Nordic Treebank Network

Participants

• Växjö University, School of Mathematics and Systems Engineering. Contact: Joakim Nivre (network coordinator).
• Copenhagen Business School, Department of Computational Linguistics. Contact: Matthias Trautner Kromann.
• CSC Scientific Computing Ltd, The Language Bank of Finland. Contact: Manne Miettinen
• Göteborg University, Department of Linguistics Contact: Jens Allwood
• Göteborg University, Department of Swedish, Natural Language Processing Section Contact: Lars Borin.
• KTH, Department of Speech, Music and Hearing. Contact: Beata Megyesi.
• NTNU, Department of Linguistics. Contact: Torbjørn Nordgård.
• Stockholm University, Department of Linguistics. Contact: Martin Volk.
• University of Bergen, Department of Linguistics. Contact: Koenraad de Smedt.
• University of Helsinki, Department of General Linguistics. Contact: Kimmo Koskenniemi.
• University of Iceland. Contact: Eirikur Rögnvaldsson.
• University of Oslo, The Text Laboratory. Contact: Janne Bondi Johannessen.
• University of Southern Denmark, Institute of Language and Communication, VISL project. Contact: Eckhard Bick.
• University of Tartu, Institute of Computer Science. Contact: Heli Uibo.
• Uppsala University, Department of Linguistics. Contact: Eva Ejerhed.

Network Activities

• Network meetings:
  1. First general meeting: 17-18 September 2003, Fefor, Norway: Minutes.
  2. Internal workshop: 5-7 March 2004, Stockholm, Sweden: Minutes.
  3. Second general meeting: 8-10 September 2004, Tartu, Estonia: Minutes.
• International workshops:
  1. The Second Workshop on Treebanks and Linguistic Theories, 14-15 November 2003, Växjö.
  2. The Third Workshop on Treebanks and Linguistic Theories, 10-11 December 2004, Tübingen.
• PhD Courses and summer schools:
  1. Treebanks: Formats, Tools and Usage, 1-5 March 2004, Stockholm University.
  2. Empirical Methods in Natural Language Processing, 9-14 August 2004, Tartu, Estonia.
• Nordic parallel treebank (password protected for copyright reasons).
• Documentation of projects, tools and resources:
  1. Treebank representation formats
  2. Treebank tools (creation, usage, evaluation)
  3. Projects and resources at network sites

Projects and Resources within the Network

Copenhagen Business School

• Danish Dependency Treebank (DDT). The DDT currently contains 100.000 words of syntactically annotated text from the Danish PAROLE corpus. The underlying linguistic principles are documented in the tagging manual for DDT.
• DTAG dependency annotation tool. DTAG is a linguistic command language interpreter for editing, searching, comparing and displaying dependency-annotated texts within a treebank.
• Discontinuous Grammar. The purpose of this project is to create a linguistic theory for discontinuous (non-projective) dependency grammars with a cost component, and to design heuristic algorithms for (1) serial parsing with repair, and (2) learning a massively ambiguous dependency lexicon from a treebank.

CSC Scientific Computing

• Finnish text collection. The Language Bank of Finland provides access to roughly 180 million words of Finnish text, mainly newspaper articles. All the text are encoded in XML. 60 % of the texts are automatically morphosyntactically analyzed with Kielikone's Textmorfo analyzer.
• Finland-Swedish text collection. The Language Bank of Finland provides access to roughly 32 million words of Finland-Swedish (Swedish text produced by Swedish speaking Finns) text, mainly newspaper articles. All the text are encoded in XML. All the texts are automatically morphosyntactically analyzed with Lingsoft's SWECG analyzer.
• Lemmie is a corpus query tool with a web user interface. It is based on a object oriented Perl API and allows the user to query texts in the Finnish and Finland-Swedish text collections.

Göteborg University - Linguistics

• Spoken Language and Semantics This research group has collected several corpora, mainly of spoken language. The main corpus is called GSLC (Göteborg Spoken Language Corpus), and it is an incrementally growing corpus of spoken language from different social activities.
• NordTalk The NorFA network on corpus based research on spoken language: Cooperation between general linguistics, computational linguistics, language technology and speech technology in the Nordic countries

Göteborg University - Swedish

• Språkbanken -- the Bank of Swedish at Göteborg University is a constantly growing and evolving national repository of annotated corpora and lexical resources for Swedish, including some syntactically annotated material and also parallel corpora. Many of Språkbanken's resources are searchable with a web interface.

KTH

• Data-driven Morpho-Syntactic and Shallow Syntactic Analysis of Swedish The purpose of this project is to automatically model the morpho-syntactic and constituent structure of Swedish by using various data-driven learning algorithms and combine these by ensemble techniques.
• Swedish Treebank The goal of this project is to build a large-scale treebank for Swedish, containing both spoken and written language data, and annotated with both constituent structure and dependency structures. The project is a cooperation between six Swedish universities.
• The GROG project: Boundaries and Groupings - The Structuring of Speech in Different Communicative Situations The purpose of this project is to model the structuring of speech in terms of prosodic boundaries and groupings in various communicative situations. The modeling aims at a structured and optimized description of the relations between the acoustic and linguistic structure on the one hand, and the prosodic/perceptual annotations on the other.
• Learning Lexical Semantic Relations for Swedish The aim of this project is to automatically learn information about lexical semantic relations for Swedish data. The aim is also to introduce and evaluate the use of this lexical knowledge in communicative systems.
• Open source softwares Open source tools developed at CTT, TMH, KTH.
• Speecon: Speech-driven Interfaces for Consumer Devices SPEECON is a project focusing on collecting linguistic data for speech recogniser training. SPEECON is funded as a shared-cost project under Human Language Technologies (HLT), which is part of the European Commission's Information Societies Technologies (IST) Programme. The partners will collect speech data for 18 languages or dialectal zones, including most of the languages spoken in the EU. TMH, KTH is a subcontractor and prepared the recordings of the Swedish data. See also the Speechdat project which consists of a series of speech data collection projects funded by the European Union. http://www.speechdat.org/

NTNU

• Projects

Stockholm University

• Converting SynTag to XML for TIGER-Search SynTag is a Swedish treebank dating back to 1986. Its annotation comprises part-of-speech tags and predicate-argument structures. In this project we convert the predicate-argument structure into a tree structure that follows the TIGER-XML coding scheme. This will allow the SynTag treebank to be loaded into TIGER-Search, a powerful treebank search tool.
• Adapting the TIGER annotation guidelines to Swedish This projects aims at investigating to what degree the annotation format and guidelines developed in the TIGER (and NEGRA) project for the German treebank will be applicable to Swedish. The project will result in 100 consistently annotated Swedish sentences in the spirit of TIGER/NEGRA and a detailed description of the advantages and disadvantages of using the TIGER/NEGRA guidelines for Swedish (with suggestions for alternative annotations for specific cases).

University of Bergen - Section for linguistic studies and Language technology group at AKSIS

• TREPIL investigates methods for building a Norwegian treebank semi-automatically. The project focuses on rich syntactic and semantic annotation. The goal is not to build a full scale treebank, but to acquire the necessary know-how, including linguistic principles, database design, resources, tools, evaluation, etc. for building a treebank with multiple structures for each sentence. The approach is tested on a small scale treebank. The project is coordinated by Prof. Koenraad de Smedt and is administered by AKSIS. Financed by the Norwegian Research Council under the KUNSTI program, the project runs from April 2004 until March2007 and cooperates actively with other research groups, in particular the language technology group at PARC (Palo Alto), the TIGER-project (Stuttgart), and the Nordic treebank network.

University of Helsinki

• The University of Helsinki Language Corpus Server (UHLCS) Computer corpora of more than 50 languages, including samples of minority languages and extensive corpora representing different text types. In 2000, the corpora of the Uralic, Turkic, Tungusic, Mongolic, Palaeo-Siberian, Iranian and Caucasian languages.

University of Oslo

• PaNoLa - Parsing Nordic Languages The project, coordinated by the University of Southern Denmark, Odense, aims at enhancing the Nordic element within the VISL system. Interactive syntax learning and grammar computer games will be available for Norwegian, Swedish and Danish. C
• Nomen Nescio-prosjektet - en automatisk navnegjenkjenner for norsk, svensk og dansk The project, coordinated in Oslo, aims at developing an automatic named entity recogniser for Norwegian, Swedish and Danish.
• Oslo-Bergen-taggeren The Oslo-Bergen tagger is based on the Constraint Grammar formalism, and is used to tag the Oslo Corpus of Tagged, Norwegian Texts.
• Big Brother-korpuset This corpus is under development at the University of Oslo. It is based on the first Norwegian series of the reality show Big Brother, and will be a spoken language corpus with transcribed text, sound and video.
• GREI The aim of the GREI project has been to extend the number of Norwegian sentences on the VISL site in University of Southern Denmark, and improve the user giudelines for Norwegian pupils, everything done in cooperation with Norwegian teachers.
• LOGON The LOGON is a large project, coordinated by the University of Oslo, aiming at automatic translation from Norwegian to English. Part of the project will be a parallel tourist corpus, of which a prototype can be viewed here.
• LDB corpus The LDB corpus is coordinated by the Department of Lexicography at the University of Oslo, as a resource for lexicographic research. Click for a search example.
• Usenet-korpus for norsk
This 140 million word corpus of Norwegian, developed at the University of Oslo, is from the newslist Usenet domain.
• Opus - Open Source Parallel Corpus The OPUS contains translated texts from the web, converted and aligned, and tagged, at the moment: 30,000,000 words in 60 languages. Developed at the Universities of Uppsala and Oslo.
• The Oslo Corpus of Tagged Norwegian Texts . Its user interface is web-based. A wide variety of search options are available for Bokmål and Nynorsk in this 20 mill. word corpus.
• A statistical tagger for Norwegian This tagger was developed by the University of Oslo for Inxight, Grenoble, but since then, a number of other types of statistical taggers have also been tested out.
• A corpus of spoken Norwegian language This corpus is only at its planning stages at this moment, its realisation depends on funding. Planned jointly by the Universities of Oslo and Bergen.

University of Southern Denmark

• Arboretum (Danish treebank) The Arboretum project, launched in 2002, aims at building a large, multi-format Danish treebank from automatically parsed and manually revised running text data (10 million words). Analyses are based on a Constraint Grammar parser (DanGram) with various lexical resources, and further processed with PSG rules and dependency transformers. About 100.000 words have been manually revised in both formats.
• Korpus 90/2000 Search interface for the treebank's corpus data base, in both text and CG-annotated format. Korpus90/2000 was compiled by DSL and annotated by Eckhard Bick (VISL) in 2002.
• Arboretum search interface for both text, form/function and structural searches. With a different interface, the same data can be accessed as part of VISL's internet-based grammar teaching initiative.
• DanGram parser On-line access to a live running text CG tagger and syntactic tree parser, with graphical tree visualisation. The info-section explains the tags and categories used in the Arboretum tree bank and other VISL corpus projects for Danish.

University of Tartu

• Text corpora (written Estonian)
  1) A corpus for different periods (from 1890s to 1990s) marked up to the sentence level, total 3 million words
  2) Morphologically and syntactically annotated and disambiguated corpora (300 000 and 100 000 words, respectively)
  3) Estonian news agencies current text corpus (about 23 million words).
• Corpus of Spoken Estonian The corpus is transcribed by the transcription of conversational analysis (CA). The goal is to collect various types of oral speech, the usage of both everyday and institutional conversation, spontaneous and planned speech, monologues and dialogues, face-to-face interaction and media texts. 659 texts (403,000 text units).
• The database of fixed expressions Currently contains 20,794 records, the majority of expressions (17,493) are phrasal verbs.
• Morphological analyzer A morphological analyser and generator for practical applications (hyphenator, spelling checker) have been developed.It is based on the Concise Morphological Dictionary of Estonian (46 000 lexical units). Recall 99%. Morphological tools are usable over WWW.
• Constraint Grammar shallow syntactic parser of Estonian The constraint grammar of Estonian includes 1240 constraints for morphological disambiguation, 180 syntactic mapping rules and 1118 syntactic constraints. Recall 83-90 %, less than 3,5 % errors. Several experiments for practical applications have been caried out on the basis of the parser: noun phrase extractor, automatic summarizer, grammar checker.
• Annotation tool A user interface for syntactic hand-annotation and disambiguation. The goal of the tool is to avoid typos usually made by hand-annotators. The user can select the correct tag by a mouse-click.
• The instruction for syntactic annotationThe instruction includes the description of syntactic tags used by CG syntactic parser of Estonian, with examples.
• Syntactically analyzed and disambiguated text corpus The corpus is intended as a test corpus for the Constraint Grammar syntactic parser of Estonian. The goal is to correctly annotate 200 000 words of written texts (mostly fiction + some newspapers).

Uppsala University

• Projects
  • KOMA: Corpus-based machine translation
  • MATS: Methodology and application of a translation system
  • Scaniakorpusprojektet: Multilingual document management
  • ETAP: Organisation and usage of parallel corpora
  • FASTY: Faster typing for disabled persons
  • Scarrie: Computer support for Scandinavian proof-reading
  • MIA: Multra at work (machine translation)
  • PLUG: Parallel corpus project at LiU, UU and GU
• Corpora
• Tools

Växjö University

• Swedish Treebank The goal of this project is to build a large-scale treebank for Swedish, containing both spoken and written language data, and annotated with both constituent structure and dependency structures. The project is a cooperation between six Swedish universities.
• Stochastic Dependency Grammars for Natural Language Parsing The purpose of this project is to study stochastic dependency grammars from three different perspectives: the formal perspective, the machine learning perspective, and the language technology perspective. Treebanks play a central role in the latter two areas.