In this newsletter:

- 2010 Publication Pipeline Update -

- Updated LDC Data Sheets and Papers Pages -

New publications:

- LDC Standard Arabic Morphological Analyzer (SAMA) Version 3.1 -

- NIST 2004 Open Machine Translation (OpenMT) Evaluation -

2010 Publication Pipeline Update

Membership Year (MY) 2010 has included a strong selection of publications including updates to the Arabic and Chinese treebanks, Spanish telephone speech and transcript data from the Fisher collection, and Chinese word n-grams collected from the web .  Please consult our corpus catalog for a full list of publications distributed by LDC. As we are now in the second half of this membership year, we would like to provide information on what publications you can expect for the remainder of MY2010.  Our pipeline includes the following:

Arabic Treebank Part 1 Version 4.1 ~ a revision of Arabic Treebank: Part 1 v 3.0 (POS with full vocalization + syntactic analysis) (LDC2005T02) (ATB1), according to the new Arabic Treebank (ATB) annotation guidelines.  The Arabic Treebank project consists of two distinct phases: (a) Part-of-Speech (POS) tagging which divides the text into lexical tokens, and gives relevant information about each token such as lexical category, inflectional features, and a gloss, and (b) Arabic Treebanking which characterizes the constituent structures of word sequences, provides categories for each non-terminal node, and identifies null elements, co-reference, traces, etc. on-terminal node.   Arabic Treebank Part 1 Version 4.1 represents the manual revision of the syntactic tree annotation in ATB1, the automatic revision and updating of certain part-of-speech tags, and the manual revision of certain targeted POS tags (function words, in particular).  The source data consists of 734 newswire stories from Agence France Presse.

Microsoft Research India POS-Tagged Bengali - to support the task of Part-of-Speech Tagging (POS) and other forms of data-driven linguistic research on Indian languages in general, Microsoft Research India has developed POS labeled data for Hindi, Bengali, and Sanskrit as a part of the Indian Language – Part-of-Speech Tagset (IL-POST) project.  The corpora are based on the IL-POST framework. IL-POST is a POS-tagset framework which has been designed to cover the morph-syntactic details of Indian languages. It supports a three-level hierarchy of Categories, Types and Attributes. The Bengali corpus consists of two different levels of information for each lexical token: (a) lexical category and types, and (b) set morphological attributes and their associated values in the context.  The data consists of 7168 manually annotated sentences (102933 words) targeted to cover written modern standard Bengali from various sources, including blogs, Multikulti, and Wikipedia. .

TRECVID 2006 Keyframes and Transcripts ~ TREC Video Retrieval Evaluation (TRECVID) is sponsored by NIST to promote progress in content-based retrieval from digital video via open, metrics-based evaluation. The keyframes in this release were extracted for use in the NIST TRECVID 2006 Evaluation.  The source data includes approximately 158.6 hours of English, Arabic and Chinese language video data collected by LDC from NBC, CNN, MSN, New Tang Dynasty TV, Phoenix TV, Lebanese Broadcasting Corp.,  and China Central TV.  The keyframes were selected by going to the middle frame of the shot boundary, then parsing left and right of that frame to locate the nearest I-Frame. This then became the keyframe and was extracted. Keyframes have been provided at both the subshot (NRKF) and master shot (RKF) levels.

Uda Walawe Asian Elephant Vocalizations ~ partially-annotated corpus of Asian Elephant communication/vocalization. The data set contains vocalizations primarily by adult female and juvenile Asian elephants. This corpus is intended to enable researchers in acoustic communication of elephants and other species to compare acoustic features and repertoire diversity to this population. Of particular interest is whether there may be regional dialects that differ among Asian elephant populations in the wild and in captivity. A second interest is in whether structural commonalities exist between this and other species that shed light on underlying social and ecological factors shaping communication systems.

2010 Subscription Members are automatically sent all MY2010 data as it is released.  2010 Standard Members are entitled to request 16 corpora for free from MY2010.   Non-members may license most data for research use.

[ top ]

Updated LDC Data Sheets and Papers Pages

LDC is pleased to announce that both our LDC Data Sheets and LDC Papers pages recently have been updated.  On our Data Sheets page, you'll find our growing collection of LDC Data Sheets, each of which highlights a key aspect of the Consortium’s research and development tasks.  Recent additions include a data sheet covering Arabic and English treebanking at LDC and one that provides an overview of LDC's role in sponsored projects.   Our updated papers page contains several papers from LREC2010:  Seventh International Conference on Language Resources and Evaluation, as well as other conferences and journals, dating from 1998 forward.  Most papers are available for download in pdf format; presentations slides and posters are available for several papers as well.

On our Papers page, you can read about LDC's efforts to apply treebank annotation to Arabic broadcast news (Maamouri et al).  Broadcast news (BN) transcript data posed new challenges; for instance, the transcript data included metadata which conveys information in addition to the text of what is being said.   Some forms of metadata were ignored, such as indications of coughs or laughter, while others, such as speech effects including discourse markers and word fragments, were annotated.  Annotators also had to handle indistinct audio signal wherein speech could be heard, but not fully understood, so the words could only be inferred from context rather than from the audio signal.  In these cases, the annotation must convey information not contained in the audio signal that accounts for the annotation in that region.  The improved Arabic Treebank (ATB) pipeline and revised annotation guidelines proved robust enough to carry out this task with few changes. This paper discusses where some adaptation was necessary and describes the overall pipeline as used in the production of BN ATB data.

Additionally, you can learn about LDC's role in resource creation for the Knowledge Base Population (KBP) Track of the Text Analysis Conference (TAC) organized by NIST (Simpson et al).  The KBP track of TAC is a hybrid descendant of the TREC Question Answering track and the Automated Content Extraction (ACE) evaluation program and is designed to support development of systems that are capable of automatically populating a knowledge base with information about entities mined from unstructured text. An important component of the KBP evaluation is the Entity Linking task, where systems must accurately associate text mentions of unknown Person (PER), Organization (ORG), and Geopolitical (GPE) names to entries in a knowledge base. This paper describes the 2009 resource creation efforts, with particular focus on the selection and development of named entity mentions for the Entity Linking task evaluation.
 
[ top ]

New Publications

(1)  The LDC Standard Arabic Morphological Analyzer (SAMA) Version 3.1 was developed by researchers at LDC. SAMA 3.1 is based on, and updates Tim Buckwalter's Buckwalter Arabic Morphological Analyzer (BAMA) 2.0 (LDC2004L02). Since this is the first public release of SAMA, it has been numbered continuously to reflect the continuity between this release and previous BAMA releases.  SAMA 3.1 is a software tool for the morphological analysis of Standard Arabic. SAMA 3.1 considers each Arabic word token in all possible 'prefix-stem-suffix' segmentations, and lists all known/possible annotation solutions, with assignment of all diacritic marks, morpheme boundaries (separating clitics and inflectional morphemes from stems), and all Part-of-Speech (POS) labels and glosses for each morpheme segment. The generated output may then be reviewed by users, and the most appropriate annotation selected from among several choices.

The software layer of SAMA 3.1 relies on a data layer that consists primarily of three Arabic-English lexicon files: prefixes (1328 entries), suffixes (945 entries), and stems (79318 entries representing 40654 lemmas). The lexicons are supplemented by three morphological compatibility tables used for controlling prefix-stem combinations (2497 entries), stem-suffix combinations (1632 entries), and prefix-suffix combinations (1180 entries).

The input format, output format, and data layer of SAMA 3.1 were designed to be backward compatible with BAMA. Incremental changes to the data layer in SAMA have resulted in:

• increased lexicon coverage in the dictionary files
• important changes and additions to the inventory of POS tags
• more possible solutions generated for numerous word forms

The software implementation has been updated to allow more input/output options, installation and configuration options, and smoother incorporation in other Perl tools/services. The structure of the dictionary and morphotactic tables has remained the same (the tables provided with SAMA 3.1 differ from the BAMA 2.0 tables only in size and content, not in format). Logical separation between the software layer and data layer allows the new software tools to be used with previous versions of the tables (instructions are provided with software documentation).  The basic logic that implements the segmentation and analysis look-up for Arabic words is essentially unchanged since BAMA 2.0.

The data layer is now accessed through Berkeley DB, with result-caching enabled by default, leading to improved performance. Various utility scripts have also been added to the software package to facilitate more flexible interaction with tools and data.

LDC Standard Arabic Morphological Analyzer (SAMA) Version 3.1 is distributed via web download.

2010 Subscription Members will automatically receive two copies of this corpus on disc, provided that they have submitted a completed copy of the User License Agreement for LDC Standard Arabic Morphological Analyzer (SAMA) Version 3.1 (LDC2010L01).  2010 Standard Members may request a copy as part of their 16 free membership corpora. As a Members-Only release, LDC Standard Arabic Morphological Analyzer (SAMA) Version 3.1 is not available for non-member licensing.

[ top ]

(2)  NIST 2004 Open Machine Translation (OpenMT) Evaluation is a package containing source data, reference translations, and scoring software used in the NIST 2004 OpenMT evaluation. It is designed to help evaluate the effectiveness of machine translation systems. The package was compiled and scoring software was developed by researchers at NIST, making use of newswire source data and reference translations collected and developed by LDC.

The objective of the NIST OpenMT evaluation series is to support research in, and help advance the state of the art of, machine translation (MT) technologies -- technologies that translate text between human languages. Input may include all forms of text. The goal is for the output to be an adequate and fluent translation of the original.  The 2004 task was to evaluate translation from Chinese to English and from Arabic to English. Additional information about these evaluations may be found at the NIST Open Machine Translation (OpenMT) Evaluation web site.

This evaluation kit includes a single perl script (mteval-v11a.pl) that may be used to produce a translation quality score for one (or more) MT systems. The script works by comparing the system output translation with a set of (expert) reference translations of the same source text. Comparison is based on finding sequences of words in the reference translations that match word sequences in the system output translation.

This corpus consists of 150 Arabic newswire documents, 150 Chinese newswire documents, and 29 Chinese 'prepared speech' documents. For each language, the test set consists of two files: a source and a reference file. Each reference file contains four independent translations of the data set. The evaluation year, source language, test set, version of the data, and source vs. reference file are reflected in the file name.

NIST 2004 Open Machine Translation (OpenMT) Evaluation is distributed via web download.

2010 Subscription Members will automatically receive two copies of this corpus on disc.  2010 Standard Members may request a copy as part of their 16 free membership corpora. Non-members may license this data for US$150.

[ top ]