In this newsletter:
      

-  LDC
          at ICASSP 2013  -
     

-  Early
          renewing members save on fees  -
     

-  Commercial
          use and LDC data  -
           
      New publications
     

-  GALE
          Arabic-English Parallel Aligned Treebank -- Newswire          -
     

-  MADCAT
          Phase 2 Training Set  -

LDC at ICASSP 2013

LDC will be at ICASSP 2013, the world’s       largest and most comprehensive technical conference focused on       signal processing and its applications. The event will be held       over May 26-31 and we look forward to interacting with members of       this community at our exhibit table and during our poster and       paper presentations:

     

Tuesday, May 28, 15:30 - 17:30, Poster Area D        

   

     

ARTICULATORY TRAJECTORIES FOR           LARGE-VOCABULARY SPEECH RECOGNITION

   

     

Authors: Vikramjit Mitra, Wen Wang, Andreas         Stolcke, Hosung Nam, Colleen Richey, Jiahong Yuan (LDC),         Mark Liberman (LDC)

   

     

Tuesday, May 28, 16:30 - 16:50, Room 2011

   

     

SCALE-SPACE EXPANSION OF           ACOUSTIC FEATURES IMPROVES SPEECH EVENT DETECTION

   

     

Authors: Neville Ryant, Jiahong Yuan,           Mark Liberman (all LDC)

   

     

Wednesday, May 29, 15:20 - 17:20, Poster Area         D

   

     

USING MULTIPLE VERSIONS OF SPEECH INPUT IN           PHONE RECOGNITION

   

     

Authors: Mark Liberman (LDC), Jiahong Yuan         (LDC), Andreas Stolcke, Wen Wang, Vikramjit Mitra

   

Please look for LDC’s exhibition at Booth #53       in the Vancouver Convention Centre. We hope to see you there!

Early renewing members save         on fees

      To date just over 100 organizations have       joined for Membership Year (MY) 2013.   For the sixth straight year,       LDC's early renewal discount program has resulted in significant       savings for our members.  Organizations
      that renewed membership or joined early for MY2013 saved over       US$50,000! MY 2012 members are still eligible for a 5% discount       when renewing for MY2013. This discount will apply throughout       2013.
     
      Organizations joining LDC can take advantage of membership       benefits including free membership year data as well as discounts       on older LDC corpora.  For-profit members can use most LDC data       for commercial applications.  Please visit our Members
        FAQ for further information.

Commercial use and LDC data

      Has your company obtained an LDC database as a       non-member?  For-profit organizations are reminded that an LDC       membership is a pre-requisite for obtaining a commercial license       to almost all LDC databases.  Non-member organizations, including       non-member for-profit organizations, cannot use LDC data to       develop or test products for commercialization, nor can they use       LDC data in any commercial product or for any commercial purpose.        LDC data users should consult corpus-specific license agreements       for limitations on the use of certain corpora. In the case of a       small group of corpora such as American National Corpus (ANC)       Second Release (LDC2005T35), Buckwalter Arabic Morphological       Analyzer Version 2.0 (LDC2004L02), CELEX2 (LDC96L14) and all CSLU       corpora, commercial licenses must be obtained separately from the       owners of the data even if an organization is a for-profit member.

New publications

      (1) GALE
        Arabic-English Parallel Aligned Treebank -- Newswire       (LDC2013T10) was developed by LDC and contains 267,520 tokens of       word aligned Arabic and English parallel text with treebank       annotations. This material was used as training data in the DARPA       GALE  (Global Autonomous Language Exploitation) program. Parallel       aligned treebanks are treebanks annotated with morphological and       syntactic structures aligned at the sentence level and the       sub-sentence level. Such data sets are useful for natural language       processing and related fields, including automatic word alignment       system training and evaluation, transfer-rule extraction, word       sense disambiguation, translation lexicon extraction and cultural       heritage and cross-linguistic studies. With respect to machine       translation system development, parallel aligned treebanks may       improve system performance with enhanced syntactic parsers, better       rules and knowledge about language pairs and reduced word error       rate.

In this release, the source Arabic data was       translated into English. Arabic and English treebank annotations       were performed independently. The parallel texts were then word       aligned. The material in this corpus corresponds to the Arabic       treebanked data appearing in Arabic Treebank: Part 3 v 3.2 (LDC2010T08)       (ATB) and to the English treebanked data in English Translation       Treebank: An-Nahar Newswire (LDC2012T02).

The source data consists of Arabic newswire       from the Lebanese publication An Nahar collected by LDC in 2002.       All data is encoded as UTF-8. A count of files, words, tokens and       segments is below.

      Note: Word count is based on the untokenized Arabic source and       token count is based on the ATB-tokenized Arabic source.

The purpose of the GALE word alignment task was       to find correspondences between words, phrases or groups of words       in a set of parallel texts. Arabic-English word alignment       annotation consisted of the following tasks:

     

Identifying different types of links:         translated (correct or incorrect) and not translated (correct or         incorrect)

   

     

Identifying sentence segments not suitable         for annotation, e.g., blank segments, incorrectly-segmented         segments, segments with foreign languages

   

     

Tagging unmatched words attached to other         words or phrases

   

GALE Arabic-English Parallel Aligned Treebank       -- Newswire is distributed via web download.

2013 Subscription Members will automatically       receive two copies of this data on disc. 2013 Standard Members may       request a copy as part of their 16 free membership corpora.        Non-members may license this data for US$$1750.
     

*

(2) MADCAT Phase 2
        Training Set (LDC2013T09) contains all training data created       by LDC to support Phase 2 of the DARPA MADCAT (Multilingual       Automatic Document Classification Analysis and       Translation)Program. The data in this release consists of       handwritten Arabic documents, scanned at high resolution and       annotated for the physical coordinates of each line and token.       Digital transcripts and English translations of each document are       also provided, with the various content and annotation layers       integrated in a single MADCAT XML output.

The goal of the MADCAT program is to       automatically convert foreign text images into English       transcripts. MADCAT Phase 2 data was collected from Arabic source       documents in three genres: newswire, weblog and newsgroup text.       Arabic speaking scribes copied documents by hand, following       specific instructions on writing style (fast, normal, careful),       writing implement (pen, pencil) and paper (lined, unlined). Prior       to assignment, source documents were processed to optimize their       appearance for the handwriting task, which resulted in some       original source documents being broken into multiple pages for       handwriting. Each resulting handwritten page was assigned to up to       five independent scribes, using different writing conditions.

The handwritten, transcribed documents were       checked for quality and completeness, then each page was scanned       at a high resolution (600 dpi, greyscale) to create a digital       version of the handwritten document. The scanned images were then       annotated to indicate the physical coordinates of each line and       token. Explicit reading order was also labeled, along with any       errors produced by the scribes when copying the text. The       annotation results in GEDI XML output files (gedi.xml), which       include ground truth annotations and source transcripts

The final step was to produce a unified data       format that takes multiple data streams and generates a single       MADCAT XML output file with all required information. The       resulting madcat.xml file has these distinct components: (1) a       text layer that consists of the source text, tokenization and       sentence segmentation, (2)  an
      image layer that consist of bounding boxes, (3) a scribe       demographic layer that consists of scribe ID and partition       (train/test) and (4) a document metadata layer.

This release includes 27,814 annotation files       in both GEDI XML and MADCAT XML formats (gedi.xml and madcat.xml)       along with their corresponding scanned image files in TIFF format.

MADCAT Phase 2 Training Set is distributed on       six DVD-ROM.

2013 Subscription Members will automatically       receive two copies of this data on disc. 2013 Standard Members may       request a copy as part of their 16 free membership corpora.        Non-members may license this data for US$2500.