Monday, August 15, 2011

LDC August 2011 Newsletter

Fall 2011 LDC Data Scholarship Program

Checking in with previous LDC Data Scholarship recipients

Weizmann Institute students are introduced to LDC data

LDC Exhibiting at Interspeech 2011, Florence Italy

New publications:

LDC2011S06

- 2005 Spring NIST Rich Transcription (RT-05S) Evaluation Set -

LDC2011S05

- 2008 NIST Speaker Recognition Evaluation Training Set Part 1 -

LDC2011T09

- Arabic Treebank: Part 2 v 3.1 -


Fall 2011 LDC Data Scholarship Program


Applications are now being accepted through September 15, 2011 for the Fall 2011 LDC Data Scholarship program! The LDC Data Scholarship program provides university students with access to LDC data at no-cost. During the previous two cycles of the program, LDC has awarded no-cost copies of LDC data valued at over US$25,000.

This program is open to students pursuing both undergraduate and graduate studies in an accredited college or university. LDC Data Scholarships are not restricted to any particular field of study; however, students must demonstrate a well-developed research agenda and a bona fide inability to pay. The selection process is highly competitive.


The application consists of two parts:


(1)
Data Use Proposal. Applicants must submit a proposal describing their intended use of the data. The proposal must contain the applicant's name, university, and field of study. The proposal should state which data the student plans to use and contain a description of their research project.

Applicants should consult the
LDC Corpus Catalog for a complete list of data distributed by LDC. Due to certain restrictions, a handful of LDC corpora are restricted to members of the Consortium. Applicants are advised to select a maximum of one to two datasets; students may apply for additional datasets during the following cycle once they have completed processing of the initial datasets and publish or present work in some juried venue.

(2)
Letter of Support. Applicants must submit one letter of support from their thesis adviser or department chair. The letter must confirm that the department or university lacks the funding to pay the full Non-member Fee for the data and verify the student's need for data.

For further information on application materials and program rules, please visit the
LDC Data Scholarship page.

Students can email their applications to the
LDC Data Scholarship program. Decisions will be sent by email from the same address.

The deadline for the Fall 2011 program cycle is September 15, 2011.

Checking in with previous LDC Data Scholarship recipients


LDC introduced the Data Scholarship program during the Fall 2010 semester. Since that time, more than fifteen individual students and student research groups have been awarded no-cost copies of LDC data for their research endeavors. Here is an update on the work of a few of our student recipients:

  • Zachary Brooks - University of Arizona (USA), PhD Candidate, Second Language Acquisition and Teaching. Zachary and his research group were awarded a copy of ECI Multilingual Text (LDC94T5) for research in eye movement tracking by native and non-natives readers. Zachary used the ECI Multilingual Text data to test how second language readers process high and low frequency words in German. The results thus far show that processing a low frequency word can make it harder to process words that come next. The group's bilingual reading processes research is ongoing and Zachary anticipates the need to utilize additional speech and text corpora for future work.
  • Benjamin Martinez Elizalde - Monterrey Institute of Technology and Superior Studies, ITESM (Mexico), graduate student, Computer Science. Benjamin was awarded a copy of Switchboard-1 Release 2 (LDC97S62) and 2002 NIST Speaker Recognition Evaluation (SRE) (LDC2004S04) to support his research in speaker verification modeling. Benjamin's group has prepared a robust Universal Background Model (UBM) and will use the Switchboard and 2002 NIST SRE data to run enrollment and test experiments once a lower baseline is achieved. The Switchboard and SRE data will also be used to prepare the system for the 2012 NIST SRE.
  • Xiaohui Huang - Harbin Institute of Technology (China), Shenzhen Graduate School. Xiaohui and his research group were awarded a copy of TDT5 Topics and Annotations (LDC2006T19) for his work in topic detection and tracking for large-scale web data. Xiaohui extracted 607 documents from TDT5 Multilingual Text (LDC2006T18) and designed a new clustering approach for this data set. TDT5 Topics and Annotations (LDC2006T19 ) was used to label for measuring the precision of clustering. Xiaohui next compared his clustering approach with other text clustering approaches such as k-means and agglomerative hierarchical clustering and was able to achieve good performance. Since his group's method has been validated on small test data sets, next they will look to validate the system using larger text databases and time-series databases.
  • Muhua Zhu - Northeastern University (China), graduate student, Natural Language Processing. Muhua was awarded a copy of Chinese Treebank (CTB) 7.0 (LDC2010T07) to support the development of a high-accuracy Chinese parser. Currently, Muhua is writing a survey paper on Chinese syntactic parsing which studies the performance of different parsing models on the versions of LDC's CTBs. Muhua had expected that parsing accuracy would increase with the additional data from CTB7.0, but accuracy decreased in some instances perhaps because of the inclusion of web text in CTB 7.0. Muhua next plans to use re-ranking methods for syntactic parsing and to extract a Combinatory Categorial Grammar bank (CCG bank) from CTB7.0.


We would like to thanks these students for providing an update on their research. Stay tuned for further reports from other data scholarship recipients.

Weizmann Institute students are introduced to LDC data


LDC data was featured in an introductory speech recognition course at the Weizmann Institute of Science in Rehovot, Israel. Visiting professor, Karen Livescu, of Toyota Technological Institute at Chicago and University of Chicago, Department of Computer Science used several LDC corpora, including CSR-I (WSJ0) Complete (LDC93S6A), Switchboard-1 Release 2 (LDC97S62), TIDIGITS (LDC93S10), and TIMIT Acoustic-Phonetic Continuous Speech Corpus (LDC93S1) for homework and term projects, with a few examples shown during in-class demonstration.

The students enrolled in the course were computer science and mathematics graduate students and all were new to automatic speech recognition (ASR). They had backgrounds in probability, but no significant experience with the probabilistic models used in ASR, such as hidden Markov models and Gaussian mixtures. Livescu provided baseline recognizers that the students could modify, so that even beginning students could focus on specific components, while using real data with results in the literature to compare against.


Since the students were provided with real data that the research community actively uses, students were motivated by the potential for 'real' results if their projects went as planned. As Livescu noted, 'while starting out in ASR from scratch is very difficult, the availability of toolkits and LDC data makes it possible for students in an introductory class to do productive research quite quickly'.


Many thanks to Karen Livescu for sharing an example of how LDC data can be used for teaching purposes.

LDC Exhibiting at Interspeech 2011, Florence Italy

LDC is returning to Europe to participate in Interspeech 2011. The conference will be held from August 28-31 at the Firenze Fiera, conveniently located near the Stazione di Santa Maria Novella. Please stop by LDC’s exhibition booth to say hello and learn more about current happenings at the Consortium.

Interspeech 2011’s theme is ‘Speech Science and Technology for Real Life’. You may learn more about the conference here.

The main conference will feature keynotes on the following topics:

Speaking More Like You: Entrainment in Conversational Speech, Prof. Julia Hirschberg

Neural Representations of Word Meanings, Prof. Tom Mitchell

Honest Signals, Prof. Sandy Pentland

Conference organizers have also scheduled a roundtable discussion for August 31st on ‘Future and Applications of Speech and Language Technologies for the Good Health of Society’ which will be led by Profs. Gabriele Miceli, Björn Granström and Hiroshi Ishiguro.

You are encouraged to keep track of LDC’s Interspeech preparations on our Facebook page. We hope to see you there!

New Publications

(1) 2005 Spring NIST Rich Transcription (RT-05S) Conference Meeting Evaluation Set was developed by LDC and the National Institute of Standards and Technology (NIST). It contains approximately 78 hours of English meeting speech, reference transcripts and other material used in the RT Spring 2005 evaluation. Rich Transcription (RT) is broadly defined as a fusion of speech-to-text (STT) technology and metadata extraction technologies providing the bases for the generation of more usable transcriptions of human-human speech in meetings.

RT-05S included the following tasks in the meeting domain:

Speech-To-Text (STT) - convert spoken words into streams of text

Speaker Diarization (SPKR) - find the segments of time within a meeting in which each meeting participant is talking

Speech Activity Detection (SAD) - detect when someone in a meeting space is talking

Further information about the evaluation is available on the RT-05 Spring Evaluation Website.

The data in this release consists of portions of meeting speech collected between 2001 and 2005 by the IDIAP Research Institute's Augmented Multi-Party Interaction project (AMI), Martigny, Switzerland; International Computer Science Institute (ICSI) at University of California, Berkeley; Interactive Systems Laboratories (ISL) at Carnegie Mellon University (CMU), Pittsburgh, PA; NIST; and Virginia Polytechnic Institute and State University (VT), Blacksburg, VA. Each meeting excerpt contains a head-mic recording for each subject and one or more distant microphone recordings.

Reference transcripts for the evaluation excerpts were prepared by LDC according to its Meeting Recording Careful Transcription Guidelines. Those specifications are designed to provide an accurate, verbatim (word-for-word) transcription, time-aligned with the audio file and including the identification of additional audio and speech signals with special mark-up.

2005 Spring NIST Rich Transcription (RT-05S) Conference Meeting Evaluation Set is distributed on 3 DVD-ROM. 2011 Subscription Members will automatically receive two copies of this corpus. 2011 Standard Members may request a copy as part of their 16 free membership corpora. Non-members may license this data for $2250.

*

(2) 2008 NIST Speaker Recognition Evaluation Training Set Part 1 was developed by LDC and the National Institute of Standards and Technology (NIST). It contains 640 hours of multilingual telephone speech and English interview speech along with transcripts and other materials used as training data in the 2008 NIST Speaker Recognition Evaluation (SRE).

SRE is part of an ongoing series of evaluations conducted by NIST. These evaluations are an important contribution to the direction of research efforts and the calibration of technical capabilities. They are intended to be of interest to all researchers working on the general problem of text independent speaker recognition.

The 2008 evaluation was distinguished from prior evaluations, in particular those in 2005 and 2006, by including not only conversational telephone speech data but also conversational speech data of comparable duration recorded over a microphone channel involving an interview scenario.

The speech data in this release was collected in 2007 by LDC at its Human Subjects Data Collection Laboratories in Philadelphia and by the International Computer Science InstituteMixer 5 project, which was designed to support the development of robust speaker recognition technology by providing carefully collected and audited speech from a large pool of speakers recorded simultaneously across numerous microphones and in different communicative situations and/or in multiple languages. Mixer participants were native English and bilingual English speakers. The telephone speech in this corpus is predominately English; all interview segments are in English. Telephone speech represents approximately 565 hours of the data, where as microphone speech represents the other 75 hours. (ICSI) at the University of California, Berkley. This collection was part of the

The telephone speech segments include excerpts in the range of 8-12 seconds and 5 minutes from longer original conversations. The interview material includes short conversation interview segments of approximately 3 minutes from a longer interview session. English language transcripts in .cfm format were produced using an automatic speech recognition (ASR) system.

2008 NIST Speaker Recognition Evaluation Training Set Part 1 is distributed on 9 DVD-ROM.
2011 Subscription Members will automatically receive two copies of this corpus. 2011 Standard Members may request a copy as part of their 16 free membership corpora. Non-members may license this data for $2000.

*

(3) Arabic Treebank: Part 2 (ATB2) v 3.1 was developed at LDC. It consists of 501 newswire stories from Ummah Press with part-of-speech (POS), morphology, gloss and syntactic treebank annotation in accordance with the Penn Arabic Treebank (PATB) Guidelines developed in 2008 and 2009. This release represents a significant revision of LDC's previous ATB2 publication: Arabic Treebank: Part 2 v 2.0 LDC2004T02.

The ongoing PATB project supports research in Arabic-language natural language processing and human language technology development. The methodology and work leading to the release of this publication are described in detail in the documentation accompanying this corpus and in two research papers: Enhancing the Arabic Treebank: A Collaborative Effort toward New Annotation Guidelines and Consistent and Flexible Integration of Morphological Annotation in the Arabic Treebank.

ATB2 v 3.1 contains a total of 144,199 source tokens before clitics are split, and 169,319 tree tokens after clitics are separated for the treebank annotation. Source texts were selected from Ummah Press news archives covering the period from July 2001 through September 2002.

Arabic Treebank: Part 2 v 3.1 is distributed via web download. 2011 Subscription Members will automatically receive two copies of this corpus on disc. 2011 Standard Members may request a copy as part of their 16 free membership corpora. Non-members may license this data for $4500.