Monday, June 18, 2012

LDC June 2012 Newsletter

New publications:

LDC at LREC 2012

LDC attended the 8th Language Resource Evaluation Conference (LREC2012), hosted by ELRA, the European Language Resource Association. The conference was held in Istanbul, Turkey and featured a broad range of sessions on language resource and human language technologies research. Fourteen LDC staff members presented current work on a wide range of topics, including handwriting recognition, word alignment, treebanks, machine translation and information retrieval as well as initiatives for synchronizing metadata practices in sociolinguistic data collection.

The LDC Papers page now includes research papers presented at LREC 2012.  Most papers are available for download in pdf format; presentations slides and posters are available for several papers as well. On the Papers page, you can read about LDC's role in resource creation to support handwriting recognition and translation technology (Song et al 2012). LDC is developing resources to support two research programs:  Multilingual Automatic Document Classification, Analysis and Translations (MADCAT) and Open Handwriting Recognition and Translation (OpenHaRT). To support these programs, LDC is collecting handwritten samples of pre-processed Arabic and Chinese data that had previously been translated into English. To date, LDC has collected and annotated over 225,000 handwriting images.

Additionally, you can learn about LDC's efforts to collect and annotate very large corpora of user-contributed content in multiple languages (Garland et al, 2012). For the Broad Operational Language Translation (BOLT) program, LDC is developing resources to support genre-independent machine translation and information retrieval systems. In the current phase of BOLT, LDC is collecting and annotating threaded posts from online discussion forums, targeting at least 500 millions words each in three languages:  English, Chinese, and Egyptian Arabic. A portion of the data undergoes manual, multi-layered linguistic annotation.

As we mark LDC's 20th anniversary, we will feature the work behind these LREC papers as well as other ongoing research in upcoming newsletters.

New publications

(1) Arabic-Dialect/English Parallel Text was developed by Raytheon BBN Technologies (BBN), LDC and Sakhr Software and contains approximately 3.5 million tokens of Arabic dialect sentences and their English translations. 

The data in this corpus consists of Arabic web text as follows:

1. Filtered automatically from large Arabic text corpora harvested from the web by LDC. The LDC corpora consisted largely of weblog and online user groups and amounted to around 350 million Arabic words. Documents that contained a large percentage of non-Arabic or Modern Standard Arabic (MSA) words were eliminated. A list of dialect words was manually selected by culling through the Levantine Fisher (LDC2005S07, LDC2005T03, LDC2007S02 and LDC2007T04) and Egyptian CALLHOME speech corpora (LDC97S45, LDC2002S37, LDC97T19 and LDC2002T38) distributed by LDC. That list was then used to retain documents that contained a certain number of matches. The resulting subset of the web corpora contained around four million words. Documents were automatically segmented into passages using formatting information from the raw data.

2. Manually harvested by Sakhr Software from Arabic dialect web sites.

Dialect classification and sentence segmentation, as needed, and translation into English were performed by BBN through Amazon's Mechanical Turk. Arabic annotators from Mechanical Turk classified filtered passages as being either MSA or one of four regional dialects: Egyptian, Levantine, Gulf/Iraqi or Maghrebi. An additional "General" dialect option was allowed for ambiguous passages. The classification was applied to whole passages rather than individual sentences. Only the passages labeled Levantine and Egyptian were further processed. The segmented Levantine and Egyptian sentences were then translated. Annotators were instructed to translate completely and accurately and to transliterate Arabic names. They were also provided with examples. All segments of a passage were presented in the same translation task to provide context.
Arabic-Dialect/English Parallel Text is distributed via web download. 2012 Subscription Members will automatically receive two copies of this data on disc. 2012 Standard Members may request a copy as part of their 16 free membership corpora. Non-members may license this data for US$2250.

(2) Prague Czech-English Dependency Treebank (PCEDT) 2.0 was developed by the Institute of Formal and Applied Linguistics at Charles University in Prague, Czech Republic. It is a corpus of Czech-English parallel resources translated, aligned and manually annotated for dependency structure, semantic labeling, argument structure, ellipsis and anaphora resolution. This release updates Prague Czech-English Dependency Treebank 1.0 (LDC2004T25) by adding English newswire texts so that it now contains over two million words in close to 100,000 sentences. 

The principal new material in PCEDT 2.0 is the inclusion of the entire Wall Street Journal data from Treebank-3 (LDC99T42). Not included from PCEDT 1.0 are the Reader's Digest material,  the Czech monolingual corpus and  the English-Czech dictionary. Each section is enhanced with a comprehensive manual linguistic annotation in the Prague Dependency Treebank style (LDC2006T01), Prague Dependency Treebank 2.0). The main features of this annotation style are:
-dependency structure of the content words and coordinating and similar structures (function words are attached as their attribute values)
-semantic labeling of content words and types of coordinating structures
-argument structure, including an argument structure ("valency") lexicon for both languages
-ellipsis and anaphora resolution
This annotation style is called tectogrammatical annotation, and it constitutes the tectogrammatical layer in the corpus. Please consult the PCEDT website for more information and documentation.Prague Czech-English Dependency Treebank (PCEDT) 2.0 is distributed on one DVD. 2012 Subscription Members will automatically receive two copies of this data.  2012 Standard Members may request a copy as part of their 16 free membership corpora.  Non-members may license this data for US$100.