1、raising standards worldwideNO COPYING WITHOUT BSI PERMISSION EXCEPT AS PERMITTED BY COPYRIGHT LAWBSI Standards PublicationBS ISO 24612:2012Language resourcemanagement Linguisticannotation framework (LAF)BS ISO 24612:2012 BRITISH STANDARDNational forewordThis British Standard is the UK implementation
2、 of ISO 24612:2012.The UK participation in its preparation was entrusted to TechnicalCommittee TS/1, Terminology.A list of organizations represented on this committee can beobtained on request to its secretary.This publication does not purport to include all the necessaryprovisions of a contract. Us
3、ers are responsible for its correctapplication. The British Standards Institution 2012. Published by BSI StandardsLimited 2012ISBN 978 0 580 54235 0ICS 01.020Compliance with a British Standard cannot confer immunity fromlegal obligations.This British Standard was published under the authority of the
4、Standards Policy and Strategy Committee on 30 November 2012.Amendments issued since publicationDate Text affectedBS ISO 24612:2012Reference numberISO 24612:2012(E)ISO 2012INTERNATIONAL STANDARD ISO24612First edition2012-06-15Language resource management Linguistic annotation framework (LAF) Gestion
5、des ressources langagires Cadre dannotation linguistique (LAF) BS ISO 24612:2012ISO 24612:2012(E) COPYRIGHT PROTECTED DOCUMENT ISO 2012 All rights reserved. Unless otherwise specified, no part of this publication may be reproduced or utilized in any form or by any means, electronic or mechanical, in
6、cluding photocopying and microfilm, without permission in writing from either ISO at the address below or ISOs member body in the country of the requester. ISO copyright office Case postale 56 CH-1211 Geneva 20 Tel. + 41 22 749 01 11 Fax + 41 22 749 09 47 E-mail copyrightiso.org Web www.iso.org Publ
7、ished in Switzerland ii ISO 2012 All rights reservedBS ISO 24612:2012ISO 24612:2012(E) ISO 2012 All rights reserved iiiContents Page Foreword iv Introduction . v 1 Scope 1 2 Terms and definitions . 1 3 LAF specification . 3 3.1 Overview . 3 3.2 LAF data model 3 3.3 LAF architecture 4 3.4 XML pivot f
8、ormat . 6 3.5 XML elements for the resource header . 11 3.6 Elements in the primary data document header 16 Bibliography 19 BS ISO 24612:2012ISO 24612:2012(E) iv ISO 2012 All rights reservedForeword ISO (the International Organization for Standardization) is a worldwide federation of national standa
9、rds bodies (ISO member bodies). The work of preparing International Standards is normally carried out through ISO technical committees. Each member body interested in a subject for which a technical committee has been established has the right to be represented on that committee. International organ
10、izations, governmental and non-governmental, in liaison with ISO, also take part in the work. ISO collaborates closely with the International Electrotechnical Commission (IEC) on all matters of electrotechnical standardization. International Standards are drafted in accordance with the rules given i
11、n the ISO/IEC Directives, Part 2. The main task of technical committees is to prepare International Standards. Draft International Standards adopted by the technical committees are circulated to the member bodies for voting. Publication as an International Standard requires approval by at least 75 %
12、 of the member bodies casting a vote. Attention is drawn to the possibility that some of the elements of this document may be the subject of patent rights. ISO shall not be held responsible for identifying any or all such patent rights. ISO 24612 was prepared by Technical Committee ISO/TC 37, Termin
13、ology and other language and content resources, Subcommittee SC 4, Language resource management. BS ISO 24612:2012ISO 24612:2012(E) ISO 2012 All rights reserved vIntroduction Effective creation, encoding, processing and management of language resources is facilitated by a single high-level data mode
14、l that supports analysis and design of both annotation schemes and representation formats. This International Standard is designed to support the development and use of computer applications relying on linguistically annotated resources and the exchange of these resources among different application
15、s. BS ISO 24612:2012BS ISO 24612:2012INTERNATIONAL STANDARD ISO 24612:2012(E) ISO 2012 All rights reserved 1Language resource management Linguistic annotation framework (LAF) 1 Scope This International Standard specifies a linguistic annotation framework (LAF) for representing linguistic annotations
16、 of language data such as corpora, speech signal and video. The framework includes an abstract data model and an XML serialization of that model for representing annotations of primary data. The serialization serves as a pivot format to allow annotations expressed in one representation format to be
17、mapped onto another. NOTE Standardization of linguistic data categories that provide annotation content is provided by ISO 12620 and other related International Standards. 2 Terms and definitions For the purposes of this document, the following terms and definitions apply. 2.1 primary data electroni
18、c representation of language data EXAMPLE Text, image, speech signal. Note to entry: Typically, primary data objects are addressed by “locations” in an electronic file, for example, the span of characters comprising a sentence or word, or a point at which a given temporal event begins or ends (as in
19、 speech annotation). More complex data objects may consist of a list or set of contiguous or non-contiguous locations in primary data. 2.2 annotate, verb process of adding linguistic information to primary data (2.1) 2.3 annotation, noun linguistic information added to primary data (2.1), independen
20、t of its representation 2.4 representation format in which the annotation (2.3) is rendered, independent of its content EXAMPLE XML, list or bracketed format, tab-delimited text. 2.5 segmentation annotation annotation (2.3) that delimits linguistic elements that appear in the primary data (2.1) Note
21、 to entry: These elements include (1) continuous segments (appearing contiguously in the primary data), (2) super- and sub-segments, where groups of segments will comprise the parts of a larger segment (e.g. contiguous word segment typically comprise a sentence segment), (3) discontinuous segments (
22、linking continuous segments), and (4) landmarks BS ISO 24612:2012ISO 24612:2012(E) 2 ISO 2012 All rights reserved(e.g. timestamp) that note a point in the primary data. In current practice, segmental information may or may not appear in the document containing the primary data itself. 2.6 linguistic
23、 annotation annotation (2.3) that provides linguistic information about the segments in the primary data (2.1) EXAMPLE Morphosyntactic annotation in which a part of speech and lemma are associated with each segment in the data. Note to entry: The identification of a segment as a word, sentence, noun
24、 phrase, etc. also constitutes linguistic annotation. In current practice, when it is possible to do so, segmentation and identification of the linguistic role or properties of that segment are often combined (e.g. syntactic bracketing, or delimiting each word in the document with an XML element tha
25、t identifies the segment as a word or sentence). 2.7 stand-off annotation annotation (2.3) layered over primary data (2.1) and serialized in a document separate from that containing the primary data Note to entry: Stand-off annotations refer to specific locations in the primary data, by addressing c
26、haracter offsets, elements, etc. to which the annotation applies. Multiple stand-off annotation documents for a given type of annotation can refer to the same primary document (e.g. two different part of speech annotations for a given text). 2.8 annotation document XML document containing annotation
27、s (2.3) 2.9 anchor fixed, immutable position in the primary data (2.1) being annotated (2.2) Note to entry: The medium determines how an anchor is described. For example, text anchors may be character offsets, audio anchors may be time offsets, video anchors may be time offsets or frame indices, ima
28、ge anchors may be coordinates. 2.10 region area in the primary data (2.1) defined by a non-empty, ordered list of anchors (2.9) 2.11 original artefact artefact or annotation (2.3) from which the primary data (2.1) is derived 2.12 graph set of nodes (vertices) V(G) and a set of edges E(G) 2.13 node v
29、ertex terminal point in a graph G, or the intersection of edges in G Note to entry: The terms node and vertex are used interchangeably in this document. 2.14 edge ordered pair of nodes u,v from V(G) Note to entry: The order of the nodes determines the direction of the edge. BS ISO 24612:2012ISO 2461
30、2:2012(E) ISO 2012 All rights reserved 33 LAF specification 3.1 Overview LAF consists of the following. A data model for linguistic annotations and the data to which they apply. An architecture for representing language data and its annotations. An XML serialization of the data model, which describe
31、s the referential structure of annotations associated with language data, consisting of a directed graph or graphs. Nodes in the graph may be linked to regions of primary data. Nodes and edges may be associated with feature structures describing linguistic properties of regions of primary data linke
32、d to reachable nodes. 3.2 LAF data model The LAF data model consists of a) a structure for describing media, consisting of anchors that reference locations in primary data and regions defined in terms of these anchors, b) a graph structure, consisting of nodes, edges and links to regions, and c) an
33、annotation structure for representing annotation content with feature structures. The data model for annotations thus comprises a directed graph referencing n-dimensional regions of primary data as well as other annotations, in which nodes are associated with feature structures providing the annotat
34、ion content. LAF conformance requires that an annotation scheme shall be (or be rendered via the mapping) isomorphic to the LAF data model. NOTE LAF does not include specifications for annotation content categories (i.e. the contents of the associated linguistic phenomena). Figure 1 LAF data model B
35、S ISO 24612:2012ISO 24612:2012(E) 4 ISO 2012 All rights reserved3.3 LAF architecture 3.3.1 Overview Language resources conforming to the LAF architecture consist of the following, described in more detail in 3.3.2 to 3.3.5. One or more primary data documents (see 3.3.2). Any number of annotation doc
36、uments containing nodes, edges and feature structures associated with some or all of the nodes and/or edges in a directed graph. All nodes reference either a base segmentation document (in which case the node has no outgoing edges) or other nodes in the same or other annotation documents via edges.
37、(See 3.3.3). One or more documents defining regions that reference each primary data document, which serve as the base segmentation for annotations (see 3.3.4.) A set of headers, including a resource header describing a collection of primary data documents and annotations, as well as headers for eac
38、h primary data document and each annotation document in the collection (see 3.3.5). It is recommended that whenever possible, each primary data document also be associated with an original artefact containing the source from which the primary data was adapted or extracted for annotation (e.g. the or
39、iginal text in the file format of a particular word processor or file viewer). 3.3.2 Primary data Primary data consists of electronic data in any format, including character (text), image, audio and video. Primary data in a LAF-compliant resources are frozen as “read-only” to preserve the integrity
40、of references to locations within the document or documents. Corrections and modifications to the primary data are treated as annotations and stored in a separate annotation document. Primary data documents containing textual data are encoded in UTF-8 (default) or UTF-16. In the general case, primar
41、y data does not contain markup of any kind. If markup does exist in primary data (e.g. HTML or XML tags), it is treated as a part of the data stream by referring annotations; no distinction is made between markup and other characters in the data when referring to locations in the document. 3.3.3 Ann
42、otation documents Annotation documents contain linguistic information describing primary data. Annotations are always associated with a node in a graph that directly references regions defined over primary data, either directly or via a path through reachable nodes. In the latter case, the annotatio
43、ns are said to be layered over the primary data. LAF recommends representing each of the linguistic layers defined in language resource management, in a separate annotation document for the purposes of exchange. The granularity of the annotation i.e. the smallest information unit to which the annota
44、tion applies is dependent on the application. For example, a single annotation over text may cover a phoneme, word, sentence, paragraph, document, or an entire corpus; for audio it may cover any temporal interval, including a temporal “instant” (timeslot, timestamp, etc.). 3.3.4 References to primar
45、y data Direct reference to locations in primary data is accomplished using anchors. In most cases, these nodes are located between the base units of the primary data representation. Anchors are medium-dependent. Regions of a resource may be defined by specifying the anchors that bound the region. Re
46、gions in artefacts such as an image map or video may be defined in terms of anchors specifying BS ISO 24612:2012ISO 24612:2012(E) ISO 2012 All rights reserved 5one or more coordinates, frame indexes, etc. Regions in audio data may be referenced in terms of anchors that refer to one or more points in
47、 the medium (e.g. an “instant” or “timestamp”). Anchors are represented by n-tuples consisting of sets of spatial and temporal offsets. For example, consider the text “My dog has fleas”: 1 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 |M|y| |d|o|g| |h|a|s| |f|l|e|a|s| The anchors for each word are the following
48、: My: start=0, end=2 dog: start=3, end=6 has: start=7, end=10 fleas: start=11, end=16 A set of regions defined over a document containing primary data need not be contiguous (i.e. there may be portions of the primary data not included in any region), but they should not, in general, overlap. Overlap
49、ping regions should be treated as composed of finer-grained sub-components. For example, two spans, and , can be reconstrued as three spans, a = , b = . Two graph nodes can then be created that reference nodes and , thereby providing the coverage of regions and . Discontiguous regions are referenced by creating nodes referencing each component region and adding a node that is in turn linked to them. The media types included in the resource are defined in the resource header. Each medium is associated with one or
