File size: 4,939 Bytes
1cbd990 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 |
P A T T R : P A T E N T T R A N S L A T I O N R E S O U R C E
Download link: http://www.cl.uni-heidelberg.de/statnlpgroup/pattr/
Author: Katharina Wäschle (waeschle@cl.uni-heidelberg.de)
Date: 28/02/2013
PatTR is a parallel corpus extracted from documents in the MAREC patent
collection [1]. The first release contains 23 million German-English
parallel sentences collected from all patent text sections.
0. TERMS OF USE
PatTR is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 3.0 Unported License (see LICENSE). Please cite
Wäschle & Riezler (2012b), if you use the corpus in your work.
1. FILES
abstract/
pattr.de-en.abstract.de
pattr.de-en.abstract.en
pattr.de-en.abstract.meta
claims/
pattr.de-en.claims.de
pattr.de-en.claims.en
pattr.de-en.claims.meta
description/
pattr.de-en.description.de
pattr.de-en.description.de.meta
pattr.de-en.description.en
pattr.de-en.description.en.meta
title/
pattr.de-en.title.de
pattr.de-en.title.en
pattr.de-en.title.meta
*.de files contain German sentences, *.en files corresponding English
sentences. *.meta contain information about the document the
sentences were extracted from as tab-separated values:
- document id
- patent family id
- publication data
- IPC up to subclass level, comma-separated
For the description data, where the bitext has been collected from two
separate documents, there is a metadata file for each of the source
documents (*.de.meta for the German document from the EPO corpus,
*.en.meta for the English document from the USPTO corpus).
2. DATA
The corpus is split into files according to the text sections of a
patent document: title, abstract, claims and description.
Parallel data from the title, abstract and claims sections were
extracted from documents belonging to the European Patent Office
(EPO) [2] and the World Intellectual Property Organization (WIPO) [3]
corpora in MAREC. Both resources feature multilingual documents that
contain for example both an English and a German abstract.
Since there are no multilingual descriptions, data from this section
were collected by exploiting patent families to align German documents
from the EPO corpus to English documents from the United States Patent
and Trademark Office (USPTO) [4] corpus, following Utiyama and Isahara
(2007).
All sections were sentence-aligned using the Gargantua aligner [5].
Preprocessing was done automatically. Sentence boundaries were detected
using the Europarl processing tools [6].
4. STATISTICS
Section Sentences EN tokens DE tokens
title 2,101,107 16,457,527 13,212,645
abstract 720,571 30,942,571 26,803,868
claims 8,346,863 501,373,533 435,117,827
description 11,829,816 498,948,414 386,920,744
total 22,998,357 1,047,722,045 862,055,084
5. TEST SETS
The training and test sets used in Wäschle & Riezler (2012a) can be
provided on request to waeschle@cl.uni-heidelberg.de. For creating
custom training and test sets, an easy option is to split the corpus by
document publication date. Note, that abstract and claims data contain
a small amount (less than 1%) of duplicate and near-duplicate sentences
due to multiple instances of the same patent document in the two
corpora. To prevent overlap, make sure family ids of test and training
set are disjunct. Furthermore, about 7% of the description data are
duplicates. This is caused by the patent writing process, where whole
paragraphs are copied verbatim from other documents, e.g. when parts of
an invention are similar to a previously filed one. These documents do
not share a patent id, so they cannot be easily identified. Indicators
are mutual citations and documents filed by the same company. We did
not remove these duplicates because they are a feature of patent
corpora. Since patent titles are very short and general, 15% of title
data are natural duplicates.
6. ACKNOWLEDGEMENTS
The work was in part supported by the "Cross-language Learning-to-Rank
for Patent Retrieval" project funded by the Deutsche
Forschungsgemeinschaft (DFG).
PUBLICATIONS
Wäschle, K. and Riezler, S. (2012a). Structural and Topical Dimensions
in Multi-Task Patent Translation. Proceedings of the 13th Conference of
the European Chapter of the Association for Computational Linguistics
(EACL 2012), Avignon, France.
http://www.aclweb.org/anthology-new/E/E12/E12-1083.pdf
Wäschle, K. and Riezler, S. (2012b). Analyzing Parallelism and Domain
Similarities in the MAREC Patent Corpus. Multidisciplinary Information
Retrieval, pp. 12-27.
http://www.cl.uni-heidelberg.de/~riezler/publications/papers/IRF2012.pdf
LINKS
1. http://www.ir-facility.org/prototypes/marec
2. http://www.epo.org
3. http://www.wipo.int
4. http://www.uspto.gov
5. http://sourceforge.net/projects/gargantua
6. http://www.statmt.org/europarl
|