Main Conference

WSPP — 2009-12-17T15:59:52Z

The Workshop on Language Technology for Patent Data: Language Resources and Evaluation is held in conjunction with the LREC 2012 conference organised by ELRA. - Main Conference / ouvert_rubrique

Call for Papers

WSPP — 2009-12-17T15:50:32Z

In the last few years, the use of patents in automatic processing has shown a growing interest in the NLP community. This has been particularly the case in the context of Machine Translation (MT) or Cross-Lingual Information Retrieval (CLIR). Nowadays this has become a major topic and besides the development of the technology itself, some key points remain regarding the resources available and the way of evaluating the quality of the technology.

A large number of language resources is already available for the community, but the development of systems, in particular the statistical ones, always requires more and more data. As there is a growing interest for patents and their processing, a workshop on the topic which gathers all those involved in the different aspects concerned is a good opportunity to move forward.

The domain of patents itself is increasing and the amount of potential material does not cease to increase. It is this potential material that gives hope to the community for improving the systems. For instance, in China, the number of patents have been multiplied by 3 in 5 years and they exceed 1 million published documents per year by now. EPO (the European Patent Office) uses more than 150 translation pairs per day. Every patent office receives more and more patents every day, needs a daily use of automatic tools to translate the documents, looks for existing patents and their translation, manages complex content, etc. As we can see, this is a domain in considerable demand and since the content of the patents is technical and needs high skills in a specific domain, providing documents that are sufficiently understandable to the end users is very complex. This is a real challenge for all NLP developers.

Above all, this challenge is about corpora and their management. The main topic concerns their acquisition and how to collect useful data. For most of the researchers, this consists in harvesting web pages, cleaning them, getting the useful content according to a specific task, aligning the sentences, etc. The acquisition task may also be done using OCR tools on PDF. Monolingual corpora are easier to retrieve (e.g. from databases) compared to parallel corpora. However, parallel translations exist and aligned corpora as well, or corpora that could be easily aligned. Following the question of the acquisition of such documents, there is that of database management. One could say that all these questions are not only related to patent data, however this workshop would like focus on this particular domain and make some effort to improve things.

Currently, the corpora are mainly used for MT. For a technical end-user in a patent office, the end goal is to manage to understand the content of a document. This may not require a very high quality translation since this person only needs to grasp the relevance of the document. However, in MT, we still need to measure quantitatively the performance of the systems. This is basically made using automatic and/or human measures, while most of the system developers are using typical automatic metrics such as BLEU to get their results. Even if the drawbacks of such metrics are well-known, it could be still relevant, for instance, to compare different versions of a system. However, even when using BLEU, the content of patent documents is very particular, which implies that different kinds of linguistic specificity need to be tackled: these include the already expected terminological level, but also a syntactic level, a semantic one, and even the structure of the documents may be different from that of other documents (for instance, patents typically comprise of a title, an abstract, a technical description of the invention, and a list of novel claims). Human measures may be also difficult to apply as patent documents are written in a way which makes them difficult to read for the layman. Furthermore, both automatic and human evaluations should have the chance to realise a deep analysis of the results, which is not trivial working with patents. However, given the often formulaic nature of the text found in patents – which is enforced on the author due to legal constraints – there may be opportunities to exploit this for evaluation. For instance, claims are constructed as a single sentence with an introductory phrase and a body linked by frequently occurring terms such as “in a certain embodiment”, “consisting essentially of”, and clauses and lists introduced using colons, e.g. “comprising: …”

The use of patents in CLIR suffers from the same kind of issues, either for the evaluation of systems or for the collection of corpora. Sentence alignment may also have specific issues related to the content of the documents, and many other types of tools may have their own thoughts using patents. Through all those technologies, one can see their usage implies several challenges, such as the integration of tools into patent information applications. The different tools should help end-users to search, examine or classify patent documents, most of the time from translations and not available in English. Web services should also be an extension of the tools and web services should be connected through workflows, helping end-users in their daily work.

Among all the topics previously mentioned, we would like to contribute to the improvement of the challenging patent field, by sharing the knowledge from the whole community.

The different topics addressed during the workshop will be (but are not limited to):

Corpora aspects: collecting data, cleaning, alignment, parallel corpora, etc.;
Evaluation of technologies: definition of metrics, patent specificity;
Integration of patent applications: web services, end-user applications;
IPR issues and licensing.

Submissions

WSPP — 2009-12-17T15:49:21Z

Full papers up to 8 pages should be formatted according to LREC 2010 guidelines and be submitted through the online submission form on START.

The templates for paper are published on the LREC web site

When submitting a paper through the START page, authors will be kindly asked to provide relevant information about the resources that have been used for the work described in their paper or that are the outcome of their research. For further information on this new initiative, please refer to http://www.lrec-conf.org/lrec2012/?LRE-Map-2012

Important Dates

WSPP — 2009-12-17T15:25:59Z

Extended deadline for submission: Friday 2 March 2012 Notification of acceptance: Friday 23 March 2012 Final version due: Friday 30 March 2012 Workshop : 27 May 2012 (afternoon) - Important Dates / ouvert_rubrique

Committees

WSPP — 2009-12-17T15:24:07Z

Table of contents

Organizing Committee, p1
Programme Committee, p2

1 2

Organizing Committee

Olivier Hamon (ELDA – Evaluations and Language resources Distribution Agency, France)
John Tinsley (PLUTO - Patent Language Translations Online, Ireland)
Heidi Depraetere (Crosslang, Belgium)

1 2

Contact

WSPP — 2009-12-17T14:33:35Z

For further queries, please contact Olivier Hamon at hamon_at_elda_dot_org. - Contact / ouvert_rubrique

LTPD 2012 Workshop