Some really useful links to corpora on the web

Stay up-to-date with the corpora@cs email list

You may want to subscribe to our corpora mailing list (ask Benjamin Van Durme to add you to the list). This list distributess information about our corpus environment. You can also use it ask/answer questions about the local corpus environment. There are several groups that use corpora. At the very least you will find useful tools and corpora in /p/hlp/ and /p/nl/ (the natural language processing group in CS). There is also a corpus inventory and RSS feed. You can use a web browser or email client with built-in RSS capabilities (e.g. Mozilla Firefox, Mozilla Thunderbird), a web-based aggregator (e.g. on My Yahoo!, Google), or a standalone aggregator. We have added further corpora on /p/hlp/corpora/ and more is to come. An updated page reflecting our corpus and software inventory is in the works.

Corpora

1. Plain text corpora

1.1. Gigaword corpora

1.2. Reuters - NIST Corpus

Reuters newswire from 1996-08-20 to 1997-08-19.

To use this corpus you as an individual must sign a license agreement, which is then kept on file in the lab. After signing you will be added to the reuters Unix group on the HLP system. This corpus is not available on the CS system.

See http://trec.nist.gov/data/reuters/reuters.html for more detail.

2. Audio and video corpora

Audio corpora (the sound files of corpora) are stored under /p/hlp/corpora/Audio/.

2.1. Switchboard sound files

We've installed the sound files of those Switchboard dialogues (Switchboard 1 release 2) that are part of the Penn Treebank (release 3, Marcus et al. 1999) and the Edinburgh-Stanford Paraphrase Switchboard. Jason Brenier has developed Python scripts that map the Switchboard annotation layers to the sound files, making it possible (via intermediate steps) to e.g. conduct syntactic searches and then extract acoustic information from the sound files over the syntactic match.

2.2. Buckeye Corpus

2.3. Czech Broadcast Conversation

Speech and MDE Transcripts

2.4. CALLHOME American English

Transcripts

3. Syntactically annotated corpora

3.1. Treebanks

Title

File

LDC Catalog number/Original name

Language

#word

#sentence

#story

Original format

Arabic Treebank Part 1 V3

ATB1_V3/

LDC2005T02

Arabic

145,386

734

Arabic Treebank Part 2 V2

ATB2_V2/

LDC2004T02

Arabic

144,199

501

Arabic Treebank Part 3 V1

ATB3_V1/

LDC2004T11

Arabic

340,281

600

Chinese Treebank V5.1

ChineseTreebank5.1/

LDC2005T01U01

Chinese

507,222

18,782

Chinese Treebank V6.0

ChineseTreebank6.0/

LDC2007T36

Chinese

781,351

28,295

Chinese Treebank v7.0

ChineseTreebank7.0/

LDC2010T07

Chinese

840,000

Penn Discourse Treebank Version 2.0

pdtb_v2/

LDC2008T05

English

Prague Dependency Treebank 2.0

pdt_2/

LDC2006T01

Czech

2,000,000

Danish Dependency Treebank V1.0

ddt1.0/

ddt-1.0.tar

Danish

5540

NEGRA corpus V2.0

Negra2.0/

negra-corpus.tar.gz

German

20,602

export/Penn Treebank

Merge files

mrg/

Tübingen Treebank of Spoken English

TuebaES/

English

310,000

30,000

export/Penn Treebank/XML

Tübingen Treebank of Spoken German

TuebaDS/

German

360,000

38,000

export/Penn Treebank/XML

Tübingen Treebank of Spoken Japanese

TuebaJS/

Japanese

160,000

18,000

export/XML/CoNLL-X Shared Task dependency

3.1.1. Penn Chinese Treebank

Segmentation Guide
POS Tagging Guide
Bracketing Guide

3.2. TGrep2able

All of our syntactically annotated corpora have been processed to make them usable with the TGrep2 tool. See Corpus Tools for more info on TGrep2.

Filename

Description

BNC.parsed.t2c.gz

The full British National Corpus

BNC_spoken.parsed.t2c.gz

Just the spoken part of the BNC

BNC_written.parsed.t2c.gz

Just the written part of the BNC

arabic-collapsed.t2c.gz

arabic-treebank-with-vowels.t2c.gz

arabic-treebank-without-vowels.t2c.gz

brown.t2c.gz

chtb2.t2c.gz

chtb4.t2c.gz

chtb5.1.t2c

commented-brown.t2c.gz

icegb.t2c.gz

negra.t2c.gz

sw.backtrans.convid_030507.t2c.gz

sw.backtrans.t2c.gz

swbd.t2c.gz

tiger.t2c.gz

tuebaes.t2c.gz

Tübingen Treebank of Spoken English

tuebads.t2c.gz

Tübingen Treebank of Spoken German

wsj-commented.t2c.gz

wsj_mrg.t2c.gz

ycoe.t2c.gz

3.3. TIGER Corpora

3.3.1. Tiger2 Corpus

4. Edinburgh-Stanford Paraphrase Switchboard

This corpus combines numerous annotations for the Penn Treebank (release 3, Marcus et al. 1999) portion of the Switchboard corpus (Godfrey et al. 1992). In addition to the part-of-speech, grammatical function, and syntactic annotation of the Treebank, the corpus includes annotation for turn-taking, disfluencies (Taylor et al. 1995), dialogue acts (Shirberg et al. 1998), animacy (Zaenen et al. 2003), coreference and information status (Nissim et al. 2001), kontrast and kontrast-triggers (Calhoun 2006), as well as time-aligned orthographic transcripts (automatically aligned down to the syllable and segment level), some prosodic phrasing and accent annotation, and some more. For a detailed summary and manuals, see the Edinburgh webpages on Switchboard in NXT. We have both the XML version and a version that has been back-translated to Penn Treebank format, so that it is TGrep2able (thanks to Jean Carlette, Jason Brenier, Sasha Calhoun, and Neal Snider).

The time-alignment of the orthographic words that are the terminals of the Treebank Switchboard corpus make it possible to go from the syntactic searches to the corresponding sound files of the Switchboard conversations to extract acoustic information from them (or for listening pleasure ;-)).

NB: You must be a member of the pswbd group to have access to this corpus.

HlpLab: Corpora (last edited 2011-10-07 19:58:33 by AndrewWatts)

MoinMoin Appliance - Powered by TurnKey Linux