DASH

News and Events

Aron Henriksson’s Licentiate Dissertation Defense

Health Informatics, Licentiate Seminar, Photo, SYSLAB

Aron Henriksson convincingly defended his Licentiate thesis on October 18, 2013. The thesis, with the title Semantic Spaces of Clinical Text – Leveraging Distributional Semantics for Natural Language Processing of Electronic Health Records, investigates the applicability and utility of distributional semantics to facilitate information extraction from electronic health records. This was demonstrated in several use-cases; finding synonyms, identifying and explaining abbreviations, detecting adverse reactions to medication, and assigning clinical codes to texts. The mined health records were written in Swedish and, in one instance, in English.

Professor Jussi Karlgren from KTH conducted the opposition. Given his extensive hands-on experience with the aforementioned techniques in other domains, both as a researcher at SICS and as a founder of the company Gavagai, Jussi led a very interesting discussion with the respondent. The opponent and the examiner had many good things to say about the thesis. The examiner from DSV was Professor Paul Johannesson. Aron’s advisors are Dr Martin Duneld and Professor Hercules Dalianis, who will continue to work with Aron in his future research studies.

Aron defending his thesis, the opponent Jussi Karlgren is leading the discussion.   The respondent, the opponent, the examiner and the two advisors.

Congratulations!

Two papers presented at Nodalida 2011

Health Informatics, Information Systems, Research Paper, SYSLAB

Two papers which I have co-written was presented at the The 18th Nordic Conference of Computational Linguistics, Riga, Latvia, May 11–13, 2011. The conference started out as focusing on NLP for the Nordic languages but now attracts participants from a score of countries. Attending this year were 130 people from mostly Europe.

The first paper, with the title Something Old, Something New – Applying a Pre-trained Parsing Model to Clinical Swedish, is co-written with Aron Henriksson and Sumithra Velupillai.

Abstract: “Information access from clinical text is a research area which has gained a large amount of interest in recent years. Automatic syntactic analysis for the creation of deeper language models is potentially very useful for such methods. However, syntactic parsers that are tailored to accommodate for the distinctive properties of clinical language are rare and costly to build. We present an initial study on the applicability of an existing parser, pre-trained on general Swedish, to clinical text in Swedish. We manually evaluate twelve documents and obtain a 92.4% part-of-speech tagging accuracy and a 76.6% labeled attachment score for the syntactic dependency parsing.”

The second paper, with the title The Impact of Part-of-Speech Filtering on Generation of a Swedish-Japanese Dictionary Using English as Pivot Language, is co-written with Ingemar Hjälmstad and Maria Skeppstedt. The paper is an adaptation of Ingemar’s Master thesis with the same title. Good work!

Abstract: “A common problem when combining two bilingual dictionaries to make a third, using one common language as a pivot language, is the emergence of false translations due to lexical ambiguity between words in the languages involved. This paper examines if the translation accuracy improves when using part-of-speech filtering of translation candidates. To examine this, two different Japanese-Swedish lexicons were created, one with part-of-speech filtering, one without. The results show 33% less translation candidates and a higher quality lexicon when using part-of-speech filtering. It also resulted in a free lexicon of Swedish translations to 40 716 Japanese entries with a 90% precision, and the conclusion that part-of-speech filtering is an easy way of improving the translation quality in this context.”

Second HEXAnord Meeting, Åre 12-14 April 2010

Health Informatics, Information Systems, Networking, SYSLAB

12-14 April the IT for Health group at DSV arranged the second HEXAnord network meeting in Åre, Sweden. Apart from Hercules Dalianis, Sumithra Velupillai, Maria Skeppstedt, Elin Carlsson, Helen Allvin and me from DSV, there were participants from all participating countries (Finland, Norway, Denmark, Estonia and Lithuania) as well as visitors from Karolinska Institutet/KTH and Östra Sjukhuset, Västra Götalandsregionen.

The meeting consisted of seminars mainly on Machine Learning and Clinical Text together with work on the coming the 2nd Louhi Workshop on Text and Data Mining of Health Documents in L.A., June 5 2010. In conjunction to this work no less than three papers from the group, that had passed the double blind peer-review, were finalized.

Nine working groups were created for Intra Nordic cooperation dealing with Automated GTT-Global Trigger Tool, Visual Analytics, Requirements engineering, Release of Data, Negation detection, Decision making and reasoning processes, Characteristics of clinical texts, Compare clinical and general language and finally Comorbidity, all of them in the perspective of health text analysis.

On the agenda was also planning of the 3rd Louhi Workshop, probably to be held next year, as well as further planning for the PhD course to be held later this year.

See more photos below:

Two papers accepted at RANLP 2009

Information Systems, Publication, Research Paper, SYSLAB

Today I recieved notification that two papers where I am one of the authors have been accepted as posters at the international conference RANLP 2009 – Recent Advances in Natural Language Processing, which is in Borovets, Bulgaria in September 14-16.

Apart from being an exceptionally pleasant conference, RANLP is ranked among the most influential NLP conferences by the site Computer Science Conference Ranking. According to this list it is one of the top 67 conferences among 701 considered in the fields of Artificial Intelligence, Machine Learning,Robotics and Human Computer Interaction.

The two accepted papers are:

Identification of Parallel Text Pairs Using Fingerprints by Martin Hassel and Hercules Dalianis.

Global Evaluation of Random Indexing through Swedish Word Clustering Compared to the People’s Dictionary of Synonyms by Magnus Rosell, Martin Hassel and Viggo Kann.

Dr. Magnus Rosell and Professor Viggo Kann both reside at the School of Computer Science and Communication, KTH – Royal Institute of Technology.

Project proposal submitted to VR: VESPTEC

Information Systems, Project Proposal, SYSLAB

Today I submitted a project proposal to the Swedish Research Council (Vetenskapsrådet) with the title “VESPTEC – Vector space representations of textual content”. Collaborating with me on this proposal are Magnus Rosell and Viggo Kann at KTH CSC as well as Jussi Karlgren at SICS and Hercules Dalianis here at DSV.

Abstract:
Since the 1960s vector space models have been used extensively for representation of semantics, especially in information-retrieval systems such as Google. These vector spaces are usually multi-dimensional and the terms and documents are represented by very large matrices. There is no greater regard to context. For instance, how a term occurs in a document is almost completely disregarded. Texts are thus viewed as mere bags-of-words. Much of the research so far has either focused on the application of these representations on specific tasks, or on the efficiency of this application by reducing the dimensionality of the original space in some way. This project proposes the study of vector space representations of textual content in a more systematic manner.

We have identified two main tasks. One is to explore the notion of intrinsic dimensionality and the spatial metaphor often used in describing “likeness” between documents. The other, and perhaps more intriguing task is that of moving from a bag-of-words representation to a more informed document space, modeling more than just the cooccurrence of lexical items within documents. These models will be systematically validated on a diverse array of text processing tasks and well established test sets with built-in success criteria. A better representation of textual content is interesting in itself, but will also lead to better underlying models that will improve applications, such as search engines and text summarization.

DSV research group joins Human Language Technology Center

Information Systems, Networking, SYSLAB

The Human Language Technology Group here at DSV has now officially joined the mid-Sweden Human Language Technology Center.

The objective of the center is to facilitate collaboration in order to promote excellence in research and increase the competitiveness and visibility of the participating research groups, building on their unique combination of competence in both text and speech technology.

Research on human language technology is concerned with systems that process human language, in written or spoken form. This includes fundamental research on modeling human language processing as well as research targeted at practical applications, such as machine translation and spoken dialogue systems. The Human Language Technology Center (HLTC) brings together leading research groups on both text and speech based at Uppsala University, Stockholm University and KTH.

Seminar on three Human Language Technology research projects here at DSV

Health Informatics, Information Systems, Presentation, SYSLAB

Hercules Dalianis, Sumithra Velupillai and Martin Hassel held a lunch seminar on three Human Language Technology research projects here at DSV.

Hercules presented results from the project Tvärsök which aims to bridge the language gap between the Nordic countries. More specifically, if you search for documents in Swedish you should be able to find corresponding information in Danish or Norwegian if such are lacking for Swedish. The idea is that while you might have active knowledge of Swedish, in the sense that you can formulate search queries in that language, you might still be able to read and comprehend closely related languages even if you cannot effectively formulate queries in those languages (passive knowledge). This work has resulted in new lexical resources in the mobility domain as well as new methods for identifying parallel text [Hercules’ slides]

Sumithra presented ongoing work in the KEA project which concerns the medical domain. In this project we have access to a very large set of medical records, on which we aim to apply an array of text mining techniques. However, these patient files need to be de-identified before they can be exposed to a broader spectrum of researchers. A major part so far has been the development and porting of de-identification methods to Swedish. Parallel tracks are the exposing of relations between loosely defined features hidden in the vast amount of clinical data, as well as the identification of (un)certainty in diagnoses [Sumithra’s slides]

Martin presented two related projects on language technology based services for eGoverment. The first one, IMAIL, which just recently has been approved, concerns the automation of responses to requests sent by e-mail to the Swedish Social Insurance Agency (Försäkringskassan). These requests will be classified as recognised, in which case a predefined response can automatically be sent to the customer, or unrecognised and forwarded to a government official. In the latter case a tool already employed in the KEA project – Infomat – can be adapted to present the official with a view over previous transactions related to incoming request. The second project, still awaiting approval, “Intelligent cross-language services for eGovernment” is a quite ambitious extension to IMAIL which aims at further dealing with translation, multilingual authoring and cross-lingual information extraction for the Swedish minority languages [Martin’s slides]

Project proposal: Vector space representations of textual content

Information Systems, Project Proposal, SYSLAB

Today I handed in a project proposal to the Swedish Research Council (Vetenskapsrådet). Collaborating with me on this proposal are Magnus Rosell and Viggo Kann at KTH CSC as well as Magnus Sahlgren, Jussi Karlgren and Oscar Täckström at SICS. The title of the project is Vector space representations of textual content.

Abstract: Since the 1960s vector space models have been used extensively for representation of semantics, especially in information-retrieval systems such as Google. These vector spaces are usually multi-dimensional and the terms and documents are represented by very large matrices. There is no greater regard to context. For instance, how a term occurs in other documents is almost completely disregarded. Texts are thus viewed as mere bags-of-words.

Much of the research so far has either focused on the application of these representations on specific tasks, or on the efficiency of this application by, for example, reducing the dimensionality of the original space in some way. This project proposes the study of vector space representations of textual content in a more systematic manner.

We have identified two main tasks. One is to explore the notion of intrinsic dimensionality and the spatial metaphor often used in describing “likeness” between terms and documents. The other task is that of moving from a bag-of-words representation to a more informed document space, modeling more than just the cooccurrence of lexical items within documents. These models will be systematically validated on a diverse array of tasks and well established test sets with built-in success criteria.

A better representation of textual content is interesting in itself, but will also lead to better underlying models that will improve useful applications, such as search engines and text summarization.