression patterns extension
9.7. Conclusions
9.8. Acknowledgments
9.9. Bibliography
Chapter 10. UBY-LMF – Exploring the Boundaries of Language-Independent Lexicon Models
10.1. Introduction
10.2. Architecture of UBY-LMF
10.3. Language independence of UBY-LMF
10.4. FrameNet in UBY-LMF
10.5. Conclusion
10.6. Acknowledgments
10.7. Bibliography
Chapter 11. Conversion of Lexicon-Grammar Tables to LMF: Application to French
11.1. Motivation
11.2. The Lexicon-Grammar
11.3. Lexical entries
11.4. Subcategorization frames
11.5. Results
11.6. Conclusion
11.7. Bibliography
Chapter 12. Collaborative Tools: From Wiktionary to LMF, for Synchronic and Diachronic Language Data
12.1. Introduction
12.2. Wiktionary
12.3. Related work
12.4. Additional challenges: how to encode the diversity of Wiktionary lexicon in LMF?
12.5. Conclusion
12.6. Bibliography
Chapter 13. LMF Experiments on Format Conversions for Resource Merging: Converters and Problems
13.1. Introduction
13.2. Automatic merging of resources
13.3. Moving from PAROLE Genelex to LMF
13.4. Conclusion
13.5. Availability of resources
13.6. Bibliography
Chapter 14. LMF as a Foundation for Servicized Lexical Resources
14.1. Introduction
14.2. Lexical resources as lexical Web services
14.3. LMF-aware Web services in the RESTful style
14.4. Implementation showcases
14.5. Summary
14.6. Bibliography
Chapter 15. Creating a Serialization of LMF: The Experience of the RELISH Project
15.1. Introduction
15.2. Overview of the RELISH interchange format
15.3. Mapping of equivalent elements
15.4. Complex mappings
15.5. Harmonization of linguistic concepts
15.6. Conclusions and future work
15.7. Bibliography
Chapter 16. Global Atlas: Proper Nouns, From Wikipedia to LMF
16.1. Motivation
16.2. Preparing recognition
16.3. Context of usage
16.4. Ontology of types
16.5. Main source: Wikipedia
16.6. Extraction
16.7. Auxiliary machine learning
16.8. LMF structures
16.9. Example
16.10. Results
16.11. Current limitations and planned improvements
16.12. LMF limitations
16.13. Related work
16.14. Conclusion
16.15. Bibliography
Chapter 17. LMF in U.S. Government Language Resource Management
17.1. Introduction
17.2. Wordscape overview
17.3. The goal
17.4. The importance of data standards
17.5. Language base exchange
17.6. Managing multilingual representations
17.7. Managing grammatical information
17.8. Grammatical information, an MRD example
17.9. Managing LBX schema and document instances
17.10. Data exchange using LBX
17.11. Summary
List of Authors
Index
Preface
Following a long series of successful scientific projects and collaborations, the community responsible for developing lexicons for Natural Language Processing (NLP) and Machine Readable Dictionaries (MRDs) decided to jump start their International Organization for Standardization (ISO) standardization activities in 2003. A group of 60 researchers (cited herein as the “LMF team”) spent 5 years gathering requirements and developing the ideas which resulted in the LMF standard.
The task was not easy because of theoretical divergences, differences in language features and structures, and differences in application types and objectives. Many long (and interesting) discussions took place over many years. An LMF website was used and is still used to share ideas (see www.lexicalmarkupframework.org, associated with a mailing list).
The first important result was the establishment of a common and well-defined terminology. This point seems anecdotal, but, in fact, it was crucial to the success of the project. For instance, among the linguistic community, there is no universal agreement concerning apparently basic terms such as “root”, “word” and “paradigm”. Thus, the LMF team had to select, define and achieve consensus on the “right” set of terms.
The second result was a formal specification for representing lexicons. The LMF team had to identify the common notions shared by all lexicons in order to specify a common skeleton (called the core model) and understand the various requirements coming from different groups of users. These requirements produced a set of eight LMF extensions, which are optional.
The challenge in developing LMF was to follow a narrow line between the need to specify a formal tool that is able to express the large diversity of lexicons, on the one hand, and the need to establish a strict specification that enables the development of hard-to-implement services like lexicon merging, on the other hand.
The LMF specification is a success. Numerous lexicon managers currently use LMF in different languages and contexts. This book is dedicated to reporting on a number of these applications.
It is structured as follows:
Despite this success, we do not claim that LMF is perfect. Indeed, several chapters describe a number of limitations and/or proposals for its improvement.
Chapter 1
LMF – Historical Context and Perspectives
1.1. Introduction
The value of agreeing on standards for lexical resources was first recognized in the 1980s, with the pioneering initiatives in the field of machine-readable dictionaries, and afterwards with EC-sponsored projects ACQUILEX, MULTILEX and GENELEX. Later on, the importance of designing standards for language resources (LR) was firmly established, starting with the Expert Advisory Group for Language Engineering (EAGLES) and International Standards for Language Engineering (ISLE) initiatives. EAGLES drew inspiration from the results of previous major projects, set up the basic methodological principles for standardization and contributed to advancing the common understanding of harmonization issues. ISLE consolidated the uncontroversial basic notion of a lexical metamodel, that is an abstract representation format for lexical entries, the Multilingual ISLE Lexical Entry (MILE). MILE was a general schema for the encoding of multilingual lexical information, and was intended as a common representational layer for multilingual lexical resources. As such, all these initiatives contain the seeds of what later evolved into Lexical Markup Framework (LMF). From a methodological point of view, MILE was based on a very extended survey of common practices in lexical encoding, and was the result of cooperative work toward a consensual view, carried out by several groups of experts worldwide. Both EAGLES and ISLE stressed the importance of reaching a consensus on (linguistic and non-linguistic) “content”, in addition to agreement on formats and encoding issues, and also began to address the needs of content processing and Semantic Web technologies. The recommendations for standards and best practices issued within these projects then became, through the INTERA and mainly the LIRICS project, the International Organization for Standardization (ISO) within the ISO TC37/SC4 committee, where LMF was developed. Thanks to the results of these initiatives that culminated in LMF, there is worldwide recognition that the EU is at the forefront in the areas of LRs and standards. LMF now testifies the full maturity reached by the field of LRs.
1.2. The context
The 1990s saw a widespread acknowledgment of the crucial role covered by LRs in language technology (LT). LR started to be considered as having an infrastructural role, that is as an enabling component of Human Language Technologies (HLTs). HLTs (i.e. natural language processing tools, systems, applications and evaluations) depend on LRs, which also strongly influence their quality and indirectly generate value for producers and users.
This recognition was also shown through the financial support from the European Commission to projects aiming at designing and building different types of LRs. Under the support of US agencies (NSF, DARPA, NSA, etc.) and the EC, LRs were unanimously indicated as themes of utmost priority.
One of the major tenets was the recognition of the essential infrastructural role that LRs play as the necessary common platform on which new technologies and applications must be based. To avoid massive and wasteful duplication of effort, public funding – at least partially – of LR development is critical to ensure public availability (although not necessarily at no cost). A prerequisite to such a publicly funded effort is careful consideration of the needs of the community, in particular the needs of industry. In a multilingual setting such as today’s global economy, the need for standardized wide-coverage LRs is even stronger. Another tenet is the recognition of the need for a global strategic vision, encompassing different types of (and methodologies of building) LR, for an articulated and coherent development of this field.
The infrastructural role of LRs requires that they are (1) designed, built and validated together with potential users (therefore, the need for involving companies), (2) built reusing available “partial” resources, (3) made available to the whole community and (4) harmonized with the resources of other languages (therefore, the importance and the reference to international standards).
The major building blocks to set up an LR infrastructure are presented in [CAL 99]:
Other dimensions were soon added as a necessary complement to achieve the required robustness and data coverage and to assess results obtained with current methodologies and techniques, that is:
Crucial to LR reusability and development was the theme of the definition of operational standards, but the value of agreeing on International Standards was also suddenly recognized as critical. Without standards underlying applications and resources, users of LT would have remained ill-served. The application areas would have continued to be severely hampered and only niche or highly specialized applications would have seen success (e.g. speech aids for the disabled and spelling checkers). In general, it had never been possible to build on the results of past work, whether in terms of resources or the systems that used them.
The significance of standardization was thus recognized, in that it would open up the application field, allow an expansion of activities, sharing of expensive resources, reuse of components and rapid construction of integrated, robust, multilingual language processing environments for end-users.
1.3. The foundations: the Grosseto Workshop and the “X-Lex” projects
During the 1980s there was a dramatic growth in interest in the lexicon. The main reasons for this were, on the one hand, the theoretical developments in linguistics that placed increasing emphasis on the lexical component, and on the other hand the awareness about the wealth of information in lexicons that could be exploited by automatic NLP systems. A turning point in the field was marked by the workshop “On automating the lexicon” held at Marina di Grosseto (Italy) in 1986 [WAL 95], when a pool of actors in the field gathered to establish a baseline for the current state of research and issued a set of recommendations for the sector. The most relevant recommendation – as far as the future LMF is concerned – was the need for a metaformat for the representation of lexical entries, that is an abstract model of a computerized lexicon enabling accommodation of different theories and linguistic models. The following years saw a flourishing of events around this new notion of a “meta-entry”, for instance the workshop on “The Lexical Entry”, held in New York City immediately after Grosseto, and the meeting held in Pisa by the so-called Polytheoretical Group in 1987, where the possibilities of a neutral lexicon were explored [WAL 87].
This has contributed to the creation of a favorable climate for converging toward the common goal of demonstrating the feasibility of large lexicons, which needed to be reusable, polytheoretical and multifunctional. This reflection has led to the definition of the concept of reusability of lexical resources as (1) the possibility of reusing the wealth of information contained in machine-readable dictionaries, by converting their data for incorporation into a variety of different NLP modules; (2) the feasibility of building large-scale lexical resources that can be reused in different theoretical frameworks, for different types of application, and by different users [CAL 91].
The first sense of reusability was clearly addressed by the ACQUILEX project, funded by the European ESPRIT Basic Research Program [BOG 88]. The second sense inspired the Eurotra-7 (ET-7) project, which had the goal of providing a methodology and recommending steps toward the construction of sharable lexical resources [HEI 91].
The need for standards in the second sense of reusability was represented by other initiatives, often publicly funded, such as the EUREKA industrial project GENELEX [GEN 94], which concentrated on a generic model for monolingual reusable lexicons [ANT 94] and the CEC ESPRIT project MULTILEX, whose objective was to devise a model for multilingual lexicons [KHA 93]. GENELEX, with its generic model, fulfilled the requirements of being “theory welcoming”, and having a wide linguistic coverage. A standardized format was designed as a means for encoding information originating from different lexicographic theories, with the aim to make it possible to exchange lexical data and to allow the development of a set of tools for a lexicographic workstation.
These “X-Lex” projects assessed the feasibility of some elementary standards for the description of lexical entries at different levels of linguistic description (phonetic, phonological, etc.) and laid the foundations for all the subsequent standardization initiatives.
It became evident that progress in NLP and speech applications were hampered by a lack of generic technologies and reusable LRs, by a proliferation of different information formats, by variable linguistic specificity of existing information and by the high cost of development of resources. This had to be changed to be able to build on the results of past work, whether in terms of resources or the systems that use them.
1.4. EAGLES and ISLE
EAGLES, started in 1993, is a direct descendant of the previous initiatives, and represented the bridge between them and a number of subsequent projects funded by the EC [CAL 96]. EAGLES was set up to improve the situation of many lexical initiatives, through bringing together representatives of major collaborative European R&D projects in relevant areas, to determine which aspects of our field are open to short-term de facto standardization and to encourage the development of such standards for the benefit of consumers and producers of LT. This work was conducted with a view to providing the foundation for any future recommendations for International Standards that may be formulated under the aegis of ISO.
The aim of EAGLES was to support academic and industrial research and development in HLT by accelerating the provision of standards, common guidelines and best practice recommendations for:
The structure of EAGLES resulted from recommendations made by leading industrial and academic centers, and by the EC Language Engineering strategy committees. More than 30 research centers, industrial organizations, professional associations and networks across the EU provided labor toward the common effort, and more than 100 sites were involved in different EAGLES groups or subgroups. In addition, reports from EC Language Engineering strategy committees had strongly endorsed standardization efforts in language engineering.
Moreover, there was a recognition that standardization work is not only important, but is also a necessary component of any strategic program to create a coherent market, which demands sustained effort and investment. ISLE, a standard-oriented transatlantic initiative under the HLT program, started in 2000, was a continuation of the long-standing European EAGLES initiative [CAL 01, CAL 02].
It is important to note that the work of EAGLES/ISLE must be seen in a long-term perspective. This is especially true for any attempt aiming at standardization in terms of international standards. EAGLES did not and could not result in standards of such an impact: this is the preserve of the ISO. The basic idea behind EAGLES/ISLE work was for the group to act as a catalyst in order to pool concrete results coming from major international/national/industrial projects.
1.5. Setting up methodologies and principles for standards
From a retrospective point of view, it is important to note that EAGLES and its guidelines were the first attempt at defining standards directly responding to commonly perceived needs in order to overcome common problems. In terms of offering workable, compromise solutions, they must be based on a solid platform of accepted facts and acceptable practices.
Since the formation of EAGLES, the work related to standards in the EU has largely been concentrated within this initiative. Related efforts elsewhere were closely linked with EAGLES and feed off it. The Lexicon and Corpus groups’ recommendations were soon applied in a large number of European and national projects. Indeed, EAGLES has acted as a catalyst and testing ground.
EAGLES drew strong inspiration from the results of major projects whose results had contributed to advancing our understanding of harmonization issues. Relevant common practices or upcoming standards were used where appropriate as input to EAGLES/ISLE work. Several LRE projects have been active in contributing comments and in testing EAGLES proposals, thus offering a concrete industry-related setting. Given the amount of industrial participation in EAGLES itself, it is notable that there has been significant advances in Language Engineering Standards, thus re-emphasizing the need to involve industry in such efforts in targeting clearly identified and motivated standardization goals. EAGLES results are to be seen as a first step on the path toward standardization for language engineering purposes.
The major efforts in EAGLES concentrate on the following types of activities:
This method of work has proven useful in the process of reaching consensual de facto standards in a bottom-up approach and was also at the basis of ISLE work.
1.5.1. The MILE methodology: toward LMF
The new awareness created by EAGLES regarding the need to reconcile different approaches to LR building was the direct inspiration for the new concept of “edited union”. This term, coined by Gerald Gazdar in one of the first EAGLES meetings, refers to the idea of conciliating what exists in major lexicons/models/dictionaries. This concept shaped the MILE, that is a highly modular and layered structure, with different levels of recommendations [BER 04]. The MILE was intended as a meta-entry, acting as a common representational layer for multilingual lexical resources. The key ideas underlying the design of a meta-entry can be summarized as follows. Different theoretical frameworks appear to impose different requirements on how lexical information should be represented. One way of tackling the issue of theoretical compatibility stems from the observation that existing representational frameworks mostly differ in the way pieces of linguistic information are mutually implied, rather than in the intrinsic nature of this information.
MILE is the direct ancestor of LMF. We will not describe MILE in detail here, but we will just introduce some of the basic notions at the basis of MILE, because these notions are also important for LMF.
The MILE was designed to meet the following desiderata:
All these requirements served the main purpose of making the lexical meta-entry open to task- and system-dependent parameterization.
The MILE lexicon architecture built, in particular, on the results of the EUREKA GENELEX and the ESPRIT MULTILEX projects, to design a multilingual and multifunctional lexicon model. Such architecture embodied three levels of linguistic information: obligatory, recommended and optional (optional splits furthermore into language independent and language dependent). In this way, the MILE modularity addressed three basic principles: (1) flexibility of the representation, (2) easiness of customization and integration of existing resources and (3) usability by different systems which may need different portions of the data.
The descriptive granularity of the MILE aimed at reaching a maximal decomposition into minimal basic information units. Therefore, small units can be assembled, in different frameworks, according to different (theory/application dependent) generalization principles. For instance, the MILE allowed us to decompose a theory-specific complex notion, such as “synset”, into theory-neutral minimal basic units, such as “senses”, “semantic relations”, where “synonymy” is a particular instance of semantic relation.
On the other side, past EAGLES experience had shown that it was useful in many cases to accept underspecification with respect to recommendations for the representation of some phenomenon (and hierarchical structure of the basic notions, attributes, values, etc.): (1) to allow for agreement on a minimal level of specificity especially in cases where we cannot reach wider agreement and/or (2) enable mapping and comparability of different lexicons, with different granularity, at the minimal common level of specificity (or maximal generality). For example, the work on syntactic subcategorization in EAGLES proved that it was problematic to reach agreement on a few notions, for example it seemed unrealistic to agree on a set of grammatical functions. This has led to an underspecified recommendation, but nevertheless recommendation that was useful.
Another key strategy adopted was the continuous, cyclic interaction between EAGLES and a large number of topic-specific R&D projects and applications.
1.6. EAGLES/ISLE legacy
EAGLES/ISLE, thus, was very influential for the field in providing the mold that shaped the representation of LRs for the years to come. Its heritage gave rise to a burning activity in the development and annotation of LRs, and directly informed the work later on carried out within the ISO Committee devoted to Language Resource Management and Representation. Beside this theoretical legacy, the other main achievement of EAGLES/ISLE was that it provided cohesion to the community engaged in the LR and technology sector.
We identify at least three main footprints. The first two refer to low-level specifications, that is recommendations related to the linguistic categories used for linguistic representation. The third refers to an abstract representation level, as a set of high-level objects used for describing the structural components of LRs.
First, a common core of morphosyntactic distinctions to be encoded in corpora and lexicons. Comparison of how morphosyntactic phenomena are encoded for all EU languages has led to a proposal for encoding a common core of morphosyntactic distinctions in a multilayered structure with applications for all the EU languages (also Eastern Europe), which gives the user more flexibility thus (1) allowing him/her to choose the most appropriate level of granularity and (2) providing a straightforward framework for extensions and updating. These specifications represent the basis on which the data categories of the ISO-12620 were developed within the morphosyntactic Thematic Domain Group, and now embodied in ISOCat.
Second, a common approach to subcategorization in syntax. Comparison of how different systems and theories in different European languages classify and deal with subcategorization phenomena has led to a preliminary classificatory scheme and to the proposal of a set of standardized basic notions for subcategorization, using a frame-based structure.
The EAGLES morphosyntactic guidelines [MON 96, LEE 96] were applied – and consequently tested and evaluated – in the LE-PAROLE Project for the syntactic layer of 12 EU languages, and in a very large number of other national and European projects, such as LRE DELIS, RENOS, CRATER, MECOLB, MULTEXT, COPERNICUS MULTEXT-East and TELRI, MLAP-PAROLE, ESPRIT-ELSNET, French GRACE, German Textcorpora und Erschliessungswerkzeuge, LE-SPARKLE, EUROWORDNET and Italian national projects.
Third, the provision of a proposal for a multilingual and multifunctional model for a lexicon, viewed as a resource out of which to extract specific application lexicons.
EAGLES results in many areas, through their application in numerous projects, became de facto widely adopted standards, and became a well-known trademark and a point of reference for HLT projects and products. EAGLES work toward de facto standards allowed the field of LRs to establish a broad consensus on key issues for some well-established areas, thus providing a key opportunity for further consolidation and a basis for technological advance.
The idea of a standard model for lexicon architecture originated here: the LMF [FRA 06] standard adopts a modular organization to cope with the challenge that actual lexicons differ very much both in complexity and type of encoded information. LMF is made up of a core model, a sort of simple skeleton and various semi-independent packages of notions, used for the various linguistic layers that make up a lexicon.
We wish to highlight here the importance of having both a standard model and core LRs (e.g. corpora and lexicons) also encoded according to the standard – or even more – for applications in the humanities. It may be in fact a big advantage to have the possibility of referring to and adopting available guidelines and possibly reusing available harmonized LRs, thus concentrating research efforts on issues more pertinent to the specific field of interest.
EAGLES results in the Lexicon and Corpus areas were adopted by an impressive number of European – and also national – projects, thus becoming “the de-facto standard” for LR in Europe. This is a very good measure of the impact – and of the need – of such a standardization initiative in the HLT sector. To mention just a few key examples, the LE PAROLE/SIMPLE resources (morphological/syntactic/semantic lexicons and corpora for 12 EU languages) [RUI 98, LEN 99, BEL 00] rely on EAGLES results [EAG 96b, EAG 99], and were then enlarged at the national level through many national projects. The fact that the core PAROLE/SIMPLE resources were enlarged to real-size lexicons within national projects in at least eight EU countries was a big step toward a very large infrastructural platform of harmonized lexicons in Europe, sharing the same model. Moreover, the ELRA Validation Manuals for Lexicons [UND 97] and Corpora [BUR 97] are based on EAGLES guidelines.
1.6.1. Lessons learned for standard design
From a retrospective point of view, the experience gained in those years was influential, in particular from the point of view of the leading principles that must guide the standardization process. Standards must emerge from state-of-the-art developments and as such they are not to be imposed. Consolidation of a standard’s proposal must be viewed, by necessity, as a slow process and, by definition, as a non-innovative action. The process of standardization, although by its own nature not intrinsically innovative, must – and actually does – proceed shoulder to shoulder with the most advanced research. Since EAGLES involved many bodies active in EU–US NLP and speech projects, close collaboration with these projects was assured and, significantly, in many cases, free manpower has been contributed by the projects, which is a sign of both the commitment of these groups/companies and of the crucial importance they place on reusability issues.
After the phase of putting proposals forward, it must comprise a cyclical phase involving external groups and projects with:
This long process has the merit of making new areas for consensus emerge while promoting consciousness of their stability in the community at the same time.
Finally, one of the targets of standardization is to create a common parlance among the various actors (both of the scientific and the industrial R&D community) in the field of computational lexical semantics and multilingual lexicons, so that synergies will be enhanced, commonalities strengthened and resources and findings usefully shared. In other terms, the process of standard definition undertaken by EAGLES, and by the ISLE enterprise in particular, represents an essential interface between advanced research in the field of multilingual lexical semantics and the practical task of developing resources for HLT systems and applications. It is through this interface that the crucial trade-off between research practice and applicative needs can actually be achieved.
1.6.2. Moving closer to LMF
After the EAGLES/ISLE experience, and the subsequent use of their results in so many projects, the ground was ready to move from standards and best practices directly emerging from projects and research groups to an international, coordinated and structured effort ratified by standardization organizations. A new work item proposal was issued by the ISO/TC37 US delegation in Summer 2003. In Fall 2003, the French delegation issued a technical proposition for a data model dedicated to NLP lexicons. In early 2004, the ISO/TC37 committee decided to form a common ISO project with Nicoletta Calzolari (CNR-ILC Italy) as convenor and Gil Francopoulo (Tagmatica France) and Monte George (ANSI USA) as editors. This was the start of the LMF (ISO-24613). From 2005 to 2007, the ISO activities were carried out in parallel with the EU eContent project LIRICS (http://lirics.loria.fr).
The goals of this project were to provide ISO ratified standards for LT to enable the exchange and reuse of multilingual LRs, and at the same time to facilitate the implementation of these standards for end-users. Through an Industry Advisory Group and demonstration workshops, LIRICS managed to gain full industry support and input for the standard’s development. The LIRICS Consortium brought together leading experts in the field of NLP and related standards development via participation in ISO committees and National Standardization committees, closely following the procedures established by ISO.
The first step in developing LMF was to design an overall framework based on the general features of existing lexicons and to develop a consistent terminology to describe the components of those lexicons. The following step was the actual design of a comprehensive model that best represented all of the lexicons in detail. A large panel of 60 experts contributed a wide range of requirements for LMF that covered many types of NLP lexicon. The editors of LMF worked closely with the panel of experts to identify the best solutions and reach a consensus on the design of LMF. Special attention was paid to the morphology in order to provide powerful mechanisms for handling problems in several languages that were known as difficult to handle. A total of 13 versions have been written, dispatched (to the national nominated experts), commented upon and discussed during various ISO technical meetings. After 5 years of work, the editors arrived at a coherent UML model. In conclusion, LMF should be considered a synthesis of the state of the art in NLP lexicon field.
1.7. Interoperability: the keystone of the field
Since the first attempts, and after LMF, we have made big steps forward with respect to interoperability. Today, open, collaborative, shared data are at the core of a sound language strategy. Standards are fundamental to exchange, preserve, maintain and integrate data and LRs, to achieve interoperability in general, and they are an essential basis of any LR infrastructure.
What was called “reusability” in the past has evolved today into “interoperability”. Interoperability means the ability of information and communication systems to exchange data and to enable the sharing of information and knowledge. To make the notion of interoperability operational, we need to set up an interoperability framework. This can be described as a dynamic environment of language (and other) standards and guidelines, where different standards are coherently related to one another and guidelines clearly describe how the specifications may be applied to various types of resource. Such a framework should be internally coherent, that is a series of specific standards should continue to exist, but they should form a coherent system (i.e. coherence among the various standard specifications must be ensured so that they can “speak” to each other). The framework should also be dynamic, in the sense that standards must be conceived as dynamic, because they need to follow and adapt to new technologies and domains of application. As the LT field is expanding, standards need to be periodically revised, updated and integrated in order to keep pace with technological advancement.
An interoperability framework is also intended to support the provision of language service interoperability. Enterprises nowadays seem to need such a language strategy, and to be key players they must rely on interoperability, otherwise they are out of business. A recent report by TAUS [TAU 11] states that: “The lack of interoperability costs the translation industry a fortune”, where the highest price is paid mainly for adjusting data formats.
The community and funding agencies need to join forces to drive forward the use of existing and emerging standards, at least in the areas where there is some degree of consensus. The only way to ensure useful feedback to improve and advance is to use these standards on a regular basis. It will thus be even more important to enforce and promote the use of standards at all stages, from basic standardization for less-resourced languages (such as orthography normalization and transcription of oral data) to more complex areas (such as syntax and semantics).
However, enforcing standards cannot be a purely top-down process. It must be backed by information about contributions from different user communities. As most users are not very concerned about whether or not they are using standards, there should be easy-to-use tools that help them apply standards while hiding most of the technicalities. The goal would be to have standards operating in the background as “intrinsic” properties of the LT or the more generic tools that people/end-users use.
But true content interoperability is still far away. We may have solved the issue of formats, of inventories of linguistic categories for the various linguistic layers, but we have not solved the problem of relating senses, which would allow automatic integration of semantic resources. This is a challenge for the following years, and a prerequisite for both a true Lexical Web and a credible Semantic Web.
1.8. Bibliography
[ANT 94] ANTONI-LAY M.H., FRANCOPOULO G., ZAYSSER L., “A generic model for reusable lexicons: the genelex project”, in OSTLER N., ZAMPOLLI A. (eds), Literary and Linguistic Computing, vol. 9, no. 1, pp. 47–54, 1994.
[BEL 00] BEL N., BUSA F., CALZOLARI N., GOLA E., LENCI A., MONACHINI M., OGONOWSKI A., PETERS I., PETERS W., RUIMY N., VILLEGAS M., ZAMPOLLI A., “SIMPLE: a general framework for the development of multilingual lexicons”, LREC Proceedings, Athens, 2000.
[BER 04] BERTAGNA F., LENCI A., MONACHINI M., CALZOLARI N., “Content interoperability of lexical resources: open issues and ‘MILE’ perspectives”, Proceedings of the 4th International Conference on Language Resources and Evaluation, Lisbon, Portugal, ELRA, 2004.
[BOG 88] BOGURAEV B., BRISCOE E.J., CALZOLARI N., CATER A., MEIJS W., ZAMPOLLI A., Acquisition of lexical knowledge for natural language processing systems (ACQUILEX), Proposal for ESPRIT Basic Research Actions No. 3030, Cambridge, UK, 1988.
[BUR 97] BURNARD L., BAKER P., MCENERY A., WILSON A., An analytic framework for the validation of language corpora, Report of the ELRA Corpus Validation Group, Paris, 1997.
[CAL 91] CALZOLARI N., “Lexical databases and textual corpora: perspectives of integration for a Lexical Knowledge Base”, in ZERNIK U. (ed.), Lexical Acquisition: Using on-Line Resources to Build a Lexicon, Erlbaum Ass., New York, 1991.
[CAL 01] CALZOLARI N., LENCI A., ZAMPOLLI A., BEL N., VILLEGAS V., THURMAIR G., “The ISLE in the Ocean. Transatlantic standards for Multilingual Lexicons (with an eye to machine translation)”, Proceedings of MT Summit VIII, Santiago De Compostela, Spain, 2001.
[CAL 96] CALZOLARI N., MC NAUGHT J., ZAMPOLLI A., Eagles Final Report: EAGLES Editors’ Introduction, EAG-EB-EI, Pisa, 1996.
[CAL 99] CALZOLARI N., ZAMPOLLI A., “Harmonised large-scale syntactic/semantic lexicons: a European multilingual infrastructure”, MT Summit Proceedings, Singapore, pp. 358–365, 1999.
[CAL 02] CALZOLARI N., ZAMPOLLI A., LENCI A., “Towards a standard for a multilingual lexical entry: the EAGLES/ISLE initiative”, in GELBUKH A.F. (ed.), Computational Linguistics and Intelligent Text Processing, 3rd International Conference, CICLing 2002, Mexico City, Mexico, Springer, pp. 264–279, 17–23 February, 2002.
[EAG 96a] EAGLES, Evaluation of natural language processing systems, Final Report, Center for Sprogteknologi, Copenhagen, 1996.
[EAG 96b] EAGLES Subcategorization Standards, EAGLES, CNR-ILC, Pisa, 1996.
[EAG 99] EAGLES Recommendations on Semantic Encoding, EAGLES, CNR-ILC, Pisa, 1999.
[FRA 06] FRANCOPOULO G., GEORGE M., CALZOLARI N., MONACHINI M., BEL N., PET M., SORIA C., “Lexical markup framework (LMF)”, Proceedings of LREC 2006, Genova, Italy, ELRA, pp. 233–236, 2006.
[GEN 94] GENELEX, Report on the Semantic Layer, Project EUREKA GENELEX, Version 2.1, 1994.
[GIB 97] GIBBON D., MOORE R., WINSKI R., Handbook of Standards and Resources for Spoken Language Systems, Mouton de Gruyter, Berlin, New York, 1997.
[HEI 91] HEID U., MCNAUGHT J., EUROTRA-7 study: feasibility and project definition study on the reusability of lexical and terminological resources in computerised applications, Final report, 1991.
[KHA 93] KHATCHADOURIAN H., MODIANO N., “Use and importance of standard in electronic dictionaries: the compilation approach for lexical resources”, Literary and Linguistic Computing, vol. 98, Oxford University Press, 1993.
[LEE 96] LEECH G., WILSON A., Recommendations for the morphosyntactic annotation of corpora, EAG-TCWG-MAC/R, Lancaster, 1996.
[LEN 99] LENCI A., BUSA F., RUIMY N., GOLA E., MONACHINI M., CALZOLARI N., ZAMPOLLI A., Linguistic specifications, SIMPLE Deliverable D2.1., CNR-ILC and University of Pisa, 1999.
[MON 96] MONACHINI M., CALZOLARI N., Synopsis and Comparison of Morphosyntactic Phenomena Encoded in Lexicons and Corpora: A Common Proposal and Applications to European Languages, EAGLES, CNR-ILC, Pisa, 1996.
[RUI 98] RUIMY N., CORAZZARI O., GOLA E., SPANU A., CALZOLARI N., ZAMPOLLI A., “The European LE-PAROLE project: the Italian syntactic Lexicon”, Proceedings of the 1st International Conference on Language Resources and Evaluation, Granada, Spain, ELRA, pp. 241–248, 1998.
[TAU 11] TAUS, Report on a TAUS research about translation interoperability, 25 February, 2011.
[UND 97] UNDERWOOD N., NAVARRETTA C., A draft manual for the validation of Lexica, Final ELRA Report, Copenhagen, 1997.
[WAL 87] WALKER D., ZAMPOLLI A., CALZOLARI N. (eds), Towards a polytheoretical lexical data base, CNR-ILC Report, Pisa, 1987.
[WAL 95] WALKER D., ZAMPOLLI A., CALZOLARI N. (eds), Automating the Lexicon: Research and Practice in a Multilingual Environment, Oxford University Press, Oxford, 1995.
Chapter written by Nicoletta CALZOLARI, Monica MONACHINI and Claudia SORIA.
