README FILE FOR LDC CATALOG ID: LDC2025T17 TITLE: LORELEI Sinhala Incident Language Pack AUTHORS: Jennifer Tracey, Dave Graff, Stephanie Strassel, Michael Arrigo, Jonathan Wright, Ann Bies 1.0 Introduction This corpus contains all the text data, annotations and supplemental resources for the Sinhala language that were used in the DARPA LORELEI / LoReHLT 2018 Evaluation, which was conducted by NIST in July of that year. Detailed information about the corpus content is provided in section 3 for each of the partitions ("sets") in the corpus. Combining all sets, the corpus contains approximately 8.1 million words of monoligual text in Sinhala, 70,000 words of monolingual text in English, 6.4 million words of parallel Sinhala-English text, and 50,000 words of data annotated for Entity Discovery and Linking and Situation Frames. The LORELEI (Low Resource Languages for Emergent Incidents) Program is concerned with building Human Language Technology for low resource languages in the context of emergent situations like natural disasters or disease outbreaks. Linguistic resources for LORELEI include Representative Language Packs for over 2 dozen low resource languages, comprising data, annotations, basic natural language processing tools, lexicons and grammatical resources. Representative languages are selected to provide broad typological coverage, while Incident Languages are selected to evaluate system performance on a language whose identity is disclosed at the start of the evaluation, and for which no training data has been provided. This package comprises all of the resources and test set references for Sinhala, which was one of the Program's Incident Languages. The evaluation protocol is based on a scenario in which some unforeseen event (the "incident") triggers a need for humanitarian and logistical support in a region where the predominant language (the "incident language") is one that has received little or no attention as yet in NLP research. The objective for evaluation participants is to provide NLP solutions, including information extraction and machine translation, based only on limited resources and with very little time for development. For more information about LORELEI language resources, see https://www.ldc.upenn.edu/sites/www.ldc.upenn.edu/files/lrec2016-lorelei-language-packs.pdf. Each incident language pack has one or more focal incidents (a natural disaster or other event which might trigger humanitarian needs). To support the evaluation scenario, the evaluation package contents are divided into the following subsets: set0 : "pre-incident" text data and reference resources for the language, including monolingual text, dictionaries, grammars, and parallel or comparable text (in English and the incident language); monolingual and parallel data in this set includes documents published prior to the beginning of the earliest focal incident and/or reference materials for which publication date is not relevant, such as religious materials setE : "post-incident" text data that forms the basis for scoring NLP system performance (using the scoring protocol and software developed by NIST); set E consists of monolingual text, along with reference translations and annotations setS : "post-incident" text data in English, including information that pertains to the incident itself; this was made available to systems after the initial set of scorable outputs had been submitted set1 : supplemental "post-incident" text data, made available after the initial set of scorable outputs had been submitted Each subset is presented as a directory within the data folder at the top-level of the release package. Tools for data processing are provided as part of set0 only, but are applicable to all sets. 2.0 Corpus organization 2.1 Directory Structure The directory structure and contents of the package are summarized below -- paths shown are relative to the base (root) directory of the package: ./docs/README.txt -- this file ./data/set0/ ./data/set0/tools/ -- software for data file format conversion ./data/set0/data/ -- monolingual and parallel text directories ./data/set0/dtds/ -- DTDs for all .xml data formats ./data/set0/docs/ -- lexical and grammatical resources, information about various set0 components and properties ./data/set1/data/ -- monolingual text ./data/set1/docs/ -- information about various set1 components and properties ./data/set2/data/ -- monolingual text ./data/set2/docs/ -- information about various set2 components and properties ./data/setS/data/ -- monolingual text ./data/setS/docs/ -- information about various setS components and properties ./data/setE/data/monolingual_text/ -- monolingual text directory ./data/setE/data/annotation/{eng,il10} {eng,il10}_edl.tab -- table of Entity-Detection-Linking annotations situation_frame/ -- subdirectories for entity mentions, needs, and issues tables twitter_tokenization -- ltf files containing tokenization information for Twitter data ./data/setE/data/translation/ eng/ ltf/ -- ltf.xml files psm/ -- psm.xml files il10/ ltf/ -- ltf.xml files psm/ -- psm.xml files ./data/setE/docs/ -- information about various setE components and properties 2.2 File Name Conventions All monolingual text documents are presented as distinct files with unique file names. For convenience, each file name provides a consistent set of information about the content of the file via a set of fields, as follows: - Language (ENG or IL10) - Genre (2 letters) - Source (6-digit numeric) - Date (8-digit numeric) - Unique Index Number (9 alpha-numeric characters) The language field for all Sinhala documents uses "IL10" instead of the ISO code for the language, as the practice in LORELEI was to refer to incident languages by numeric identifiers to preserve the secrecy of the language name until the start of the evaluation. The date field for news reports represents the date of original publication for the report. Where possible, discussion forum material uses the date when a given discussion thread was initiated. When date information is not available or meaningful for a given document, the date field will reflect (roughly) the time at which the content was initially collected by the LDC, and may be left "incomplete" by setting the "day" field (last two digits) to zero (e.g. "20140900"). Files containing translations from a source language have the source language identified in the "Language Code" field of the file name, and the translation language as a 3-letter extension that immediately follows the main part of the file name. Pairs of corresponding files in "found" translation may have distinct identifier strings (one with IL10 in the initial file name field, and one with ENG in that field), if they were harvested independently of each other and were later found to contain parallel content. The alignment data specifies how the IL10 and ENG files are paired. 2.3 Genres Five genres are represented in this data set, as follows: NW - news and general text harvested from news sites SN - "social network" data (i.e. Twitter) WL - weblog and newsgroup data DF - discussion forum data RF - data from "reference" materials, including religious text, government/NGO information sites, etc. Note that the SN (Twitter) data cannot be distributed directly by LDC, due to the Twitter Terms of Use. The file "docs/twitter_info.tab" (described in Section 7.0 below) provides the necessary information for users to fetch the particular tweets directly from Twitter. 3.0 Content Summary 3.1 Set 0 3.1.1 Monolingual text Document and token counts of monolingual text by genre: Genre N_Docs N_Tokens NW 49102 5391910 WL 1931 923975 3.1.2 Parallel text Parallel text document and token count by genre (counts based on Sinhala documents): Genre N_Docs N_Tokens NW 21167 2930854 RF 5521 3523428 All parallel text is aligned at the sentence level. Parallel for Sinhala and Enlgish can be found in set0/data/translation/, which contains the following structure of subdirectories: found/ sentence_alignment/ eng/{ltf,psm}/ il10/{ltf,psm}/ The "found" data set consists of files from web data sources that had parallel text content in Sinhala and English. Each "leaf" directory in the tree (*/ltf, */psm, sentence_alignment) contains a matched set of data files. Parallel file pairs were identified and harvested automatically, processed into LTF.xml format, and then aligned at the level of "segments" (putative sentences). The alignment files (*.aln.xml) contain one or more "alignment" elements, in which one or more "source" (English) segments is associated with one or more "translation" (Sinhala) segments. It's not assured that all segments in a given (Sinhala or English) data file are accounted for in a given set of alignments. The sentence alignment files contain references to the source document and the translation document (both files can be found in their respective directories), and multiple "alignment" elements, each of which contains one source element and one translation element. The "segments" attribute of the source and translation element contains space delimited segment ids referring to SEG IDs in the corresponding ltf files. NB: We refer to English as the "source" purely as a matter of convenience and consistency across language packs; we do not have confirmable evidence as to the true original language of a given data file. In fact, for some web data sources, it may be the case that documents were translated from some third language into both English and Sinhala. 3.1.3 Lexical and grammatical resources The docs/ directory contains two subdirectories: categoryI_dictionary/ This directory contains the file IL10_categoryI_dictionary.txt, which is a simple two-column tranlation lexicon. The file IL10_CategoryI_dictionaryinfo.pdf contains information about additional bilingual Sinhala-English dictionaries available online that are not distributed in this corpus. categoryII/ LORELEI Incident Language packs were required to contain (pointers to) at least 5 of the following 8 "category II" resources: -- bilingual IL-non-English dictionary -- monolingual IL dictionary -- bilingual grammar (reference grammar of the IL in English) -- monolingual grammar in the IL -- monolingual primer (grammar in the IL of the type used by school children) -- bilingual gazetteer -- monolingual gazetteer in the IL -- monolingual gazetteer in English covering the incident region The categoryII directory contains a pdf file (CategoryII_list.pdf) with additional information and URLs for the resources identified. The english_gazetteer.txt is from Geonames (www.geonames.org) and is a gazetteer for the country of Sri Lanka. 3.2 Set 1 All data in this set is monolingual text in Sinhala from the date of the incident that serves as the focus of the evaluation and later. It may contain some information about the incident, but also contains documents whose content is not relevant to the incident in any way. Genre N_Docs N_Tokens NW 10198 1396556 WL 417 205065 3.3 Set S All data in this set is monolingual text in English from the date of the incident that serves as the focus of the evaluation and later. It may contain some information about the incident, but also contains documents whose content is not relevant to the incident in any way. Genre N_Docs N_Tokens DF 1 5114 NW 20 22267 WL 11 10630 3.5 Set E 3.5.1 Monolingual Text This data set provides monolingual source data for the LORELEI 2018 Evaluation Test Set in Sinhala, as well as a smaller test set of monolingual English data. All data in this set is from the date of the incident that serves as the focus of the evaluation and later. Sinhala set Genre N_Docs N_Tokens NW 372 99568 WL 81 48264 DF 4 1758 SN 1951 25016 English set Genre N_Docs N_Tokens DF 1 537 NW 41 19266 WL 30 16417 SN 1028 17034 Because annotations obey the "full-token rule", meaning that all reference annotation extents coincide with token boundaries as provided by the automatic tokenization process, it was deemed to be important for participants in the evaluation to be able to match the LDC's tokenization for Twitter documents that they retrieved directly from the Twitter API. For this reason, in set E only, the monolingual_text directory contains "scrubbed" ltf for Twitter documents. These ltf documents contain none of the actual tweet content, but instead contain a series of underscores and whitespace which allow users to match the tokenization of the tweet via the character offsets provided in the ltf file. 3.5.2 Translation Human reference translations were provided for a subset of the Sinhala data in the test set. Genre N_Docs N_Tokens DF 4 1758 NW 75 25305 WL 41 23737 The translation/ directory under setE/data/ contains source and reference translation files, as follows: il10/{ltf,psm}/ -- contain 120 ltf/psm pairs eng/{ltf,psm}/ -- contain 120 ltf/psm pairs 3.5.3 Annotation Entity Detection and Linking and Situation Frame annotations were applied to a subset of the Sinhala data in the translation set, and to the English monolingual set, in order to identify "entities", "needs" and "issues" to be detected by systems for scoring purposes. Some of the files that received annotation did not yield annotatable content for one or more annotation types. The docs/ directory contains two file lists, one for IL10 documents and one for English documents, indicating which files were subject to annotation. If a file on one of those lists does not have any annotation present in the annotation directories, that means it did not contain any taggable content for entities, needs, or issues. The annotation/{il10,eng} directories under setE/data/ contains a tab delimited file "{il10,eng}_edl.tab" containing the entity linking annotation, as well as a set of directories containing situation frame annotation as follows: situation_frame/ -- contains subdirectories for each type: issues/ mentions/ needs/ Situation Frame annotation is designed to extract basic information about where needs (such as a need for food) and relevant issues (such as civil unrest) exist; the information is designed to be of the type that would be useful for planning a disaster response effort. For more detailed information about situation frame annotation, see https://www.ldc.upenn.edu/sites/www.ldc.upenn.edu/files/smerp2017.pdf. Guidelines for both the EDL and Situation Frame tasks are included in the docs/ directory of set0. 4.0 Data Formats The data formats described below are common across all sets. 4.1 PSM - Primary Structural Markup When original data has structural markup interleaved with the language content, we apply a filtering process that, in effect, separates the markup and language content into distinct files. The language content (with white-space normalization) goes into an RSD file (see below), and the relevant markup content goes into a corresponding PSM file, which is a simple XML stream comprising tags with attributes, and no other text content of its own. (Configuring the filter for a given data source involves determining which content and markup are "relevant"; the filter eliminates other content and markup as irrelevant, such as ads, navigation menus, etc.) Each PSM file has a single "psm" tag as its root element, and contains one or more "string" tags. Each "string" refers to some span of text in the corresponding RSD file, using "begin_offset" and "char_length" attributes, and assigns a label to it, using a "type" attribute. (Note that offsets and lengths are expressed as Unicode CHARACTER counts, not byte counts.) The "type" attribute tells what sort of markup tag was used in the original data to contain the given string (e.g. "p", "quote", etc.); when sentence segmentation can be done as part of the filtering step, a "string" tag with type="seg" is used to label the span of each detected sentence. Some structural tags in original data contain attributes that may be relevant to language research; for example, in a file that contains a thread from a discussion forum, it's useful to keep track of the dates and authors of posts within the thread. For these cases, the "string" element can contain one ore more "attribute" elements, to preserve the name and value of the given attribute - e.g.: As shown in this example, the "attribute" tag is also used, where appropriate, to assign an ID value (unique within the file) to each string of a given type; this is also used with the "seg"-type strings to assign IDs to detected sentences. PSM files appear in the data/monolingual_text/ and data/translation/ directories of each set. 4.2 LTF - LORELEI Text Format LTF was originally developed for language packs produced in the REFLEX Program ("LCTL Text Format"). This XML format uses structural tags "SEG" and "TOKEN" to mark sentence segmentation and word tokenization of the source data. The full original text of each sentence (SEG) is contained in an "ORIGINAL_TEXT" tag, and each individual word and punctuation string is contained, in order of occurrence, in a sequence of "TOKEN" elements, along with various attributes for each token. Both SEG and TOKEN attributes include character offsets relative to beginning of the raw source data ("RSD" file format, described below), with the offset of the first character being 0. LTF files appear in the data/monolingual_text/ and data/translation/ directories of each set. 4.3 EDL (Entity Detection and Linking) The file "il10_edl.tab" contains all EDL annotations for the IL10 EDL subset. The table contains eight columns, as follows: column 1: system_run_id -- "LDC" column 2: mention_id column 3: mention_text column 4: extents column 5: kb_id -- numeric-ID or "NIL"+numeric, may contain multiple KB links separated by | ("pipe" symbol) column 6: entity_type column 7: mention_type column 8: confidence When column 5 is fully numeric, it is a citation to a numbered entity in LORELEI Entity Detection and Linking Knowledge Base (distributed separately as LDC2020T10); when it consists of "NIL" plus digits, it refers to an entity that is not present in the Knowledge Base, but this label is used consistently for all mentions of the particular entity. Note that for any annotated Twitter documents, text extents have been replaced by underscore ("_") characters to comply with the prohibition against distributing the text of tweets directly. Character offsets can be used to align the annotations with the tweets once the user has downloaded them using Twitter's API. 4.4 Situation Frame Situation frame annotation consists of three parts, each presented as a separate tab-delimited file: entities, needs, and issues. The details of each table are described below. Entities, mentions, need frames, and issue frames all have IDs that follow a standard schema consisting of a prefix designating the type of ID ('Ent' for entities, 'Men' for mentions, and 'Frame' for both need and issue frames), an alphanumeric string identifying the annotation "kit", and a numeric string uniquely identifying the specific entity, mention, or frame within the document. 4.4.1 Entities The grouping of entity mentions into "selectable entities" for situation frame annotation is provided in the mentions/ subdirectory. The table has 8 columns with the following headers and descriptions: column 1: doc_id -- doc ID of source file for the annotation column 2: entity_id -- unique identifier for each grouped entity column 3: mention_id -- unique identifier for each entity mention column 4: entity_type -- one of PER, ORG, GPE, LOC column 5: mention_status -- 'representative' or 'extra'; representative mentions are the ones which have been chosen by the annotator as the representative name for that entity. Each entity has exactly one representative mention. column 6: start_char -- character offset for the start of the mention column 7: end_char -- character offset for the end of the mention column 8: mention_text -- mention string Again, note that for any annotated Twitter documents, text extents have been replaced by underscore ("_") characters to comply with the prohibition against distributing the text of tweets directly. 4.4.2 Needs Annotation of need frames is provided in the needs/ subdirectory. Each row in the table represents a need frame in the annotated document. The table has 13 columns with the following headers and descriptions: column 1: user_id -- user ID of the annotator column 2: doc_id -- doc ID of source file for the annotation column 3: frame_id -- unique identifier for each frame column 4: frame_type -- 'need' column 5: need_type -- exactly one of 'evac' (evacuation), 'food' (food supply), 'search' (search/rescue), 'utils' (utilities, energy, or sanitation), 'infra' (infrastructure), 'med' (medical assistance), 'shelter' (shelter), or 'water' (water supply) column 6: place_id -- entity ID of the LOC or GPE entity identified as the place associated with the need frame; only one place value per need frame, must match one of the entity IDs in the corresponding ent_output.tsv or be 'none' (indicating no place was named) column 7: proxy_status -- 'True' or 'False' column 8: need_status -- 'current', 'future'(future only), or 'past' (past only) column 9: scope -- '1_smallgroup', '2_largegroup', '3_municipality', '4_region', or 'none' column 10: severity -- '1_discomfort', '2_injury', '3_possibledeath', '4_certaindeath', or 'none' column 11: resolution_status -- 'sufficient' or 'insufficient' (insufficient / unknown sufficiency) column 12: reported_by -- entity ID of one or more entities reporting the need; multiple values are comma-separated, must match entity IDs in the corresponding ent_output.tsv or be 'none' column 13: resolved_by -- entity ID of one or more entities resolving the need; multiple values are comma-separated, must match entity IDs in the corresponding ent_output.tsv or be 'none' column 14: description -- string of text entered by the annotator as memory aid during annotation, no requirements for content or language, may be 'none' column 15: kb_id -- numeric-ID or "NIL"+numeric, may contain multiple KB links separated by | ("pipe" symbol) 4.4.3 Issues Annotation of issue frames is provided in the issues/ subdirectory. Each row in the table represents an issue frame in the annotated document. The table has 9 columns with the following headers and descriptions: column 1: user_id -- user ID of the annotator column 2: doc_id -- doc ID of source file for the annotation column 3: frame_id -- unique identifier for each frame column 4: frame_type -- 'issue' column 5: issue_type -- exactly one of 'regimechange' (regime change), 'crimeviolence' (civil unrest or widespread crime), or 'terrorism' (terrorism or other extreme violence) column 6: place_id -- entity ID of the LOC or GPE entity identified as the place associated with the issue frame; only one place value per issue frame, must match one of the entity IDs in the corresponding ent_output.tsv or be 'none' column 7: proxy_status -- 'True' or 'False' column 8: issue_status -- 'current' or 'not_current' column 9: scope -- '1_smallgroup', '2_largegroup', '3_municipality', '4_region', or 'none' column 10: severity -- '1_discomfort', '2_injury', '3_possibledeath', '4_certain death', or 'none' column 11: description -- string of text entered by the annotator as memory aid during annotation, no requirements for content or language, may be 'none' column 12: kb_id -- numeric-ID or "NIL"+numeric, may contain multiple KB links separated by | ("pipe" symbol) 5.0 Software tools included in this release All software tools are provided in the tools/ directory of Set 0. 5.1 "ltf2txt" (source code written in Perl) A data file in ltf.xml format (as described above) can be conditioned to recreate exactly the the "raw source data" text stream (the rsd.txt file) from which the LTF was created. The tools described here can be used to apply that conditioning, either to a directory or to a zip archive file containing ltf.xml data. In either case, the scripts validate each output rsd.txt stream by comparing its MD5 checksum against the reference MD5 checksum of the original rsd.txt file from which the LTF was created. (This reference checksum is stored as an attribute of the "DOC" element in the ltf.xml structure; there is also an attribute that stores the character count of the original rsd.txt file.) Each script contains user documentation as part of the script content; you can run "perldoc" to view the documentation as a typical unix man page, or you can simply view the script content directly by whatever means to read the documentation. Also, running either script without any command-line arguments will cause it to display a one-line synopsis of its usage, and then exit. ltf2rsd.perl -- convert ltf.xml files to rsd.txt (raw-source-data) ltfzip2rsd.perl -- extract and convert ltf.xml files from zip archives 5.2 "twitter-processing" (source code written in Ruby) Due to the Twitter Terms of Use, the text content of individual tweets cannot be redistributed by the LDC. As a result, users must download the tweet contents directly from Twitter and condition/normalize the text in a manner equivalent to what was done by the LDC, in order to reproduce the Sinhala raw text that was used by LDC for annotation. The twitter-processing software provided in the tools/ directory enables users to perform this normalization and ensure that the user's version of the tweet matches the version used by LDC, by verifying that the md5sum of the user-downloaded and processed tweet matches the md5sum provided in the twitter_info.tab file. Users must have a developer account with Twitter in order to download tweets, and the tool does not replace or circumvent the Twitter API for downloading tweets. The twitter_info.tab file provides the twitter download id for each tweet, along with the LORELEI file name assigned to that tweet and the md5sum of the processed text from the tweet. The file "README.md" in the tools/twitter-processing/ directory provides details on how to install and use the source code in this directory in order to condition text data that the user downloads directly from Twitter and produce both the normalized raw text and the segmented, tokenized LTF.xml output. 5.3 Encoding The common framework for text processing in LORELEI includes a “normalization” step, which allows for rectifying variations in orthography and/or punctuation that may occur with some frequency in this or that particular language. For overall simplicity and consistency in processing across all languages, this normalization step is always invoked; in languages such as Sinhala that require no special normalization, this step leaves the data unchanged. 6.0 Documentation included in this release Each set has its own docs directory, but the types of files found there are consistent across the sets, as described below. IL10_IncidentDescription.pdf: provides a description of the incidents that were the focus of the evaluation data set. Found in set0/docs/ only. SimpleNamedEntityGuidelines_IL10_V1.0.pdf, Entity_NAM-NOM_Annotation_Guidelines_English_V2.0.pdf, EntityLinkingGuidelines_V1.3.pdf and SituationFrameGuidelines_V4.0.pdf: guidelines for entity annotation, entity linking, and situation frame annotation. Found in set0/docs/ only. twitter_info.tab: contains tab-separated columns: doc uid, tweet id, normalized md5 of the tweet text, and tweet author id for all tweets in the release. Found in setE only (no other sets contain Twitter data). source_codes.tab: contains tab-separated columns: genre, source code, source name, and base url for each source in the release. Found in all sets. urls.tab: contains tab-separated columns: doc uid and url. Note that the url column is empty for documents from older releases for which the url is not available; they are included here so that the uids column can serve as a document list for the package. Found in all sets. eng_il10_annotated_filelist.txt and il10_annotated_filelist.txt: list of all files annotated for the EDL and Situation Frame tasks. Found in setE only. 7.0 Acknowledgement The authors would like to acknowlege the following contributors to this corpus: Song Chen, Dana Delgado, Neville Ryant, Brian Gainor, Neil Kuster, University of Maryland Applied Research Laboratory for Intelligence and Security (ARLIS), formerly UMD Center for Advanced Study of Language (CASL), and our team of Sinhala annotators. This material is based upon work supported by the Defense Advanced Research Projects Agency (DARPA) under Contract No. HR0011-15-C-0123. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of DARPA. 8.0 Copyright Portions © 2017-2018 Al Jazeera Media Network, © 2016 Associated Newspapers Ltd, © 2017 Associated Newspapers of Ceylon, © 2017-2018 BBC, © 2016-2017 Cable News Network. Turner Broadcasting System, Inc., © 2017 CBS Interactive Inc., © 2017 Colombo Gazette, © 2016 EUMETSAT, © 2016 Fairfax Media, © 2017-2018 Firstpost, © 2017 Global News, a division of Corus Entertainment Inc. Corus News, © 2016- 2018 Gossip Lanka News, © 2018 Gray Television, Inc., © 2016 Guardian News and Media Limited or its affiliated companies, © 2018 HT Media Limited, © 2018 IDN-InDepthNews, © 2016, 2018 Independent.co.uk, © 2017 International Media Investments FZ LLC, © 2018 ITN News, © 2016 Lanka Business Online (Pvt) Ltd., © 2018 Lanka Lead News, © 2016-2018 Lanka News Column, © 2018 LankaPage.com (LLC), © 2018 Lanka Soka Gakkai, © 2018 Lanka Views, © 2017 Living Media India Limited, © 2011-2018 Lotus Technologies (Pvt) Ltd., © 2018 Mediacorp 2018. Mediacorp Pte Ltd., © 2016 Ministry of National Policies and Economic Affairs, © 2017-2018 MTV Channel (Pvt) Ltd, © 2016-2018 Nasdaq, Inc., © 2017 Newsweek LLC, © 2017 Northeastern University, © 2017 Office of the Cabinet of Ministers, Sri Lanka, © 2018 Philadelphia Media Network (Digital), LLC, © 2018 Printline Media Pvt. Ltd., © 2017 Ravaya Newspaper, © 2016-2018 Reuters, © 2016, 2018 Rivira Management Consultant (pvt) Ltd, © 2016 Roar Media, © 2016 SBS, © 2016 Sri Lanka Air Force Information Technology Unit, © 2018 Sri Lanka Broadcasting Corporation, © 2017 Sri Lanka Mirror, © 2014 Telegraph Media Group Limited, © 2016 The Associated Newspapers of Ceylon Ltd., © 2017-2018 The Diplomat, © 2017 The Hindu, © 2017 The Peninsula, © 2017 The Quint, © 2016 Time Inc., © 2018 TRT World, © 2018 ucanews.com, © 2018 USA TODAY, a division of Gannett Satellite Information Network, LLC, © 2018 Watch Tower Bible and Tract Society of Pennsylvania, © 2016-2018 Wijeya Newspapers Ltd, © 2018, 2025 Trustees of the University of Pennsylvania 9.0 Contacts If you have questions about this data release, please contact the following personnel at LDC. Dana Delgado - LORELEI Project Manager