Cognitive Systems

Biological Evolution and the Diversification of Languages

2011-11-24T18:18:00.010+02:00

BEDLAN (Biological Evolution and the Diversification of Languages) is a project funded by Kone Foundation. The background of the project and its main objectives were presented by its director, Prof. Urho Määttä from University of Tampere. The project is a collaboration between universities of Tampere, Turku and Helsinki and The Research Institute for the Languages of Finland. The presentation took place in a meeting organized by the Society for the Study of Finnish.

BEDLAN conducts research in two main areas:

development of dialects ("microevolution")
development of languages ("macroevalution")

The project is interdisciplinary including researchers from linguistics, biology and philosophy. Methodology applied in the BEDLAN project includes methods used for modeling complex dynamical systems (stemming from population genetics) and methods of historical linguistics.

Kaj Syrjänen was unable to attend the meeting but detailed description of the research results was given by his collaborators Jyri Lehtonen and Terhi Honkonen. Lehtinen introduced a Uralic vocabulary data collection used in the project. The data includes 17 languages with information on connections between lexical items in these languages (e.g. Finnish, Sami, Estonian, Komi, Udmurt, Hungarian, Mordvin, Mansi, Khanty, Livonian, Tundra Nenets, Karelian and Veps). Etymological dictionaries were used to analyze the historical connection. The number of words was 226. Examples of words include "meet", "moon" and "mother". These are in Finnish "liha", "kuu" and "äiti", in Karelian "liha", "kuu" and "emä", and in Veps "liha", "ku" and "mam".

Terhi Honkonen gave a presentation on the computational analysis of the data. In the introduction of the methodology, she referred to McMahon and McMahon (2005): "Language Classification by Numbers" (Oxford), and Atkinson and Gray (2006): "Curious parallels and curious connections - Phylogenetic thinking in biology and historical linguistics" (Systematic Biology, 54:513-526). The method used was Bayesian phylogenetic analysis (using a program called MrBayes).

A strong merit of this kind of research is that conclusions on the relationships between languages and dialects are made based on vocabulary patterns rather than on individual word instances.

Jyri Lehtonen continued by presenting research on using network analysis methods. First he discussed the differences between tree-based models and network models (e.g. Heggarty et al. 2010) and showed results of network analysis on Uralic languages. The network analysis divided the languages into groups of Baltic-Finnic, Saami, Samoyedic, Ugric and Permic languages. Meadow Mari and Mordvin were not clearly connected with any of these groups. Lehtonen mentioned the classical lexico-statistical research by Swadesh in 1950s and continued by presenting research on the effect of using more or less central vocabulary. Usually central vocabulary is used where the lexical items are typically stable and morphologically simple. In the Loanword Typology Project, 1400 meanings are considered. This has further lead to Leipzig-Jakarta list which includes 100 "most central" meanings. Lehtonen argued that less central vocabulary may help in detecting language connections in a fine-grained manner.

Lehtonen's presentation inspired to think about the future of scientific representation and the role of animations in it. In this case, it seems that an animation of the development of the network structure could be useful.

In the end of the meeting, Terhi Honkonen presented research results on analyzing the timing of divergence of languages. The method used for the analysis is BEAST (Bayesian Evolutionary Analysis Sampling Trees), originally developed for Bayesian MCMC analysis of molecular sequences.

WSOM 2012 in Chile

2011-11-23T13:54:00.010+02:00

WSOM 2012, 9th Workshop on Self-Organizing Maps will be organized in Viña del Mar, Chile between 12th and 14th of December 2012. A call for papers and more information on the conference is available at the WSOM 2012 web page. Prof. Pablo Estévez from Universidad de Chile serves as the general chair, Prof. Pablo Zegers from Universidad de los Andes, Chile as the program chair and Prof. Guilherme Barreto from Universidade Federal do Crera, Brazil as the publicity chair.

Viña del Mar is a city on central Chile's Pacific coast. It is located 5 kilometers from Valparaiso. The average high temperature in Viña del Mar in December is 23 degrees Celcius.

Timo Honkela, Pablo Estévez and Teuvo Kohonen during WSOM 2011 in Espoo, Finland

Heikki Hyötyniemi: Theory of Emergent Information

2011-11-01T18:41:00.005+02:00

In a meeting of the Finnish Society for Natural Philosophy, emeritus professor Heikki Hyötyniemi is giving a talk on his theory of emergent information. Hyötyniemi is an original thinker who considers philosophical issues in the framework of mathematical formulation. In this work, he studies complex phenomena in a systems theoretical framework. Hyötyniemi has coined the term neocybernetics that seems to be closely related to the second-order cybernetics of von Foerster, Pask, Maturana and others.

Hyötyniemi's presentation covered a wide range of topics ranging from multivariate statistics to control-theoretical and systems-theoretical considerations and further to metaphysical issues. The presentation was concluded by discussing implications on how to understand life and ethics. It seems important that ethics is considered from the systems-theoretical point of view (see Von Foerster's views on ethics).

An excerpt of one of Hyötyniemi's slides illuminates the contents of the presentation:

One needs no conscious adaptation to take place, it is an illusion when looking the group from outside, after "survival"
Emformation makes difference, it becomes visible; the fittest entity has evolutionary advantage most efficiently
One can interpret that there is hunger for life (livelihood)
It seems that notorious vitalism might be coming back
It is possible to approach questions of the type WHY.
Analyses starting from below, through emergence ("birth") in a distributed manner without central control ("subject")
Yes, this is not science, this is natural philosophy!

In the discussion, Hyötyniemi pointed out that even though process philosophy has some related insights, it is usually limited by relying on writings in natural language and does not use mathematical tools in theory formation.

Teemu Ruokolainen: Conditional Random Fields

2011-10-07T10:18:00.002+03:00

In a Computational Cognitive Systems group meeting, Teemu Ruokolainen gave a presentation on how Conditional Random Fields (CRFs) can be used to conduct labeling of sequential data and, more specifically, part-of-speech tagging. A CRF is a a discriminative undirected probabilistic graphical model for encoding relationships between observations and construct consistent interpretations. The presentation raised discussions on the relationship between CRFs, Hidden Markov Models and Weighted Finite State Machines. Ruokolainen conducts his research with the support from Langnet, Finnish Graduate School in Language Studies.

Bernardo Huberman: Social Media and Attention

2011-10-06T14:16:00.013+03:00

Bernardo Huberman from Hewlett-Packard Laboratories and Stanford University is visiting Aalto University School of Science. He is giving a joint ICS Forum/Aalto Physics Colloquium talk today, on 6st of October.

In his talk, Prof. Huberman discussed the explosion of online interactions: the web has speeded up the metabolism of thought and decision about work, social issues and play. Twitter is delivering over 200 million messages a day. Facebook has more than 800 million users and Google+ has gained more than 40 million users in one month. Rate of sharing content in social media is growing exponentially.

Huberman further pointed out that attention is limited by brain capacity. Attention is a scarce and valuable resource. Wherever attention flows, issues surface and ideas are discussed - and money often follows. Almost anything else except attention can be manufactured as a commodity. Actually, attention is the coordination mechanism that powers the progress of science. Within academia attention has a symmetry property: academics seek attention from those who also seek it from them. Topics that get a lot of attention become part of the research agenda of the community. Attention is used to determined professional standing. He also mentioned that productivity can be measured the same way. This raises, however, the question of time delays: the importance of a finding or an innovation may be understood only much later. Regardless of this, the value of the work is measured according to more or less recent citations.

Most of the attention that we pay today is reflected and propagated in social media. Huberman referred to a study of 15 million recommendations from amazon.com by Lefkovec, Adamic and Huberman (ACM Transactions on the Web, 2007). The network allocates attention in a highly nonlinear but predictable way. For example, the analysis of the behavior of 1 million users of digg.com follows lognormal distribution as predicted. In a publication by Wu and Huberman (2007), they noticed that novelty decays in a predictable manner. HP has built on these kinds of research results and developed a technology called i-catcher that dynamically charts the attention paid to each piece of online content on a web page.

Huberman continued into a very interesting topic of predicting the future with social media. He used the popularity of movies as an example and referred to a service called Hollywood Stock Exchange. He explained the use of sentiment analysis and the Mechanical Turk. Huberman defined subjectivity in this context as the ratio of positive-versus-negative tweets to neutral. Polarity is then the ratio of positive to negative tweets. In a publication by Asur and Huberman (2010) in the Proceedings of the ACM Conference on Web Intelligence, the authors present the result of some predictions.

One of the interesting aspects of social media, is that it shows what is general social agenda. Earlier, the mass media decided what are the topics of discussion in the society. In the attention economy, what "balloons" to the top defines the public agenda. Trends are formed as the popularity in the crowdsourced world. Trends are created by people keep tweeting. There is a certain persistence of trends. These issues are discussed by Asur, Huberman, Szabo and Wang (2011).

Huberman concluded by making a remark that with social media it has become possible to measure social phenomena, not only to create theories of them.

A lively discussion followed the presentation. Huberman noted that advertising a going through a revolution. As an example he stated that HP stopped advertising in traditional media. Propagation in a viral form is important. One question posed the issue of personalization which may prevent from receiving dissonant views and could start to resemble editorial control. In his answer, Huberman said for example that he is very careful in the use of Facebook and Google+. He also mentioned Lady Gaga and Paris Hilton as people who are experts in drawing attention by creating situations and scandals in an effective manner.

The talk by Prof. Huberman was very well attended by about one hundred researchers including many prominent professors at Aalto University including Erkki Oja, Kimmo Kaski, Risto Nieminen, Olli Simula, Pekka Orponen, Samuel Kaski and Kaisa Nyberg (list not conclusive).

META-FORUM 2011

2011-06-28T15:37:00.012+03:00

The META-FORUM 2011 is held June 27/28 in Budapest, Hungary. It is organized by META-NET, a Network of Excellence consisting of 47 research centers from 31 countries, which is dedicated to building the technological foundations of a multilingual European information society. The forum consists of keynote talks, several sessions, two panels and joined exhibition by industry and academia.

Thomas Hofmann from Google Research gave a keynote lecture on how they are bringing semantics to the web. He explained how interesting models can be scaled to web size by using curated knowledge bases (eg Freebase) and the output from accurate but slower parsers as the either the starting point to statistical analysis or training data for faster systems.

Bran Boguraev from IBM's TJ Watson Research Center talked about evidence based reasoning over natural language content. The example was of course their massively parallel Watson architecture for large-scale hypothesis generation, validation and scoring which tackles the long-standing challenge in AI to emulate human expertise.

Hans Uszkoreit introduced the outline for META-NET's strategic research agenda for language technology in Europe. The vision paper presents possible futures for innovative language technology applications in the following areas:

language-transparent web and media experience,
natural and inclusive interaction, and
efficient information management.

WSOM 2011 started

2011-06-13T15:42:00.002+03:00

The 8th Workshop on Self-Organizing WSOM 2011 started on Monday, 13th of June, at Aalto University School of Science. The program is available at www.cis.hut.fi/wsom2011/. A WSOM 2011 blog is also live.

In the event, there are participants, e.g., from Japan, Brazil, Chile, USA, France, Germany and Austria.

Content on the Multilingual Web

2011-04-20T22:23:00.034+03:00

The Internet world stats site is a fascinating source of information for predicting the trends of web usage in terms of language and locality. These statistics were frequently cited in Pisa, where the participants of the W3C coordinated MultilingualWeb project held its second workshop under the title multilingual content in the web.

To fully appreciate the demand for new developments in multilingual web technologies one only needs to look at the statistics of Internet penetration rates by geographic regions (see figure). In 2010, already 42% of Internet users are from Asia while the Internet penetration rate has only reached 21.5%. Considering that the basic building blocks of web technology (browsers, content markup syntax) were originally developed for English, it is obvious that there is much work to be done in developing web standards that enable language interoperability from the nuts-and-bolts level of character encoding all the way to statistical machine translation.

While the MultilingualWeb project in itself is a small one, the European Commission has substantial investments in developing multilingual technologies in the near future. In specific, the project officer Kimmo Rossi counted that in the year 2015 there will be a total of 50 ongoing projects in the area of multilingual technologies. Recently, the EC conducted a survey with Eurobarometer to study how Europeans use languages online. One of the findings was that 44% of people feel that they miss important information online because they don't understand the content language. Stay tuned for the full report of the survey on Eurobarometer's site.

Expertise << popularity ?

Standardization regarding multilingual web content often boils down to a compromise between sound engineering practices and the popularity of existing ad hoc solutions. The developers of elegant standards (such as XLIFF 2.0 as presented by David Filip) are often racing against the clock to finish their work before de facto standards (however good or bad) take over.

"Web standardization is a popularity contest. Facebook can get any non-standard garbage into browsers just by being popular." (anonymous browser developer in Pisa)

The increasing adoption of HTML5 is a promising development for the support for multilinguality. Some of the recent changes in HTML5 include the removal of the charset attribute in a and link elements and updates to ruby annotation syntax. Richard Ishida of the W3C invites everyone to participate in the development of the standard.

Charles McCathieNevile (presenting on behalf of Marcos Caceres) gave a tutorial on how the W3C Widget specification can be used to package a web application that can operate across device platforms. Simply by writing a small configuration file and adding the content into a zip archive I was able to create my own web widget in a matter of minutes. (Since when has implementing standards been this much fun?)

Crowdsourcing and love brands

One of the recurring topics of the workshop was the use of crowdsourcing for translation tasks on the web. Crowdsourcing (i.e. the process of distributing a task to an online community) was presented as a solution to commercial translation services as well as voluntary translation efforts. Pål Nes (of Opera Software) discussed some of the challenges they have met in crowdsourcing their translation efforts and highlighted that crowdsourcing has its costs and is not suitable for time critical tasks.

Chiara Pacella (of Facebook) presented their success in using crowdsourcing in the translation of Facebook's user interface. The overwhelming popularity of the site is reflected in an active translation community (the French translation of Facebook's interface was effectively finished within hours). So could anyone follow Facebook's example in letting the community do their localization work (and for free, did I mention). Could Microsoft use the community to translate MS Office?

"People translate Facebook for free because they love Facebook. People use Facebook because they want to. People use MS Office because they need to." Eliot Nedas (of XTM International)

There's certainly a dose of wisdom in these comments. It's quite possible that the most effective way to disseminate new technologies and standards is to make them easy and fun to work with.

HIIT Seminar presentation by Patrik Hoyer: Recommender systems for science

2011-02-21T13:20:00.013+02:00

In collaborative filtering, the correlations among ratings explicitly or implicitly provided by users are used to provide recommendations. Systems of this kind are in widespread use in recommending books, movies, songs, web pages, etc. In his presentation, Patrik Hoeyr from HIIT stated that this technology has yet to be successfully applied to the scientific research literature.

Motivation for recommender systems for science include (1) keeping watch and recommending potentially interesting papers (based on topic), and (2) having found a paper, checking whether it is worth reading. An online 'journal club' of expert discussions would be one solution.

Current systems include search engines for academic search (Google scholar, CiteSeer), bibliography management and sharing (BibSonomy, citeulike, Mendeley, Connotea, RefWorks, zotero), and archives (arXiv.org, PubMed). Other systems include whattosee, journalreview.org, getcited, techlens, pubzone, wepapers, and action science explorer.

Problems include that (1) there is not much commercial potential, (2) one has to draw users to a collaborative filtering system, given that there are few other users in the beginning, and (3) privacy issues have to be balanced with authority and trustworthiness. As a solution to the cold start problem, users have to be provided with some other service and make the use very easy.

As an experimental efforts, Hoyer presented his project grocko.org. The system provides

automated topic analysis of abstracts,
personalized recommendations of newly appeared articles,
trivial set-up, and
anonymous discussions and ratings of existing paper

Teemu Leinonen: Designing Learning Tools - Methodological Insights.

2010-12-07T18:21:00.007+02:00

Teemu Leinonen from Aalto University School of Art and Design defended today his thesis "Designing Learning Tools - Methodological Insights". The opponent, Prof. Gerhard Fischer (Department of Computer Science, University of Colorado at Boulder) is the director of the Center for Lifelong Learning and Design. Prof. Lily Díaz, School of Art and Design, served as the custos.

In his opening talk, Leinonen presented the wish to serve the others as a central motivation for his research work and stated learning to be one of the central human rights. He showed a timeline of the main paradigms of using computers in learning starting from programming/drill and practice to current use of social software and free and open content. The backbone of Leinonen's thesis is based on his responsible role in four web-based learning environment projects including Fle3, MobilED, LeMill and Wikiversity. He summarized the key findings as follows: (1) Design research relying on practice should follow a certain research-based design process, (2) Designers should aim and accept that design is often based on informed guessing, and (3) Designers should be aware of the need to move between different knowledge interests.

A particularly interesting conceptual link made by Teemu Leinonen was a combination of the philosophy of three worlds of knowledge by Karl Popper and Jürgen Habermas' three knowledge interests (prediction, understanding, emansipation). He associated the four steps of research-based design process (1. contextual inquiry, 2. participatory design, 3. product design, and 4. software prototype as hypothesis) with the segments of a matrix in which Popper's three worlds are the columns and Habermas' knowledge interests the columns.

In his questions, Prof. Fischer asked about the concept of learning and the relationship between formal and informal learning. As it has became obvious that we are surrounded by cognitive tools that help us in calculation, remembering, etc., Fischer also wanted to hear what can be condidered as the basic skills of the 21st century to be taught at school. Leinonen's answer included skills in knowledge building, learning to learn, analytical and critical thinking skills used in a community.

An interesting discussion was elicited from a question in which the opponent asked to consider a discussion with Helsinki school board. In this scenario, they would like to know whether the pupils would be given the right to use hand-held calculators (or other tools such as spelling correctors) and to which extent. Another theme of intensive discussion was related to the relationship between the concepts of usability and usefulness of tools. It seemed finally that the discussants agreed on the basic points even though they initially approached them from different angles. As a member of the Computer Human Interaction Academy, Prof. Fischer had a clear view on the field of usability.

Teemu Leinonen is known for his position as an Advisory Board member of Wikimedia Foundation. Related to this and to the Wikiversity developments, Prof. Fischer asked how does one know if information is reliable in Wikipedia. Leinonen responded that people have to be taught media literacy, i.e. how to use Wikipedia including the functionalities of editing, viewing the history of edits, and following and participating the discussions related to the edits. He also pointed out that people need to remember to check the quality of the references. Leinonen concluded that crowd sourcing is not an anarchy but there is a clear social structure. Fischer agreed and stated that media literacy and a critical position is always necessary, regardless of whether Wikipedia, Encyclopedia Britannica or a teacher/professor is inconsideration.

Antti Sorjamaa: Methodologies for Time Series Prediction

2010-11-19T12:25:00.003+02:00

Antti Sorjamaa from the Adaptive Informatics Research Centre at Aalto University School of Science and Technology is depending his PhD Thesis Methodologies for Time Series Prediction and Missing Value Imputation. He is stressing the importance of direct prediction in the case of predicting multiple future values instead of a recursive approach in which accumulation of errors is a clear problem. He also motivates the importance of considering missing values as in many applications there can be a large number of missing measurements. As an example, he uses the environmental state of Tanganjika Lake. The opponent in the defence is Prof. Guilherme de Alencar Barreto from Brazil and the custos is Prof. Olli Simula. The work has been conducted under the supervision of Dr. Amaury Lendasse who is responsible for time series analysis research in the research center.

MultilingualWeb - Where Are We?

2010-10-27T16:36:00.038+03:00

Since the beginning of the World Wide Web in the early 90s, English has been the lingua franca of web technologies. Issues related to supporting other languages have not gotten much attention as the majority of developers have adopted English as the standard language environment in everything from character encodings to browser user interfaces. Now, as the number of people using the internet is growing faster than ever, the need for standardized solutions is apparent to enable the web to scale to thousands of different languages and locales.

In Africa alone, there are more than 2000 spoken languages. More than 100 of these have over one million speakers (Denis Gikunda, Google Africa)

In India, the number of official languages is 22 out of 123 major languages and +2000 dialects (Swaran Lata, Dept. of Information Technology, Government of India)

In some areas of the world, internet access will be available before literacy (Max Froumentin, World Wide Web Foundation)

The current number of internet users is about 1.97 billion with 444.8 % growth in the past ten years (http://www.internetworldstats.com/stats.htm)

The MultilingualWeb project is exploring standards and best practices that support the creation, localization and use of multilingual web-based information. In the first workshop of the project titled "MultilingualWeb - Where Are We?" the project partners and industry representatives from browser vendors to representatives of large multilingual web sites (including Google, Facebook and Mozilla) gathered in Madrid to discuss the current state of standardization in language interoperability.

So, let's use UTF-8

While character encoding is the most obvious issue in enabling support for multiple languages, the consequences go much deeper into the locality. For example, in some languages the text flow can be right-to-left (e.g. Arabic) or even vertical (e.g. Chinese, Japanese, Korean). Combine this with the requirements of producing a portable layout with CSS and you end up with a handful of new issues which show that multilingual interoperability is indeed a big challenge.

In his talk, Richard Ishida of the W3C (World Wide Web Consortium) gave a tour of tentative web standards in the area of multilingual interoperability in the web. Some of the exciting new developments include WebFonts, CSS3, bidirectional text and Unicode locale extensions.

Max Froumentin of the World Wide Web Foundation gave some further insights into the future of internet use that the English speaking audience easily overlook. Namely, a significant proportion of "the remaining 5 billion" people that do not yet use the web are 1) not literate 2) do not speak English 3) are soon able to access the web through mobile devices. Despite these limitations, the ability to communicate through the internet is having a major impact in the organization of local governments.

In fact, it is estimated that in less than five years, the majority of web use is done through mobile devices. There is thus a huge potential for mobile web technologies.

Roberto Belo-Rovella of BBC World Services is all too familiar with the problems of internationalization as their office provides news in 32 languages in the web. It's really quite staggering that despite the availability of standards like Unicode, BBC still publishes text content embedded into images in some languages such as Hindi to circumvent the poor support for Hindi characters in the browsers of mobile phones.

Two sides to a standard

In his keynote talk, Reinhard Schäler of the Rosetta Foundation reminded that standardization is usually a coin with two sides. Standards are indeed necessary for interoperability and standards facilitate technological development since anyone can provide an improved alternative to an existing solution as long as the interface is standardized. But standards can also be restrictive, outdated and impractical. For a standard to be useful, it has to become widely adopted and developers are reluctant to comply with standards that are not up to par with the technological development.

Standardization is also a form of authority and one should question what gives the legitimacy to a standards organization. To this, Ishida's attitude is humble:

"It's your web, not ours [W3C]. Now go and make the most out of it!".

STeP 2010

2010-08-20T11:44:00.004+03:00

The 14th Finnish Artificial Intelligence Conference (STeP 2010), co-organized by FAIS and A?, was held 17-18 August in Otaniemi. The conference had invited speakers around Finland: Jarmo Alander from University of Vaasa, Timo Honkela from Aalto University, Ville Kyrki from Lappeenranta University of Technology, Tapio Salakoski from University of Turku and Harri Valpola from Aalto University and ZenRobotics Ltd. There were also presentations for ten contributed papers and a couple of demos.

Jarmo Alander talked about computational intelligence in Vaasa, where especially evolutionary algorithms and their applications are investigated.

Timo Honkela discussed subjectivity, which is mostly ignored in artificial intelligence systems. The first level of subjectivity considers the fact that each individual agent has its own model of the world. The second level of subjectivity goes deeper and takes into account that symbols in the models are not shared between agents, but have to be grounded individually for each agent.

Ville Kyrki made the claim that robots haven't really improved from Shakey, that was built in the late 60s and early 70s. The most recent significant developments have been made with robots operating in unstructured environments, such as the BigDog robot or the DARPA Urban Challenge.

Tapio Salakoski showed how methods for natural language processing (NLP) can be applied in the bio-medical domain, for instance, by extracting events from millions of scientific publications and using machine learning for finding connections between the events.

Harri Valpola discussed how artificial intelligence should be constructed gradually. They tackle the goal of constructing an artificial brain in such manner, taking into account the evolutionary development of the vertebrate brain. They also use their models for motor control, which they say is the original purpose of the brain.

The presented papers had a range of topics, including solving the Rubik's cube with a genetic algorithm, new genetic operators for image registration, building a platform for software program comprehension, domain adaption for statistical machine translation, and handling missing values with principal component analysis.

LREC 2010 in Malta

2010-05-28T09:33:00.003+03:00

The seventh international conference on Language Resources and Evaluation (LREC) was organized in Malta last week. Tommi Vatanen presented a paper written by himself, Jaakko J. Väyrynen and Sami Virpioja on Language Identification of Short Text Segments with N-gram Models. They have studied language identification task, in which the test samples have only 5-21 characters.

Before traveling to work at CERN for the summer 2010, Tommi Vatanen shared with us some experiences from the LREC conference. Issues included Jaime Carbonell's key note presentation on, for instance, CBMT, Context-Based Machine Translation method, Ralf Steinberger's keynote on NewsExplorer as well as the use of Wikipedia and active learning in which a supervised learning algorithm can actively query the teacher for labels.

Tomi Kauppinen: Thesis on Geospatio-temporal Semantic Web

2010-04-23T12:21:00.004+03:00

Tomi Kauppinen from the Semantic Compuring group at Aalto University is defending his PhD thesis "Methods for Creating and Using Geospatio-temporal Semantic Web". The basic objective of the thesis is to present and evaluate new methods for creating graded relationships for geospatial and temporal ontologies.

Werner Kuhn, Eero Hyvönen and Tomi Kauppinen
The opponent, prof. Werner Kuhn from University of Münster started by stating that many people think that ontologies should be used to model types or kinds and categories of things. In constrast, places like Helsinki are instances of these categories. He asked how did the defendant think of representing place through ontologies. In his answer, Kauppinen illuminated the relationship between categories and instances. Later Kauppinen suggested that RDF is better than OWL for representing graded relationships.

From modeling and simulation to computational science

2010-04-14T12:03:00.006+03:00

Tekes (Finnish Funding Agency for Technology and Innovation) organizes today the final seminar of the MASI (Modeling and Simulation) programme. In his presentation, professor Risto Nieminen from Aalto University School of Science and Technology described computational science as a modern paradigm that unifies experiment, theory and simulation.

Nieminen mentioned that the computational approaches have grown to cover all areas in natural and social sciences, medicine, engineering, and are rapidly gaining ground in arts and humanities. He concluded that computational science continues to revolutionise scientific enquiry and discoverym including new areas (e.g. computational sociology, psychology and behavioral studies) - these are areas in which also our Cognitive Systems research group is active. Nieminen, who was recently appointed as the first Aalto Professor, reminded that Academy of Finland has just announced the first projects that will be funded through their Computational Science Research Programme.

The event was closed by Arto Kotipelto (TEKES) and Pekka Taskinen (VTT) who have been coordinating the MASI Programme.

Seminar on the impact of science

2010-03-11T23:31:00.005+02:00

University of Helsinki (Prof. Markku Löytönen, Department of Geography), Finnish Academy of Science and Letters (Prof. Arto Mustajoki, chair of board) and Academy of Finland (Director Pirjo Hiidenmaa, Culture and Society Research Unit) organized a seminar on the impact of science.

Docent Allan Tiitta provided a view on the history of Finnish science and discussed among things the intended impact of the Royal Academy of Turku when it was founded in 1640 by Queen Christina of Sweden. Dr. Tarmo Lemola (Advansis) described in detail different ways of measuring the impact. He referred to, for instance, OECD's list of different kinds of impact including impact in science, technology, economy, culture, society, policy, organization, health, environment and training as well as symbolic impact.

Prof. Markku Löytönen emphasized the need to see the big picture and the long time scales when the impact of basic science is considered. He gave some examples including the influence that Fourier had on GSM technology. Kjell Forsén, the CEO of Vaisala Inc. described Vaisala's scientific origins that date back to the 1930s when Professor Vilho Väisälä, Vaisala's founder and long-time managing director, invented some of the operating principles of a radiosonde. Forsen also gave a view on how important active research and international collaboration with research institutions is still nowadays important for the company.

Prof. Jarkko Hautamäki provided a overview on the PISA research and how it has influenced the visibility of Finnish educational system. The Finnish educational system has approved to be the best in the world in the PISA results. It is an egalitarian Nordic system with no tuition fees and free meals. Attendance is compulsory for nine years starting at the age of seven. Prof. Mari Vaattovaara described an interesting case in which a university course and researchers' attention to the situation of a suburb had also led into significant public and political attention.

Director Pirjo Hiidenmaa emphasized the idea that the impact of science needs to be considered as a process that has two directions. She reminded that also lay persons have contextual knowledge that can be important in many cases. Our group has recently published a related research report named Modeling communities of experts – conceptual grounding of expertise. This publications considers different kinds of expertise as well as modeling its explicit and implicit forms at individual and social levels.

Prof. Arto Mustajoki provided a big picture on the impact of science. He stressed that some kinds of impacts of basic research are such that their value is immense and not measurable. Some scientific results influence profoundly how individuals and organization act through changed understanding of the world and through adoption of new kinds of methodologies. This impact should be understood as clearly as possible by politicians and policy makers to underline the need to continuously invest on basic research, not only expecting short term results of applied research.

In their concluding talks, Prof. Markku Mattila, the president of the Academy of Finland, and Director Leena Vestala, Department for Education and Science Policy, Ministry of Education also emphasized the importance of considering the impact of research. Prof. Mattila also announced that Academician Leena Peltonen-Palotie had passed away and a moment of silence was spent to honor her memory.

After the presentations, there was a lively discussion during which Dr. Jouni Tuomisto from the National Institute for Health and Welfare told about interesting developments related to Science 2.0. He has initiated Opasnet which is a wiki-based website for helping decisions about human health, and environmental factors affecting it. Opasnet is also open for anyone who wants to promote science-based decision-making in any field.

Workshop on Self-Organizing Maps, WSOM 2011

2010-03-04T14:17:00.004+02:00

WSOM 2011 will be organized at Aalto University School of Science and Technology and Dipoli Conference Center from 13th to 15th of June, 2011, co-located with ICANN 2011 conference. The eventwill bring together researchers and practitioners in the field of self-organizing systems, with a particular emphasis on the self-organizing maps. It will highlight key advances in these and closely related fields. It is the eighth conference in a series of bi-annual international conferences started with WSOM'97 in Helsinki.

Teuvo Kohonen's book on Self-Organizing Maps (Springer, 2001; in the middle) and its Japanese and recent Russian translation.

META-NET: A Network of Excellence forging Multilingual Europe Technology Alliance

2010-02-22T18:24:00.005+02:00

META-NET is a Network of Excellence dedicated to the technological foundations of the European multilingual information society consisting of applications that

enable communication and collaboration among people without language boundaries,
secure users of any language equal access to the information and knowledge society, and
include and push forward functionalities of networked information technology.

We are actively involved in this consortium that is coordinated by DFKI, Germany, and there are twelve additional founding members including Aalto University. The Kick-Off meeting is today (22nd of February, 2010). More information will be available at http://www.meta-net.eu.

Robin Gustafsson: Awareness, Institutional Entrepreneurship, and Contradictions in Emerging Technological Fields

2010-01-22T10:45:00.002+02:00

Our collaborator Robin Gustafsson defended recently his PhD thesis Awareness, Institutional Entrepreneurship, and Contradictions in Emerging Technological Fields. His thesis studies the emergence of awareness and institutional entrepreneurship in novel technological fields (such as functional foods, well-being technologies, electronic publishing and printing, and modular constructional steel), and the contradictions that result from emergence. The thesis further advances understanding about emergence and socio-cognitive dynamics in novel technological fields.

In the defense, professor Nelson Phillips from Imperial College (UK) served as the opponent.

Our collaboration with Robin Gustafsson began when Nina Janasik and Robin organized a course on Socio-Cognitive Fundamentals.

Analogy and Douglas Hofstadter

2009-10-18T23:14:00.009+03:00

Douglas Hofstadter is a well known cognitive scientist. His book Gödel, Escher, Bach: An Eternal Golden Braid has been inspiring people interested in intelligence and cognition since it was published in 1979. The concept of analogy has been central in Hofstadter's work and this was also the main theme when we met at University of Indiana Bloomington on 16th of October.

Douglas R. Hofstadter, Timo Honkela and Nina Janasik
Hofstadter wanted to emphasize that when he talks about analogy he does not talk about analogical reasoning which he finds to be much too narrow view on analogy making as a central cognitive operation. According to his view, analogy making is a basis for recognition and categorization. He agreed that analogy and similarity are closely related concepts which also lead to a discussion on Kohonen's self-organizing map as an approach in which detection of similarities is central.

Hofstadter mentioned that sometimes people have viewed his work through the examples or microdomains that he uses. One example is the Copycat model with Melanie Mitchell that produces answers to such problems as "abc is to abd as xyz is to what?". He stressed, however, that his interest is in the general cognitive processes.

We also discussed the work of Gilles Fauconnier and Mark Turner on conceptual blending which Hofstadter found important from the point of view of recognizing the importance of analogy making.

Hofstadter's impressive series of books including "Gödel, Escher, Bach: An Eternal Golden Braid", "Metamagical Themas", "Fluid Concepts and Creative Analogies" and "I am a Strange Loop" calls for continuation.

Castellani: Sociology and complexity science

2009-09-02T17:15:00.004+03:00

Brian Castellani (Kent State University) writes an excellent blog on sociology and complexity science. Among many interesting items I would specifically like to mention the map of complexity science. The map includes links with a lot of additional information. Castellani also discusses what happened to neural networking.

In 2003 Castellani published an article in which he showed how qualitative researchers can use the SOM to conduct grounded theoretical investigations of large, complex, numerical databases. It was heart warming to read how positively Castellani writes in his blog about our (Janasik, Honkela and Bruun) recent article "Text Mining in Qualitative Research" in which we continue the same line of research and extend it to text data.

Last but not least, it is worth mentioning that Castellani and Hafferty have recently authored a book called "Sociology and Complexity Science - A New Field of Inquiry". I am eagerly looking forward to reading it.

Towards Aalto University

2009-09-02T16:56:00.003+03:00

The joint opening of the 2009–2010 academic year of the Helsinki School of Economics, the University of Art and Design Helsinki, and Helsinki University of Technology took place on Tuesday, 1st of September 2009 in Finland Hall. The rectors of each university were emphasizing the opportunities provided by the merger of these three. Aalto University will begin its operations in January 2010.

Rectors Matti Pursula, Helena Hyvönen and Eero Kasanen

The talks by the rectors as well as the talk by the president of Aalto University, Dr. Tuula Teeri were rather seremonial whereas the student union representatives brought up challenges and new openings in their talks.

Tuula Teeri, Aalto University President

Cognitive Information Processing workshop in Santorini

2009-08-10T12:17:00.006+03:00

Professors Simon Haykin and Sergios Theodoridis chaired a workshop on Cognitive Information Processing in Santorini, Greece in June 2008. The keynote speakers included Simon Haykin, José Principe, Ali Sayed, Bernhard Schölkopf, Naftali Tishby and Timo Honkela. There were many interesting, mostly methodological presentations, including some presentations on the special theme of cognitive radio. The term "cognitive" was discussed and it was concluded that it has been used in different ways covering the concept at different levels of detail.

Sergios Theodoridis and Simon Haykin in Santorini. Theodoridis is
known for his excellent text book on pattern recognition with Koutroumbas.

Naftali Tishby gave an interesting keynote with a focus on the perception-action cycle. Among other things, he considered the topic from the point of view of information bottleneck. There is a predictive channel between the unknown future (channel output) and partially observed past (channel input). An organism links these so that the sensing costs and action value are taken into account.

Panel discussants at CIP'2008 (from left): José Principe, Ali Sayed, Simon Haykin, Bernhard Schölkopf, Timo Honkela, and Naftali Tishby

Zhijian Yuan: Advances in independent component analysis and nonnegative matrix factorization

2009-04-24T16:59:00.004+03:00

Our colleague, Zhijian Yuan, defended his doctoral dissertation "Advances in independent component analysis and nonnegative matrix factorization" today. The opponent was Prof. Fabian Theis from Helmholz Zentrum München and he provided very interesting and insightful discussion on Zhijian's work.

The thesis is theoretical work and develops and analyzes independent component analysis (ICA) and nonnegative matrix factorilization (NMF). They are both interesting methods from a cognitive point of view as they can find interesting and understandable features in an unsupervised fashion without any labeled training examples.

New journal: Cognitive Computation

2009-03-09T17:35:00.002+02:00

Springer has announced a new quarterly journal Cognitive Computation, with Amir Hussain as the Editor in Chief. It covers biologically-inspired computational accounts of all aspects of natural and artificial cognitive systems. The first issue is available.

Dissertation in corpus-based lexicography

2008-12-19T16:02:00.011+02:00

MSc Antti Arppe successfully defended his dissertation Univariate, bivariate, and multivariate methods in corpus-based lexicography - a study of synonymy today at the University of Helsinki. The thesis is a methodological study in modeling lexeme selection in context when considering linguistic variation such as synonymy. The work applies polytomous logistic regression to produce odds for lexeme selection using a number of predictors. The odds for each individual predictor allows interpretation as to in which kind of contexts each lexeme is typically present or absent.

The opponent, Prof. R. Harald Baayen from the University of Alberta (Canada), provided an interesting discussion. It started by the notion of movement away Chomskyan intuition-based language analysis towards corpus analysis that studies how language works with statistical methods. It was followed by complex questions: Are grammars probabilistic? If so, how does the brain handle probabilities? Why does language allow synonymy? How to handle inter-dependencies between predictors? Language changes, how about the models?

Prof. Baayen mentioned the recently started Journal of Empirical Linguistics, which accepts only replicatable submissions that have to include all scripts that can be used to re-produce the results reported in the paper.

Extracting meaning from audio signals

2008-11-20T17:46:00.001+02:00

Associate professor Jan Larsen, from the Intelligent Signal Processing Group at IMM, Technical University of Denmark, gave a presentation on extracting meaning from audio signals. He gave several examples on how to process audio signals to extract information of, for instance, used instruments and genre. Other examples showed how several information sources can be combined to create search engines for music.

He highlighted the increasing importance of cognitive systems and machine learning in future research. As current emerging interests he stated sparse unsupervised models, generation of high-level contextual information, semisupervised learning and user modeling. Games with a purpose, such as the ESP game and reCAPTCHAs, were mentioned. There users typically entertain themselves, and at the same time happen to tag images, recognize digitized text or otherwise augment raw data with high-level meta-data. The results can be used as additional training data for machine learning methods for creating better applications.

The next day, he acted as the opponent in Jarkko Salojärvi's dissertation on using wrong models.

Neural Computation and Psychology Workshop 11

2008-07-29T10:36:00.003+03:00

The eleventh Neural Computation and Psychology Workshop (NCPW11) was held this year at the Experimental Psychology department of Oxford University in historical Oxford on July 16-18. There have been several exciting presentations in the workshop, of which I summarizing a few I found most interesting.

Dr. L. Andrew Coward (Australian National University), whom I had earlier met at AKRR'05 gave an interesting talk about managing neocortex resources. In his talk, Dr. Coward told us about a his artificial neural network model that uses leaky integrators. The structure of the ANN is based on the findings that the receptive fields of the cortical pyramidal cell columns in the brain seem to overlap as little as possible. According to him, the responses in the cortical columns form the information record which serves as a declarative memory, and the association of the columns in which there was an increase in the receptive field simultaneously is the the episodic memory. The background idea of minimally overlapping (i.e. as independent as possible?) features sound very much like Independent Component Analysis (ICA) to me, except that in ICA you cannot add more features to better discriminate between categories. Further talk with Dr. Coward clarified that the outcome of the model is indeed similar to ICA, but the matemathical foundations of the two approaches are fairly different. The simulations are carried out using random patterns in different categories making sure that the inputs of the same category are statistically similar.

Dr. John Lipinski (Institut für Neuroinformatik, Ruhr-Universität Bochum) gave another interesting talk. Test subjects show a so-called spatial drift in behavioral experiments in which they are shown a cross on the computer screen, and then after a short delay, their task is to click the place where the cross was. In the experiments, there is a considerable drift toward more left/right/above prototype depending on the place of the target. According to further experiments, the effect is even stronger when a lexical category label (above, left, right) is shown. Building on these findings, he then introduced a model for simulating the integration of linguistic and non-linguistic spatial systems using dynamic field theory approach to spatial working memory integrated with a competitive spatial semantic network. According to the results shown, the simulation model captures the empirical findings.

Dr. Nicolas Ruh (Oxford Brookes University) presented OXlearn, a new neural networks simulation toolkit. The toolkit contains simple neural network models, a graphical interface in which several parameters can be altered. The toolkit is intended to be an easy-access pedagogical tool, which means no programming is needed to use it (but you can access all the Matlab-based parameters and functions, if you want to). The OXlearn is available at http://psych.brookes.ac.uk/oxlearn for free download both as a toolbox for Matlab, and as a standalone version (only for Windows XP at the moment). Within the field of computer science, one usually wants to teach the students programming as well, but there are several other disciplines in which such a program is probably quite useful.

The invited speaker in this year's workshop was Professor David Plaut from Carnegie-Mellon University (Psychology, Computer Science and the Center of the Neural Basis for Cognition). He gave an insightful presentation about the common cognitive and neural principles of face and word processing in the brain. The research had been carried out together with professor Marlene Behrmann. Distributed presentation of knowledge as a pattern of activation over multiple, hierarchically organized visual areas. According to him, the representations are different in the case of faces and words, but the computation would be same and in both cases the process relies on high visual acuity which enables making fine grained discriminations. In addition, representational coordination and competition and a bias for local connections are important. His conclusions were based by simulations carried out in two simulated systems which on these features, and the results of the simulations seem to produce behavior which is similar to humans in these cases.

The last talk of the three-day workshop was given by Dr. András Lorincz (Eötvös Loránd University, Budapest, Hungary). This talk featured a cognitive model based on factored reinforcement learning and independent component analysis. Like most of the content of the workshop, I will look forward to reading the full paper of this paper the proceedings of the workshop that should be out in December.

The NCPW12 will be probably be held in London in early 2010. I am certainly looking forward to it!

Models and simulations of consumers and rhythms of everyday life

2007-12-14T12:54:00.001+02:00

Our Kulta project organized a seminar on 13th of December at the Helsinki University of Technology. Mika Pantzar, professor at National Consumer Research Center and researcher at Helsinki School of Economics, showed interesting examples of rhythms of everyday life. The topic is related to his collaboration research with Prof. Elisabeth Shove from Lancaster University.
Pantzar presented the idea of considering everyday life through the concepts of melody (sequentiality), harmony (synchronicity) and rhythm. He showed interesting examples of various kinds of time series. For instance, the time used for eating in Spain and France is clearly clustered around lunch and dinner times whereas in Finland people are eating throughout the day rather evenly.

Dr. Amaury Lendasse, the head of the group on Time series prediction and chemoinformatics, gave a tutorial on time series analysis. He presented a number of examples in the areas of economics, physics, industry, astronomy, climatology and hydrology. Lendasse made a distinction between recursive prediction, direct prediction and hybrid prediction in long term time series prediction. He also discussed links between variable selection, scaling selection and distance measures.

In the end, we discussed some similarities and differences between various disciplines that consider future. Time series prediction aims at developing methods for predicting the future values of some numerical variables. On the other hand, scenarios are created from a qualitative point of view. Similarly, control aims at manipulating some variable values in such a way that some quantitatively measurable process can be directed into wanted direction. Decision making aims at directing some process into wanted direction through measaures that can be described mainly at qualitative level.

Merja Oja: Methods for Exploring Genomic Data Sets

2007-12-14T12:13:00.000+02:00

Merja Oja, a resercher in our laboratory, is defending her thesis "Methods for Exploring Genomic Data Sets: Application to Human Endogenous Retroviruses" for the moment. The opponent is Professor Hiroshi Mamitsuka from University of Kyoto. In Merja Oja's thesis, exploratory data analysis methods have been developed for analyzing genomic data, in particular human endogenous retrovirus (HERV) sequences and gene expression data. HERVs are remains of ancient retrovirus infections and now reside within the human genome. The work has been conducted in Professor Samuel Kaski's group on Statistical machine learning and bioinformatics.

NICSO 2007: Nature inspired cooperative strategies for optimization

2007-11-10T11:20:00.000+02:00

The second international workshop on nature inspired cooperative strategies for optimization (NICSO 2007) was held in the beautiful baroque town of Acireale in Sicily. The topics of the workstop ranged from theoretical considerations of complex network models to hardware implementations of swarm computing.

In his plenary speak, Marco Dorigo (Université Libre de Bruxelles) presented his ongoing work on swarm robotics. In a series of videos Dorigo showed, for example, how a group of simple robots emreged a collaborative self-assembly strategy so that a group of robots was able to move past physical obstacles that any single robot could not pass. In another robotics related plenary speak, Paolo Arena (University of Catania) presented their work on emergent adaptive locomotion controllers which enabled a legged robot to lear to walk and plan actions autonomously.

Xiaohui Cui's (Oak Ridge National Laboratory) presentation dealt with clustering of text documents after transferring them into a vector format using word frequence information. An inspiring solution to this computationally intensive task was to use modern graphics processing units (GPUs) which have up to 128 cores on one low cost card. Other interesting work related to hardware implementations of biologically inspired parallel computing included Francesca Palumbos's and Giovanni Busonera's papers on an architecture for self-organization on the hardware level.

Artificial immune systems, a topic of ongoing research at TKK, was referenced in two presentations. The comparison of the neural network paradigm and the immune network paradigm for prototype based clustering was the topic of our own paper while Slavisa Sarafijanovic presented a tool for using negative selection in a collaborative spam filtering environment.

Emergence of early cognition, communication and language

2007-11-05T14:22:00.001+02:00

Our research group has been organizing a language technology seminar "Emergence of early cognition, communication and language: from humans to machines", coordinated by Krista Lagus, Oskar Kohonen and Timo Honkela.

Today Don Killian was discussing the book Origins of the Modern Mind by Merlin Donald. According to Killian, Donald's book synthesizes various fields, bringing together theories of cultural origins of human cognition in a cohesive approach. In the book, three major "abrupt" cognitive transformations are brought up by which the human mind developed over millions of years.

First major change in brain structure was the transition period leading to Homo Erectus. Brain enlarged from 600 cc to 1,100 cc in late Homo Erectus. Second major change was related to the arrival of Homo Sapiens, including a further increase in brain size. The period was also characterized by a continual acceleration of cultural change, more frequent innovations, development of language, and vocal apparatus change.

In his book, Donald points out that primates could not, and cannot, reinvent symbols, only recombine already symbols given to them. Then a natural question is: What is missing with primates? One good candidate is that fact that humans possess three types of memory: procedural, episodic and semantic. Semantic memory is a system of representation that forms semantic networks and the final change. Other primates are restricted to episodic memory: a virtual "flashback" of previous performances. Thus, there is a lack of performance ability.

Donald explains that misesis is the missing link. Mimesis includes non-lingustic, self-initiated and representational intentional acts. It incorporates mimicry and imitation to a higher end, voice tones, gestures, facial expressions, and patterned wholy-body movements. According to Donald, mimesis probably did not originate as a means of communication, but the presence of mimetic skill however would have lead to a form of social communication.

Mimetic culture has included, for instance, toolmaking, fire use, and coordinated seasonal hunting etc. A change from a mimetic to a mythic culture was crucial for the emergence of modern language. Approximately 50,000 years ago the biological transition from erectus was complete. It lead to Upper Paleolithic, or stone-age culture. The descent of larynx signified cognitive change and resulted in speech and language. Mythic inventions were conceptual models of the universe.

Ivan Berazhny was giving a presentation with the title "On the trail of social semiotics?". He started by stating that being very specialized is often a kind of self-defence. Society could be presented as a system of meanings.

Berazhny gave three perspectives on meaning-making: a) Ontogenetic: an individual/development; b) Phylogenetic - a semiotic resource/evolution; and Logogenetic - a text/unfolding discourse. He asked how people learn to mean. He summarized that humans learn to representre reality, to enact social relations to organize meaning-making in the flow of time.

Berazhny was also discussing a movement from semiotics to social semiotics, and from systemic functional linguistics to multisemiotic studies. In a multilevel stratified model of language, context includes ideology (a model of social organization), genre (a context of culture, i.e. configurations of meanings as they unfold through stages) and register (a context of situation). On the other hand, content includes semantics (configurations of meaning as in figures and taxonomies) and lexicogrammatical information (configurations of meanings in wordings).

According to Berazhny, metafunctions of language include a) ideational (logical, experiential), b) interpersonal, and textual functions.

The presentation lead into active discussion. In the end of his presentation, Berazhny mentioned the conference "Linguistics and multisemiotic challenges in Europe in the world".

KITES symposium on multilingual communication and content management

2007-10-25T20:44:00.000+03:00

Kites is a new Finnish organization that coordinates collaboration between companies, universities, educational institutions and public organizations related to multilingual communication and content management. Today took place the first day of a Kites symposium, organized in Kouvola. In the first presentation, Markku Saari and
Arto Leinonen, CEO of Kielikone Ltd., provided an overall view on the mission and objectives of the organization.

Juhani Lönnroth, the director-general of the directorate-general for translation of EU commission, gave a presentation on the EU strategy for multilingualism. He started by describing the historical path from national languages into multilingualism. Lönnroth brought up challenges for the language industry which is one the fastest growing business areas in the world and which provides work for a large number of people. He emphasized the central role of language in political and business processes.

Currently in EU there are 23 official languages and thus 506 language pairs. In the EU markets, messages and products need to be localized, the use of global tilingualism in the EU. He emphasized that the basis is one's own mother tongue. If one is not able to express oneself well in one language, it is not possible to communicate efficiently in any other language. In order to facilitate root level communication among EU citizens, the EU Commission wants that all Europeans would master at least two other languages than their mother tongue. The third dimension consists of the communication between citizens and public organizations. Finally, the EU-level public organizations have to find ways for multilingual communication within themselves. Towards the end of the presentation, Lönnroth was describing areas in which Kites and EU may have collaboration including:

life-long learning programme 2007-2013 (ERASMUS, GRUNDTVIG, COMENIUS, LEONARDO DA VINCI, ERASMUS MUNDUS)

databases and translation memories (IATE, EURAMIS, etc.)

systems for content management

internet-based communication (WCMS)

localization of messages

research activities

machine translation (CAT, AI)

digitized libraries, documentation systems

Ma Keqing, Chinese ambassador to Finland, described the relationship between China and Finland in several areas, reminding that these countries have had diplomatic relations for 57 years. Several examples of active business relations were given. The value of the trade is 8 billion dollars, Finnish investments in China reach 5 billion euros, Nokia's market share in China is 35%, and Finnair is the third largest international airline operating in China.

Cultural and educational exchange between China and Finland is becoming more and more active. A Chinese-Finnish general dictionary is under preparation and a Chinese-Finnish bilingual school begins its operation in Helsinki next August. Ma Keqing mentioned that Finland is known rather well among Chinese even though mostly the Chinese only know Nokia's products, Santa Clause and sauna as Finnish phenomena. She described the rapid development of Chinese society during the past decades, emphasizing the increasing openness of the society. She described Chinese cultural aspects including the holistic thinking. She discussed the phenomenon of globalization that provide opportunities for both China and Finland for mutual benefits.

Käthe Sarparanta communication manager for Olkiluoto nuclear power station project, was describing this very large industrial construction project, the budget of which is approximately 3 billion euros. The network involved in the project consists of over 1700 companies and people working for the project are of over 30 nationalities. Sarparanta payed attention to the different cultures within the project that creates various challenges. The current 2350 workers need services in their own languages. For instance, in Rauma there is a French school. The number of documents is currently reaching one million. In the end, the documentation will take about 2.5 kilometers of shelves. To manage the translation task, a terminology is being gathered, currently with 8000 terms. Sarparanta concluded by describing the effects of the large project on the local communities.

Juhani Reiman, CEO of Lingsoft Ltd., was emphasizing the importance of language in the modern society, comparing the potential of language technology with the industries grown based on electricity. His main topic was to describe the commercialization of speech technologies. Reiman pointed out that in Finland there is much more room for services the use of which is based on spoken dialogue with machine. He gave a live demonstration of the Lingsoft speech portal using a mobile phone. The speech synthesis was based on the technology developed by Bitlips Ltd. (http://www.bitlips.com/). Reiman emphasized the idea that speech dialogue systems will be important in the future.

Esko Vario, program manager from Microsoft Finland, was describing Microsoft's activities related to multilingualism. He emphasized the importance of understanding the user, including understanding their culture and values. Relevant information can be obtained using ethnographic studies to support development of understandable and usable products. Localization tools include Trados and Microsoft's own tools. Vario told that terms are collected to a database called TermStudio, currently with 15,000 concepts, 20,000 English terms, and almost 100 languages.

My own presentation discussed how research and development on multilingual communication and content management is and can be affected be research on statistical machine learning and pattern recognition.

Volker Tresp: Infinite Hidden Relational Model

2007-10-18T15:54:00.000+03:00

Dr. Volker Tresp from Siemens AG is currently visiting our laboratory. He will act as the opponent for Timo Similä's doctoral defense tomorrow on Friday 19th of October. The title of Similä's thesis is "Advances in variable selection and visualization methods for analysis of multivariate data".

For the moment, Volker Tresp is giving a talk on "The Infinite Hidden Relational Model". The IHRM can be used to realize nonparametric relational Bayes. He suggests that the IHRM might be an interesting model for a number of relational problems. He presented a recommendation system, prediction of gene functions and medical decision support as applications.

MT Summit XI

2007-09-13T18:54:00.000+03:00

The 11th Machine Translation Summit (MT Summit XI) is taking place in Copenhagen, Denmark, and brings together researchers, developers and even some users. Invited speakers have included, for instance, Stephen Richardson from Microsoft Research and Philipp Koehn from the University of Edinburgh.

Microsoft has recently launched their free online machine translation on Translator Live Beta. It is based on Systran and includes some alternative systems from Microsoft for technical domains. Systran is a rule-based system and is the backbone of most machine translation systems provided online, including Google's translator.

Koehn introduced the EuroMatrix project that tries to facilitate resource and tool sharing between researchers, a framework for evaluation of systems for machine translation between all the official languages in the European Union. His talk and the following discussion pointed out the fact that statistical machine translation research is reliving the past of rule-based machine translation research, as there is more and more demand for implementation of syntactic and semantic analysis in the statistical machine translation framework.

Philipp Koehn giving his talk.

Visit and presentation by Prof. Simon Haykin

2007-07-04T09:59:00.000+03:00

Professor Simon Haykin was visiting our laboratory last week. Prof. Haykin is the director of the Adaptive Systems Laboratory of McMaster University. Simon is also well known for his books, including "Neural Networks: A Comprehensive Foundation".

Prof. Haykin gave a presentation on "Coherent Independent Component Analysis", related to the work by his student Kevin Kan. The basic setting was related to the question how to model processing of speech. The Coherent ICA is based on Infomax, Imax and copulas. Imax algorithms are based on the idea of maximizing the mutual information between the outputs of different network modules, and are capable of extracting higher-order features from data. Copulas are functions used in statistics for modeling dependencies between random variables. Prof. Haykin showed in his presentation that the model can be useful for solving real problems including source separation and auditory coding.

I had the chance to have an inspiring discussing with Simon on issues related to cognitive systems and agreed on various ways of future collaboration.

Reasons for a gap between theory and practice

2007-06-18T10:26:00.000+03:00

In the Third Organization Studies Summer Workshop, Alfred Kieser and Lars Leiner had a paper on “Why Collaboration with Practitioners Is often Referred to in Management Science as a
Remedy for the Rigor-Relevance-Gap and Why This Is Not a Promising Idea”. Alfred Kieser is Professor of Organizational Behavior at the University of Mannheim. Alfred mentioned in his presentation several reasons why theory and practice may not meet. There are problems of access: scientific information may not be readily available for a practitioner. It may be that relevant information exists but it presents itself as information overflow. Scientists apply a multitude of different approaches to a particular problem and it is very difficult to assess which point of view is relevant for practice. Scientists also produce contradictory results depending on their theoretical framework and many practical choices. Sometimes the results are simply trivial from the practical point of view, for instance they may just confirm existing practices and intuitions. Last but not least, scientists speak for scientists: the scientific form of presentation is often not suitable for “consumption” in practical contexts. Alfred amused the audience by showing some mathematical formulae out of context.

Alfred Kieser and Sara Rynes discussing

In general, Alfred Kieser painted a quite dark picture about the possibility of being relevant for practice. This view, maybe presented in a provocative manner on purpose, was considered too pessimistic by many. For instance, Alper Alsan and Kathleen Park gave a presentation in which they described their collaboration as a researcher and a practitioner.

In summary, I would say that building a bridge from theory to practice is a challenging task. On the other hand, I would strongly claim that scientists are in a central role in protecting and producing wellbeing in the world. It is sad to see when people do not understand and appreciate the complex paths that the scientific results take to practice. Scientific understanding of evolution, human mind, physiology, materials, complex systems, etc. have fundamentally changed the way how human beings and societies live their lives. This does not mean, however, that scientific innovations only take place in formally organized institutions and projects.

Organization Studies Workshop: Keynote by S. L. Rynes

2007-06-11T00:19:00.000+03:00

The Third Organization Studies Summer Workshop was organized in Crete in early June. The event was well organized and raised many important issues. The gap between theory and practice was one of the recurring theme. The same theme was also phrased as the question how to combine successfully scientific rigor and practical relevance.

The first keynote speaker was Sara L. Rynes, Professor at University of Iowa, who discussed the theme how to make research relevant to practice. She is the editor of Academy of Management Journal. She emphasized the values of pluralism in research and welcomed respect for co-produced research, based on an inductive and social constructive approach. This means that researchers should work in collaboration with practitioners gathering qualitative data in real-world cases. She used her own experience from the 1990s as an example.

She showed how ready-made questionnaires may not succeed in bringing up utterly important factors. Through open interviews the researchers had figured out the importance of recruitment in job choice. Sara also called for broader views of relevance and research that develops insights that helps managers. She referred to Jim March in mentioning the combination of academic and experiential knowledge. Academic research can produce generalizations and frameworks which need to be applied through human intuition and contextual understanding.

Sara Rynes also emphasized the idea that usefulness to practitioners alone is a deficient vision for organization studies. She mentioned the book Freakonomics in which N.N. brings up important findings that may be “inconvenient truths” for many. I also very much agreed with the conclusion that for longer term and for the society as a whole, researchers need to be certain kind of watchdogs. Researchers should be allowed to have the role that is commonly attributed to free press. If this role is not fully appreciated the society does not function as well as it could.

It may, of course, be convenient for the leaders to think that the researchers can support directly their aspirations and policies. This is, however, bound to lead both to bad science as well as into a declining nation. If a decision makers fail to listen to the inconvenient questions that researchers may ask they finally are going to make much less optimal decisions than through taking into account even the critical voices. This does not, of course, mean that the scientists should be made the primary decision makers.

Towards the end of the presentation, Sara Rynes mentioned that researchers within organizational studies ought to be think more about data mining. This was nice to hear for us three, Sari Stenfors, Tanja Kotro and Timo Honkela who participated the workshop as Kulta project researchers. The project develops methods for modeling and simulating changing customer needs. In addition to practice theory and qualitative research, data mining and other adaptive informatics methods are in a central role in the project.

Nodalida 2007

2007-05-26T10:56:00.000+03:00

The 16th Nordic Conference on Computational Linguistics (Nodalida 2007) is underway and presents the current status of language technology in the Nordic countries. This time the conference is held in the city of Tartu in Estonia. The weather has been favorable the whole time and we have been able to enjoy the city and the extremely well organized conference.

Diana F. McCarthy from the University of Sussex gave an interesting invited talk on lexical substitution, that has been used to evaluate various word meaning discovery methods based on different representations, principles and inventories in a shared task. Results of the competition will be published in SemEval-2007.

Walter Daelemans, from the University of Antwerp, gave another invited talk in text analysis and machine learning for stylometrics and stylogenetics. They have investigated, for instance, author verification and personality identification.

Sari Stenfors: Strategy Tools and Strategy Toys: Management Tools in Strategy Work

2007-05-04T13:11:00.000+03:00

For the very moment, Sari Stenfors who is working with us within the Kulta project is defending her thesis "Strategy Tools and Strategy Toys: Management Tools in Strategy Work" at Helsinki School of Economics. Sari has 15 years experience in managerial positions and works nowadays at Stanford University. In her lectio precursoria, Stenfors referred to the dissonance between theoretical and practical points of view and was opening discussion on how to find useful ways to bring these points of view together in a productive way.

Defendant Sari Stenfors, custos Professor Jyrki Wallenius and opponent
Professor Richard Whittington (some minutes ago at the time of writing).

The opponent, professor Richard Whittington (Said Business School, University of Oxford) gave a positive opening statement. He concluded that Sari Stenfors' thesis shows one of the ways strategy work should go. Whittington's first question was motivated by the notion of strategy toys which lead him to ask about Stenfors' philosophical view on human beings and if they are naturally homo ludens.

Statistical machine translation recipe

2007-04-25T09:40:00.001+03:00

Text-to-text statistical machine translation systems have become easy to rustle up. For instance, Proceedings of the European Parliament have been collected into the Europarl corpus, which provides the same text in 11 languages. Most importantly, a state-of-the-art statistical machine translation system, Moses, is being developed under the LGPL and it handles both training of the system and translating new sentences. These are the two main ingredients that you need to have. At least for now, the Moses system is not optimized, so don't be stingy with memory.

During the final seminar of FENIX, an interactive computing programme by Tekes, I was presenting an online demo of a Finnish-to-Swedish statistical machine translation system based on Europarl and Moses. Our research interests include improving translation quality in an unsupervised manner when translating from or into Finnish as well as other languages with compounding and agglutinative nature.

Speech-to-speech translation

2007-04-20T16:04:00.000+03:00

A first prototype of a speech-to-speech machine translation system is being built in our laboratory. David Ellis, a visitor in our lab, was demonstrating the first results yesterday. Especially translation results from Finnish to English were promising. The current system deals with biblical language due to the availability of necessary resources, including a parallel corpus of spoken language.

Dr. Mikko Kurimo (on the left), David Ellis and

Dr. Mathias Creutz pleased with the results.

Recognizing patterns

2007-03-30T01:07:00.000+03:00

On Thursday 29th of March, Matti Aksela, a researcher from our laboratory, defended his thesis on "Adaptive combinations of classifiers with application to on-line handwritten character recognition”. Opponent was Prof. Robert P.W. Duin (Delft University of Technology). A day earlier, prof. Duin had given an interesting presentation on "Non-Euclidean Problems in Human Centered Information Processing".

The continuation of the thesis party took place in restaurant Mecca in which one of the most talented contemporary Finnish violinists, Pekka Kuusisto, had a concert in which he played Cafe del Mar style music chosen by Dj Yuhis. As the last piece, Kuusisto played Bach's g minor Adagio from six Sonatas and Partitas for Violin. A great day altogether!

AKRR 2008: Adaptive Knowledge Representation and Reasoning

2007-03-16T17:09:00.000+02:00

AKRR 2005 was a successful conference that emphasized the need to take into account the adaptive aspects related to knowledge representation and reasoning. The AKRR conference series is planned to become a triennial event. According to the preliminary plans, the next conference, AKRR 2008, will be held in early June 2008.

The main themes of the conference will be:

Adaptive systems in economic sciences and organizational theory,
Sapient systems: computational wisdom,
New generation of semantic web: social and multimodal grounding of knowledge and understanding,
Adaptive systems in medical education, research and practice, and
Adaptive machine translation: towards global connection.

Actions, concepts, activities and conceptions

2007-02-19T14:42:00.000+02:00

In Southern Finland, it is the winter holiday week at schools. This week also brings in mind the concept of activity. Activity theory is a framework within social sciences that focuses on cultural and technical mediation of human activity. There is a long history of activity theory in Finland thanks to Prof. Yrjö Engeström and his colleagues. A related theme, social aspect of learning has been emphasized by Dr. Kai Hakkarainen and his research group and Prof. Kirsti Lonka.

I was inspired last Thursday by a lecture given by Timo Rauhala who is, for instance, the trainer of the golfer Mikko Ilonen. Ilonen became last weekend the first Finn to win on the European Tour. In cognitive science, one interesting research theme is the role of action. In this blog, there was earlier discussion on the relationship between the level of raw visual perceptions and the linguistic level. It appears that symbols, like words in language, need to be grounded before they can be effectively used. Concepts may not only be based on grounding in the visual domain but many of them stem from the world of actions or activities. Our actions also influence the perceptions of the world. Often the action is deliberately meant to change the perception, like in the case of attention mechanisms. One of the similarities between visual and kinesthetic realms is that they both tend to be much more detailed than the linguistic domain. Referring again to the earlier discussion, an image contains typically many more details than are truly described at linguistic level. In the same way, for instance our movements are very refined when they are compared with the expressions that are used to describe them.

Another interesting theme is the relationship between activity theory and practice theory. Let's consider this issue later.

Discussion around practice theory and some new results

2007-02-15T06:15:00.000+02:00

Oskar Korkman and Mika Pantzar wrote earlier this month an article for Talouselämä that is the largest weekly economic journal in Scandinavia. Their article, entitled "In everyday life customers are similar" discusses the importance of everyday practices as a basis for business development and consumer research. They consider, for instance, the role of segmentation in business. They also refer to Clayton Christensen who has presented the idea that segmenting markets by price point, product type and customer demographics does not reflect the way customers actually experience. This theme is brought up in a book called "The Innovator’s Solution: Creating and Sustaining Successful Growth" by Christensen and Raynor. Korkman and Pantzar continue by discussing the history of customer segmentation, for instance, how Alfred Sloan as the director of General Motors in the 1920s applied the concept of segmentation. They conclude that the point of view of everyday practices can be used as a unifying principle when the customer market is considered.

Yesterday I gave a presentation on Kulta project at Fujitsu that is one of the case study partners in the Tekes project. The project builds on practice theory that is used to model and understand changing needs of customers. The presentation outlined the basic principles of practice theory, its applications, and how the domain could be modeled using adaptive informatics methods.

A snapshot of the Pracsim simulation.

One recent development was also brought up in the presentation. Based on the core ideas of practice theory by Mika Pantzar and Elisabeth Shove, we have developed a simulation and visualization environment called Pracsim. A Java-based demo and a technical report by Lasse Lindqvist, Timo Honkela and Mika Pantzar are available. The audience agreed with a finding that the simulation is visually quite captivating. This conclusion was made also when the demo was recently shown in a meeting at RAY that is another case study partner in the Kulta project.

Subjectivity, science and measuring

2007-02-09T11:47:00.000+02:00

Since having been involved in a serious attempt to create a natural language understanding system in the 1980s, I have been interested in understanding and modeling the inherent subjectivity of understanding. Philosophical discussions within hermeneutics have pointed out related themes for some time already. However, it seems that much of the scientific practice still goes on as if we could assume that objective knowledge exists, ready to be represented with formalisms such as predicate logic. This is not true, to use this problematic word: Symbolic logic is actually of little use if we wish to deal successfully with many of the central issues in epistemology. The relationship between what is called reality and the expressions used to describe it is highly complex. One phenomenon widely recognized is vagueness or fuzziness, a concept the scientific history of which is discussed by Rudolf Seising is his interesting article. Fuzziness is, though, only one aspect among many. The dynamics and the adaptive processes in different levels of abstraction need to be taken into account. It is not realistic to say that language exists somehow miraculously without the community that learns it, uses it and molds it.

An illustration about the relationship between the "reality" in different levels
(an image, a model) and the idea of representing the reality as symbolic structures.

One of the consequences of taking the complexity of natural language understanding seriously is that the methods for philosophy, science and technology of epistemology need to include statistics, probability theory, dynamic systems theory, statistical machine learning, simulation of agent communities, etc.

Another very important consequence is related to the practices in and basic assumptions related to science: we should not consider science as a practice that aims at collecting a set of true propositions of the world. Our collaboration with researchers like Prof. Mika Pantzar has brought into our attention the idea of considering human activities as practices. It might be interesting to consider: what is "sciencing"? (rather than: what is science?) The role of language is hugely central within the practices of science. The results of science are reported and distributed mainly using language. If the statistical nature of understanding language among human beings (even among specifically educated scholars) is to be taken seriously, this should also be reflected in our practices in science. How to do it? That is worth another story.

The discussion above is largely motivated by the complex relationship between "raw perceptions" of the world and the descriptions of it using natural language. A related theme is measuring. In science and technology measuring is a central task. In quantum physics one can consider, e.g., the Heisenberg uncertainty principle, or the Kochen-Specker Theorem. In the 6th Framework Programme, the EU Commission has launced an interesting initiative within New and Emerging Technology called Measuring the Impossible:

"Measurement underpins science, but how do you measure the subjective? Can science use advances in technology to uncover the quirks and imponderables of the human mind, and how people interact with the world? The answer is that science is trying to. A new set of EU research projects is looking at the interface between different disciplines and the human experience; somewhere between psychology, engineering and physiology comes Measuring the Impossible."

ESTSP 2007: European Symposium on Time Series Prediction

2007-02-08T00:27:00.000+02:00

European Symposium on Time Series Prediction is taking place in Espoo from 7th to 9th of February. The symposium is chaired by Dr. Amaury Lendasse from our laboratory. Invited speakers include Prof. Manfred Deistler, Prof. Michel Verleysen and Prof. Donald Wunch.

Dinner at a Lappish restaurant in Helsinki; from the left:
Timo Honkela, Erkki Oja, Michel Verleysen,

Amaury Lendasse, Manfred Deistler and Donald Wunch.

Andrew Chesterman: Translation Strategies

2007-02-02T20:08:00.000+02:00

Machine translation is a great challenge. In order to make the scope of the task clearer, our research group invited prof. Andrew Chesterman to give us insights on some of the issues that human expert translators take into account. Andrew is Professor of Multilingual Communication at the Department of General Linguistics, University of Helsinki. He is the author of several books including "Can Theory Help Translators? A dialogue between the ivory tower and the wordface" (with Emma Wagner, 2002).

The presentation covered some aspects related to translation theory. For instance, the relation between translations and their source texts is a complex issue. Translations are assumed to be somehow "the same" as the original, but they are obviously different. One can ask what does the equivalence mean. A central topic of the presentation was translation strategies: what kinds of textual changes do translators make? Another important insight we got was that there are many kinds of potentially good translations, depending on the objective.

Discussion after Prof. Chesterman's presentation. Dr. Mathias Creutz

explaining issues related to statistical machine translation.

Scholarpedia: Encyclopedia of Computational Intelligence and Computational Neuroscience

2007-02-01T16:57:00.000+02:00

Scholarpedia is a free peer reviewed encyclopedia written by scholars from all around the world. It is based on an interesting concept: articles in Scholarpedia are not frozen, but subject to an ongoing process of improvement moderated by their curators. Dr. Eugene M. Izhikevich, Editor-in-Chief of Scholarpedia, continues: "Similar to Wikipedia, every registered user can revise and expand articles in Scholarpedia. The revision can be just a simple grammar fix, an attempt to rewrite an obscure paragraph, a suggestion on how to improve the quality of the article, or an in-depth review of the article with major additions and modifications. In this sense, every user is a reviewer."

Recently, the article on the Self-Organizing Map (SOM) became available in Scholarpedia. The article contains, for example, a short history of the SOM and useful information when applying it. The popularity of the SOM continues strong which I was able to see when glancing through how many articles were submitted to each topic area in the IJCNN conference.

Dr. Walter J. Freeman has written an interesting article on intentionality. He defines intentionality as the circular process of generalization/abstraction of input and specification/concretization of output by which brains achieve understanding of their environments through the cycle of prediction, action, sensation, perception, and assimilation by learning.

Our old friend Professor Robert Hecht-Nielsen (who, for example, gave the dinner speech in the WSOM'97 conference) has written a thought provoking article on something that he calls confabulation theory. The theory, in his words, offers a comprehensive detailed explanation of the mechanism of thought. Citing Hecht-Nielsen, "confabulation theory proposes that cognition is a phylogenetic outgrowth of movement and that cognition utilizes the same neural circuitry that was originally developed for movement." The idea that the motor control and movement would have a central role has certain appeal. In our university, Dr. Harri Valpola has presented thoughts along the same lines.

Language Technology Education in Finland

2007-01-25T14:50:00.000+02:00

The representatives of the KIT network on language technology education and the Graduate School of Language Technology in Finland had a meeting today at University of Helsinki. The education in the area of language technology is increasingly networked. Several universities provide courses that the students from other universities may participate.

Recently, much thanks to the active role of Prof. Kimmo Koskenniemi and Robin Cooper, there are new developments in the Nordic and Baltic level. NGLST, Nordic Graduate School of Language Technology is a collaboration between leading centres of research and teaching in language technology in the Nordic countries, with funding from NordForsk. Prof. Koskenniemi has also been active in the establishment of the Northern European Association for Language Technology (NEALT).

From the left: Konsta Koppinen (Tampere University of Technology), Stefan Werner (University of Joensuu),
Hanna Westerlund (University of Helsinki), Kimmo Koskenniemi (University of Helsinki),
Eero Sormunen (University of Tampere) and Helena Ahonen-Myka (University of Helsinki)

Cognitive development before birth

2007-01-17T10:54:00.000+02:00

The Faculty of Behavioural Sciences at University of Helsinki organized on Tuesday, 16th of January an event in which Docent Minna Huotilainen and Professor Katri Räikkönen-Talvitie were discussing the theme "Life before birth". The lectures inspired the audience to present numerous questions, leading into a lively discussion chaired by Professor Kirsti Lonka.

Huotilainen was describing the results of brain research, e.g.,on the early auditory skills. During the last three months before the birth the unborn child learns many kinds of things through what she hears. For example, a newly born baby clearly prefers music
that she has heard, whether rock or classical. Huotilainen has coined the term "audio home" referring to the sound context that the baby already learns to appreciate before the birth.

Prof. Katri Räikkönen-Talvitie and Doc. Minna Huotilainen.

Räikkönen-Talvitie discussed how weight at birth may matter in psychological life-span development. She was describing evidence on how low weight at birth correlates with development of psychological factors, such as depression, that may later lead into acquiring chronical deseases. She was also showing results on the effects of contextual factors such as parental optimism.

Finnish-Swedish machine translation challenge

2007-01-11T09:59:00.000+02:00

We organize a special course on language technology, each year with a different topic. Now the topic was Finnish-Swedish machine translation. The student groups were given the challenge to build during the fall a working prototype of a MT system. Two groups built a hybrid system and one fully statistical system. The statistical system got the best BLEU scores but with very small margin to the second group. There were also differences in the experience level among the participants in the area of MT. Mathias Creutz (on the right) announced the evaluation results.

It was concluded that the general level of the results was good, especially taking into account the short time the groups had available for the whole project. Many problems remained, of course, but the students told that they got good hands-on experience on the theme.

Participants of the Finnish-Swedish Machine Translation Challenge.

The course was organized by the Laboratory of Computer and Information Science in collaboration with the Department of Translation Studies at University of Helsinki and the Department of General Linguistics at University of Helsinki. The students were able to use the resources and tools provided by the Finnish IT center for science, CSC.

Happy New Year!

2007-01-01T14:34:00.000+02:00

It is time to wish everyone Happy New Year 2007! The change of the year was celebrated in Finland traditionally. In Helsinki, the city had organized an event in the famous senate square. The magic of the midnight moments was created by Sibelius' beautiful violin concerto and magnificient fireworks.

The large crowd of celebrating people was welcomed by the Lord Mayor of Helsinki, Jussi Pajunen. He expressed the fact that Finland has reached its most prosperous state in its history and reminded that the social responsibility to take care of those who have difficulties in their life always exists. He also warmly welcomed all those who move to Finland from abroad and reminded that they need to be welcomed by all who live here.

IJCNN 2007 Special Session on Philosophical Aspects of Neural Network Modeling

2006-12-28T20:13:00.000+02:00

In collaboration with Dr. Tarja Knuuttila and Anna-Mari Rusanen from the Department of Philosophy, University of Helsinki, we are organizing a special session on Philosophical Aspects of Neural Network Modeling for IJCNN 2007. International Joint Conference on Neural Networks will be organized in Orlando, Florida from 12th to 17th of August. More information on the session is available at http://www.cis.hut.fi/tho/ijcnn07.html. The deadline for submissions is by the end of January, 2007.

Finnish Science in International Comparison

2006-12-19T17:00:00.000+02:00

Academy of Finland organized today an event in which the report on Finnish Science in International Comparison was presented. The report, published in the Academy of Finland publication series, is written by Senior Science Adviser Annamaija Lehvo and Science Adviser Anu Nuutinen. It is available on the Academy's website. The number of publications by Finnish researchers in international scientific journals has increased 2.5-fold during the past 20 years. In 2005, Finnish researchers produced 8,300 publications, the highest figure on record. Finland ranked eight in a comparison of the citation impacts in all OECD countries.

In general, the occasion highlighted the strong position and impact of Finnish research. However, Finns are rarely completely satisfied with their own achievements but assume that something better can be achieved (even in areas in which Finland has been proven to be globally number one in international comparisons). Along these lines, Raimo Väyrynen, the President of the Academy of Finland, presented three challenges (translation from Finnish to English mine):

Is technological research able to produce enough scientific results when the investments are considered?
Are the biosciences able to produce enough societal and industrial impact, considering the investments?
How do the human and social sciences motivate their societal importance when their rather modest international achievements are considered?

These challenges were first discussed by Prof. Outi Krause (TKK) and Prof. Olli Kallioniemi (VTT). Prof. Krause emphasized that the results of bibliometric analyses are always subject to some potential critisism. For instance, she asked whether the publication and citation databases cover well enough and in a balanced way all different areas of science.

In the general discussion some attention was paid to the fact that within computer science it is still commonplace to publish in conferences rather than in journals. This practice was quite heavily critized, for example, based on the idea that conference proceedings are forgotten after their publication. This is not fully true, though. Already in our laboratory there is a substantial number of conference publications that have reached the level of one hundred citations or more. It is important to remember, first of all, that for good quality computer science conferences full papers are submitted, and, secondly, they are peer reviewed.

Many conference papers are self-contained already as such: the journal paper may not add anything substantially new compared with the earlier conference proceedings version. It may be advisable to aim at publishing also a journal paper, though. However, the delays and other complications can be substantial.

The web has also changed the "scientific marketplace" at least in the area of computer science substantially. With Google, one can easily look for publications, and those papers that are not available (for free) in the web tend to have lesser number of citations. The forum of the paper does not necessarily matter as much as the content even though the credibility of the paper becomes usually checked in one way or another before it is even read. There are many ways to make papers known. Even in this blog, I tend to bring up some of the best papers that we have, and also mention good papers from elsewhere.

Another critical remark, already a long time ago raised by Academician Teuvo Kohonen, is the fact that the impact of important text books is not taken into account if only journal articles are considered.

In summary, it might even be that the computer science research community is experimenting with new ways of publishing scientific content and these practices will be applied in other areas later. This is a radically different point of view in comparison with the idea that computer scientists may sometimes be slow in adopting "normal" practices of others. Maybe we are also then able to develop better means to measure the importance of research in better ways that take into account the differences between various areas of science.

Social and technological sciences tend to have a closer connection with the societal context than, for example, natural sciences. Therefore, a Finnish historian may be highly relevant for the local audience when publishing in Finnish. It is self-evident that a researcher within, say, molecular biology has to publish in English. The research results of technologists also have to be "translated" into the language of enterprises that take the results into practical use. Much of the research funded within computer science is expected to have a practical impact within even a very short timeframe. This takes place in projects, the main focus of which may not be in scientific publishing. The basic research within computer science has, of course, to be competitive globally. The funding for true basic research is, however, smaller than for applied research. This does not mean to say that applied research is somehow scientifically less valuable. It may be that the practical contexts, rich with complex problems, give rise to even more important research results, or at least, guide the associated basic research into societally relevant directions. Without proper grounding in reality, the basic research might even be lead into some local minima based on overly restrictive background assumptions.

In general, it may not be wise to apply same success criteria for all scientific activities but give researchers a chance to excel in various ways. The motivational basis of the research community is great, I think. Guidance to approach relevance does not need to be overly explicit but it can be naturally achieced through self-organization when enough communication within the scientific community and with the surrounding society takes place.

The report published by the Academy of Finland today provides important quantitative evidence for the substantial progress achieved by Finnish scholars during the past 20 years. Suitable combinations of quantitative and qualitative research may provide in the future means for discussing the merits of different areas of science. Today's main message should be that the scientific community has a very important role in the society.

Research on Customer Practices

2006-12-19T15:36:00.000+02:00

Today, on Tuesday 19th of December, I participated a morning seminar organized by Vectia Ltd related to the KULTA project. Elizabeth Shove (Lancaster University) and Matt Wattson (Durham University) had a talk on practices and consumption. Elizabeth discussed practice oriented product design and presented Do It Yourself and digital photography as cases. She referred to elements of practice theory that she has been developing together with Prof. Mika Pantzar.

Pantzar and Shove state in their Choreography of Everyday Life that "practice is a process of integration resulting in a structured arrangement". Elements that are integrated consist of material, image and skill. In different phases links between elements are not yet formed, formed, or broken. One of the underlying issues is the idea that activities can be considered from the point of view of practices rather than, for instance, through the properties of the individuals involved. This seems to provide interesting point of view into cognitive modeling within social settings. Namely, individuals could be considered as "containers of practices". This would, of course, not mean only a top down approach. (The image on the right refers to a situation in which the practices of crosscountry skiing, golfing and family outing are merged.)

Dr. Oskar Korkman from Vectia talked on application oriented aspects of practice theory with the title "Practice Design". He presented cases on team working and cruise consumption.

In the end, Mika Pantzar provided an interesting point of view into practice theory discussing the aspect of time. He mentioned, for instance, that "practices (as performances) have identifiable co-ordinates of space and time: some are recorded in individual appointments diaries; many more evaporate, but never entirely without trace." The abstract of the presentation describes these issues more in detail.

From Molecule to Metaphor through Conceptual Spaces

2006-12-19T04:29:00.000+02:00

The previous post to our blog was written by our young researcher Matti Pöllä who participated the cognitive science anniversary event while I was still in Santa Barbara. I mentioned already earlier our paper related to semantic web in which we discussed some problems related to those semantic web methodologies that are straightforwardly based on predicate logic and related formalisms. We also discuss complementary and alternative approaches and provide some examples of such. Matti is conducting, among other things, research on artificial immune systems that he will describe later.

Prof. Peter Gärdenfors, mentioned in the previous post by Matti, has written an excellent book called Conceptual Spaces: The Geometry of Thought (MIT Press 2000). One of the strengths of the book is that there is no jump to the symbolic level of representation through some kind of magical leap like in many traditional theories of semantics and cognition. Gärdenfors' book should be read by all who are interested in the intellectual battle between proponents of symbolic and connectionist representations.

During my visit to California, I had a chance to visit Prof. Jerome A. Feldman at University of California Berkeley. Jerry Feldman is Professor of Electrical Engineering and Computer Science at UC Berkeley and a member of the Institute for Cognitive and Brain Sciences. Feldman's book From Molecule to Metaphor is an important contribution related, for example, to the question "where does the meaning come from". In From Molecule to Metaphor, Feldman proposes a theory of language and thought that treats language not as an abstract symbol system but as a human biological ability that can be studied as a function of the brain. In the preface, he mentions "virtually everyone agrees that the scientific explanation for human language and cognition will be based on our bodies, brains, and experiences." He continues by mentioning Noam Chomsky as the major exception. However, Chomsky focuses on linguistic form and Feldman deals first with meaning that is something that Chomsky has hard time to deal with.

Prof. Jerry Feldman in his office at University of
California Berkeley in November 2006

In his book, Feldman discusses, among other things, embodied language, neural computing, and computational modeling. He continues to explore learning concrete words, spatial relation words, action words, and further understanding stories. Feldman concludes that even though cognitive science can not yet address all mysteries of the mind, an enormous amount has been achieved and the outlook for continuing progress is excellent. The years of research has made him confident to even state "I have been able to assemble a picture of language and thought that must be right in its essentials."

Feldman leaves, though, some questions open. As remaining mysteries he mentions the origins of language, and the nature of subjective experience. Interestingly enough, Gärdenfors has touched upon the question of the origins of language in his research and has authored a book called "How Homo Became Sapiens: On the Evolution of Thinking". The nature of subjective experience is an issue that has multiple facets. Personally, I have approached it from the point of view of subjectivity and intersubjectivity of understanding language.

Jerry Feldman visiting our lab at Helsinki University of Technology in June 2006,

with Mathias Creutz, Jouko Salo (Tekes) and Janne Hukkinen.

Cognitive Science 50 years symposium

2006-12-11T11:12:00.000+02:00

The Cognitive Science Unit of the Helsinki University celebrated the past 50 years of cognitive science with a symposium. In a panel discussion titled "Theoretical and Historical Foundations of Cognitive Science" professors Peter Gärdenfors (Lund University), Timo Kaitaro (University of Helsinki) and Göte Nyman (University of Helsinki) discussed the paradigms of cognitive science in the past decades and their influences to the current and future directions. The session was chaired by Pauli Brattico.

The session began with a discussion of the origin of the computer metaphor of the brain -- the notion that the brain could essentially be described as a machine executing symbol manipulation tasks and algorithms to process input information as computers do. The computer metaphor and the 'cognitive psychology revolution' of the 1950s was seen as a counter reaction to behaviorism -- which in turn was a counter reaction to the earlier paradigm of German introspective psychology.

In addition to being a relatively new area of science, cognitive science has a distinct feature of having multiple, often contradictory, views on the roles of learning and adaptation (as opposed to innate structures) and statistical information processing (as opposed to symbol manipulation).

Brattico asked each of the panelists to give their take on whether symbol manipulation has a role in human cognition. All three panelists were quite sceptical about symbol manipulation taking place in the mechanisms of the brain although the brain is able to solve symbol manipulation tasks. Quoting Timo Kaitaro - 'Does symbol manipulation occur? Yes. Does it happen in the brain? No.'

Later questions asked the panelists to elaborate what the next 50 years of cognitive science could be like. Gärdenfors' answer to this was the expansion of the concept 'cognition' outside the brain organ to consider the important components of embodiment and cultural interaction. Also, Gärdenfors noted that 'the mind' might not be a good term to use anymore as a separate entity as there is no clear border between the cognition 'leaking out' of the brain into the environment.

An interesting question left without a good answer was the nature of robots and computational cognitive systems of the future. Göte Nyman's opinion was that there is no need for computational systems to be very human-like to serve humans similarly as we don't need airplanes to look like birds. Nyman also emphasized the role of a more complex top-down system model of human cognition.

But does this apply in cases where human-computer interaction would require skills which are generally thought to be possible only for humans, such as natural language processing/learning or image segmentation?

When asked to select a single central question yet to be answered in the next 50 years of cognitive science, Kaitaro presented the question of the connection of biology and cognition and whether cognition could ever be isolated from the biological realm. Gärdenfors' question was simply 'where does the meaning come from?'.

Legrady: Making Visible the Invisible

2006-12-08T15:31:00.000+02:00

Prof. George Legrady had kindly organized my visit at UC Santa Barbara. Our joint project Pockets Full of Memories, originally exhibited in 2001 in Paris, has visited of locations during the past years, also in Helsinki.

George has a number of interesting projects. Towards the very end of my visit, he described one of the latest projects called Making Visible the Invisible. It is a commission for the Seattle Central Library. The installation consists of LCD screens located on behind the librarians’ main information desk. The screens feature real-time calculated animation visualizations generated by custom designed software that conducts varies kinds of statistical analysis.

Timo Honkela: Language, Learning and Multimodal Systems

2006-12-07T07:46:00.000+02:00

On Friday, 1st of December, I gave an invited talk in the IGERT seminar series at UCSB. The talk was titled Language, Learning and Multimodal Systems: Technological, Cognitive and Philosophical Perspectives. After the talk (pdf, 5Mb), there was a lively discussion ranging from methodological issues to societal and philosophical ones.

Many broad issues in the talk were touched upon only briefly. Below there are some references for more information (some are available online; please, ask for those which are not).

Honkela, T. Von Foerster Meets Kohonen - Approaches to Artificial Intelligence, Cognitive Science and Information Systems Development. Kybernetes, 2005. 31(1/2):40-53.

Honkela, T. Self-Organizing Maps in Symbol Processing. Hybrid Neural Systems, Stefan Wermter, Ron Sun (eds.), Springer, Heidelberg, 2000, pp. 348-362.

Honkela, T., Pulkki, V. and Kohonen, T.. Contextual Relations of Words in Grimm Tales Analyzed by Self-Organizing Map. Proceedings of International Conference on Artificial Neural Networks, ICANN-95, EC2 et Cie, Paris, 1995, pp. 3-7.

Honkela, T. and Vepsäläinen, A.M. Interpreting Imprecise Expressions: Experiments with Kohonen's Self-Organizing Maps and Associative Memory. Proceedins of ICANN-91 (Artificial Neural Networks), North-Holland, vol. I, 1991, pp. 897-902.

Jäppinen, H., Honkela, T., Hyötyniemi, H. and Lehtola, A. A Multilevel Natural Language Processing Model. Nordic Journal of Linguistics 1988, 11:69-87.

Knuuttila, T. and Honkela, T. Questioning External and Internal Representation: The Case of Scientific Models. L. Magnani, and R. Dossena (eds), Computing, Philosophy, and Cognition, College Publishing, London, 2005, pp. 209-226.

Laaksonen, J. and Viitaniemi, V. Emergence of ontological relations from visual data with Self-Organizing Maps. Proceedings of SCAI 2006, Scandinavian Conference on Artificial Intelligence, pp. 31-38.

Legrady, G. and Honkela, T. Pockets Full of Memories: an interactive museum installation. Visual Communication, 1(2): 163-169, 2002.

Lindh-Knuutila, T., Honkela, T. and Lagus, K. Simulating Meaning Negotiation using Observational Language Games. Proceedings of the Third International Symposium on the Emergence and Evolution of Linguistic Communication, Rome, Italy, September, 2006.

Sjöberg, M., Viitaniemi, V., Laaksonen, J. and Honkela, T. Analysis of Semantic Information Available in an Image Collection Augmented with Auxiliary Data. Proceedings of AIAI 2006, Athens, Greece, 2006, pp. 600-608.

Meeting with Prof. Matthew Turk

2006-12-07T07:10:00.000+02:00

On Friday, 1st of December I had a chance to have lunch with Professor Matthew Turk. He is at Computer Science Department of UCSB and the Chair of Media Arts and Technology Program. His research interests include computer vision and imaging, perceptual interfaces, multimodal interaction, human-computer interaction, and gesture recognition. Prof. Turk is also the co-director of UCSB FourEyes Lab, the research focus of which is on the "four I's" of Imaging, Interaction, and Innovative Interfaces.

We discussed among other things Matthew's research on computer vision, their current project, his background in Microsoft Research, and his connections with Finland. Related to his connections with Finland, Dr. Ismo Rakkolainen from Tampere University of Technology has been visiting UCSB and that has lead into a project called
The Interactive FogScreen.

From the point of view of TKK, it was also interesting to notice a project called 3D Hand Pose Reconstruction Using ISOSOM. Based on Kohonen's Self-Organizing Map (SOM) and Tenenbaum's ISOMAP algorithm, Haiying Guan and Matthew Turk have developed the ISOmetric Self-Organizing Mapping algorithm (ISOSOM). Instead of organizing the samples in the 2D grids by Euclidian distance, it utilizes the topological graph and geometric distance of the samples' manifold to define the metric relationships between samples. They report that the experimental results show that the ISOSOM algorithm performs better than traditional image retrieval algorithms for hand pose estimation.

"Spheres of Influence"

2006-12-07T06:10:00.000+02:00

On Friday, 1st of December, Dr. Alexander Villacorta and Julie Dillemuth with colleagues presented their project called Spheres of Influence. The project builds, for instance, on the project on Multi-Domain Geovisualization of News Stories (see also a Flash demo on the Terry Schiavo case.

From the left: Timo Honkela (TKK, Finland), Alex Villacorta,
Julie Dillemuth and colleagues at UCSB.

The objective of the project is to analyze the geographical and temporal movement of information, specifically headline news stories in newspapers and relate this movement with the location of the news events in space and time. They had used geographical maps on which they position the news. They had also conducted some experiments with the self-organizing map algorithm. We discussed how it would be interesting if it was possible to analyze the news contents regarding the differences in their point of view. In order to do so, one would need also high-quality machine translation tools to ensure a global coverage. In the project, their already use some tools to translate from some European languages into English.

"Out of the Ether"

2006-12-07T05:39:00.000+02:00

Immediately after visiting the Allosphere, a project called "Out of the Ether" was introduced by Dr. John Thompson, Prof. JoAnn Kuchera-Morin and their colleagues. Their approach is to use various sensors for natural control of computer-generated musical material in the context of live performances. The group showed me a video in which a flutist was able to direct a computer-based system. The analysis system consist of components for computer vision, audio analysis, and sensor analysis. The computer vision component tracks the position of the flute, the position of the flutist's head, and the direction in which the flutist is looking (gaze tracking).

We discussed, among other things, how music and language resemble and differ from each other and how methods such as the self-organizing map could be used in the context of the project.

CNSI Allosphere at UCSB

2006-12-07T04:38:00.000+02:00

On Thursday, 30th of November, I had chance to visit a space called Allosphere at UCSB. The Allosphere is a 3-story high spherical space in which one will be able to experience fully immersive, interactive, stereoscopic and pluriphonic virtual environments. The space was in construction but one could already experience many of its features. The Allosphere is located in the California Nanosystems Institute.

Various features of the space were kindly introduced by Dr. Xavier Amatriain, Prof. JoAnn Kuchera-Morin and their colleagues. Once equipped, the CNSI Allosphere will be one of the largest immersive instruments in the world. It will contain features such as true 3D 360-degree projection of visual and aural data, and sensing and camera tracking for interactivity. The sphere will accommodate approximately fifteen people on a bridge suspended in the middle of the instrument.

During the introduction, we were standing on the bridge and following the demonstration with glasses enabling a visual 3D experience. I was also given a device with which I could "fly" in the virtual world. The demonstration consisted of the surface layers of the cortex of one of their researchers. One of the important starting points of the Allosphere had been the idea that the environment would enable researchers and artists to develop jointly new ways of human-computer interaction.

When Xavier Amatriain (blog) was describing the 3D Audio Subsystem, he mentioned about his active connections with the Laboratory of Acoustics and Audio Signal Processing in our university. The laboratory is well known, for instance, for its research on spatial sound, and analysis and synthesis of musical sounds.

Meeting with Prof. Manjunath

2006-12-02T21:24:00.000+02:00

Prof. B.S. Manjunath is the Director of the Center for Bio-image Informatics as well as the Director of Interactive Digital Multimedia at UCSB. His research interests include image/video analysis, multimedia databases and data mining, and signal/image processing for bio-informatics. I was describing related research in our laboratory including the research on Multimodal Interfaces, Content-Based Image Retrieval, Bioinformatics, and Neuroinformatics.

Prof. Manjunath has, among other things, been active in developing multimedia content description standard. As he had just a couple of hours after our meeting a lecture for his students in which he had a plan to cover the Self-Organizing Map algorithm with applications on text mining, he invited me to give a guest lecture, which I gladly accepted.

B.S. Manjunath and Timo Honkela,
with a projection of some ICA features on their face.

Prof. Manjunath's research groups are conducting very interesting research in several areas. For instance, the vision research lab conducts research on bio-molecular image databases. High resolution imaging of molecules and cells can be used to understand complex systems such as the nervous system.

Visiting UCSB, Santa Barbara

2006-12-02T20:40:00.000+02:00

After two intensive weeks around Palo Alto in Stanford, Berkeley, San Francisco, Mountain View and San Jose, a visit to University of California Santa Barbara was scheduled for this third week. UCSB is a lively university with about 18,000 students. Its researchers have received five Nobel prizes. The university has a a beautiful location, with the ocean on one side and mountains on the other.

UCSB Lagoon as seen from the campus.

I was invited to UCSB by Prof. George Legrady who is a co-principal in the National Science Foundation IGERT Interactive Multimedia Research Program. With George, I had with my colleagues from the UIAH Media Lab a joint project in which we created an interactive museum installation called Pockets Full of Memories for the Pompidou Center in Paris. The installation was exhibited from 10 April to 3 September 2001.

During the visit, I have had a chance to meet many researchers with very interesting projects. I will later describe these in some detail.

Talk in Berkeley, visiting BISC

2006-11-30T07:59:00.000+02:00

In the afternoon, after visiting San Jose and Stanford, it was time to visit Berkeley. Dr. Masoud Nikravesh of BISC had invited me to give a presentation in their seminar series. The title of the talk was "Inherent fuzziness of language: cognitive, philosophical and computational aspects".

A starting point for the talk was the argument Prof. Lotfi Zadeh has presented for the idea that the brain has a crucial ability to manipulate perceptions. Manipulation of perceptions plays a key role in human recognition, decision and execution processes. Zadeh has also pointed out that measurements are crisp whereas perceptions are fuzzy. In general, the relationship between perceptions and their linguistic descriptions is not as straightforward as often suggested, for instance, by many logicians.

Prof. Jonathan Evans has pointed out that reasoning is highly contextualized by relevant prior knowledge and belief. He has also referred to the dual process theories of reasoning that make a division between a heuristic system and an analytical system. The heuristic system has evolved early, it is shared with animals, it is rapid and parallel, has high capacity and is pragmatic. According to Evans, the analytic system is conscious and it has evolved late in the evolution.

In the presentation, I was discussing how the division into heuristic and analytic system supports Zadeh's arguments. I also showed how some issues in philosophy of language can become resolved when fuzzy systems thinking is applied.

The discussion after the presentation was lively. One active discussant was Henry Story who has commented some of the items in his blog and provided some ideas on how to apply fuzzy sets within semantic web. Those who are interested in the semantic web might wish to check our recent paper: 'Describing Rich Content: Future Directions for the Semantic Web' by Timo Honkela and Matti Pöllä (appeared in: New Developments in Artificial Intelligence and Semantic Web, Proceedings of STeP 2006, pp. 143-148).

Tekes San Jose and SCANCOR

2006-11-28T08:55:00.000+02:00

On Tuesday 21st of November, I rented a car as we needed to travel around the Bay area intensively. Next week I would also travel to University of California Santa Barbara which is very difficult to reach by public transportation. (There is a small airport in Santa Barbara but the flights there appeared to be expensive.)

With Sari Stenfors we first visited Tekes office in San Jose. The head of the office, Reijo Kangas told us, among other things, that Tekes is going to move to a new office space in Santa Clara in January 2007. The move is based on the formation of FinNode, an innovation centre that will serve Finnish and American companies and researchers. The innovation centre has been established by Finpro, Sitra, the Academy of Finland, Tekes, and VTT.

From San Jose I headed quickly back to Palo Alto and Stanford University where I had a meeting with Prof. Walter W. Powell who is the director of SCANCOR. Woody, as he is often called, gave interesting examples of their research. One paper that he discussed was "Roads to Institutionalization: The Remaking of Boundaries Between Public and Private Science" by Jeannette A. Colyvas and Walter W. Powell (Research in Organizational Behavior, vol. 27, 305-353). In the paper they illustrate how archival materials may be systematically assessed to capture notable changes in organizational practices and categories.

Visiting SFO and Google

2006-11-27T20:00:00.000+02:00

On Monday 20th of November, I had first a short company visit in San Francisco. The travel by train from Palo Alto to the city was an interesting experience. The travel was convenient but the user interface of a ticket vending machine was seemingly slightly too complicated for me. In the end, however, everything went well.

In the afternoon, I continued discussions with Dr. Peter Norvig. Sari Stenfors kindly drove me to 1600 Amphitheatre Parkway in Mountain View where Google headquarters are located.

Multidisciplinary view on consumer models

2006-11-27T15:42:00.000+02:00

After listening to Sherry Turkle's presentation, we headed quickly towards the Wallenberg hall at Stanford University in which we gave a presentation on the Kulta project, funded by Tekes and collaborators. The title of the presentation was "Multidisciplinary view of consumer models: Hobbyism and playful strategies meet adaptive models". The objective of our project is to develop models and methods that can be used in understanding, conceptualizing and anticipating the changing needs of consumers. Our study takes a constructivist perspective: we study how meanings are constructed in the practical and interactive use of models.

Tanja Kotro was presenting the point of view of hobbyism based on her dissertation. Hobbyism refers to employees’ passion for a hobby (for instance sports) and communities related to the hobby as an important reference for the product development team members for understanding users.

Sari Stenfors' presentation was based on her research on
playful strategies. Most strategy tools are namely used to improve organizational efficiency and only some are used to explore and innovate. Both tools that support efficiency and toys that improve creativity are needed in organization’s tool-box to maintain sustainability.

Timo Honkela was considering how computational tools and methods could support analytical, reflective and creative processes in organizations. The constructivist perspective emphasizes the need to have open ended tools in which some predefined frameworks do not determine the results. In order to deal with change, the models need to be adaptive: they alter themselves as new information on the consumers and related phenomena is available.

Sherry Turkle: Evocative Objects

2006-11-24T01:50:00.000+02:00

On Friday, 17th of November we had a chance to hear Sherry Turkle's presentation on "New Complicities for Companionship: A Nascent Robotics Culture". Turkle is the professor of Social Studies of Science and Technology at MIT. She is the author of several well known books including "Life on the Screen: Identity in the Age of the Internet" (1995). The talk was a part of Stanford's Seminar on Science, Technology, and Society.

In her presentation, Turkle emphasized that technologies are never just tools but that they are evocative objects: things we think with (which is the name of a book that will be published by MIT Press this year). She discussed relational artifacts that present themselves as having a mind of its own. She explained that there is a movement in computation from users projecting themselves on the screen to computational entities (agents/robots) becoming relational companions. A point that Turkle underlined in her presentation was that the art and science of the construction of relational artifacts is related to understanding human psychology and human vulnerability. Interactions with computational objects moves more and more towards the psychology of engagement and object relations.

The presentation led into a lively discussion. Turkle correctly pointed out that there is a tendency to present AI systems as having some human like qualities and skills which is, to a large degree, only an illusion. I presented her a question about the possibility of having in the future a system that would be able to recognize some real psychological patterns. For instance, in the future we could have an agent that would follow a conversation and could suggest something like: "You seem to be unnecessarily harsh towards your spouse which might stem from the fact that he resembles your parent that was unfair towards you in your childhood." Turkle answered briefly that that needs to be thought about. The shortness of her answer made me wonder if she thought that I did not understand the core point of her presentation. However, I think there is a chance to develop systems that could to some degree model psychological phenomena to such level of detail that they could help people to gain increased understanding of themselves.

Weizenbaum's Eliza and all its successors only react to some keywords in an illusionary therapeutic conversation. An alternative would be to develop systems that would collect their "understanding" of psychological phenomena from a large number of interactions among people over an extended period of time. A modest attempt to this direction was our paper "Emotional Disorders in Autonomous Agents?" in which we outlined a minimalistic model of anxiety, depression and mania (Hyvärinen and Honkela, Proceedings of ECAL'99, European Conference on Artificial Life, Springer, 1999, pp. 350-354). The paper itself is rather sketchy but I think the core ideas are still valid.

Social AI revisited

2006-11-20T06:08:00.000+02:00

Based on my presentation on Wednesday at Nokia RC Palo Alto, Joe McCarthy wrote an excellent summary.

Social Artificial Intelligence

2006-11-16T06:54:00.000+02:00

Wednesday in the Bay Area was sunny and interesting visits continued. In the morning, I bicycled from the hotel to the Semlab, the Computational Semantics Laboratory at the Center for the Study of Language and Information at Stanford University.

Dr. Elizabeth Owen Bratt was hosting the visit and had organized a number of interesting demos shown by her colleagues. The interesting demos included dialogue systems being developed in the lab as well as a system that analyzes meeting transcripts.

I had also a chance to discuss with Prof. Keith Devlin who is the director of the Center for the Study of Language and Information. Devlin was telling about the increasing connections between Finnish researchers and Stanford University.

In the afternoon, we visited the newly opened Nokia Research Center Palo Alto. The visit was hosted by Dr. Joe McCarthy (check Joe's blog). I gave a presentation on "Social Artificial Intelligence" expressing that not only are the needs and priorities different among people but also their knowledge and language are unique. One cannot assume that expressions would mean the same for all those who receive the message. Social AI technologies are needed to model those differences. The presentatation was followed by a lively discussion.

Joe McCarthy, Timo Honkela and Sari Stenfors at
Nokia Research Center Palo Alto.

Berkeley: meetings with Feldman, Hautamäki, Nikravesh and Norvig

2006-11-15T09:01:00.000+02:00

Today on Tuesday, 14th of November, Sari Stenfors took Tanja Kotro and myself to Berkeley. We first had a meeting with Prof. Jerome Feldman who provided information on ICSI. Dr. Antti Hautamäki from Sitra accompanied us. He is visiting Berkeley for two years researching innovation networks.

Jerry Feldman invited us to participate a seminar of the speech group. Malcolm Slaney from Yahoo! Research gave a presentation on issues such as timbre analysis, analysis of song dissimilarity and image relevance. The issue on image relevance reminded of the successful work in our lab on PicSOM, a content-based information browsing and retrieval system.

With Jerry Feldman we also discussed his latest book "From Molecule to Metaphor: A Neural Theory of Language". It emphasizes the important point of view that language should be considered not as an abstract symbol system but as a human biological ability that can be studied as a function of the brain.

Later the same day, we walked through the beautiful campus to Soda Hall in which BISC, Berkeley Initiative on Soft Computing is located. Dr. Peter Norvig, Director of Research, Google, gave a presentation entitled "AI as the Future of Search". The seminar was chaired by Dr. Masoud Nikravesh from BISC. Peter Norvig was discussing issues such as the use of clustering algorithms, spelling, query refinement, statistical machine translation and learning from text or "machine reading". As we know, many results based on the vast data gathered by Google are convincing. Peter Norvig was also providing information on how Google collaborates with and supports the research community.

In the end of the day, Dr. Nikravesh kindly invited us to join a dinner in a Persian restaurant.

Masoud Nikravesh, Timo Honkela and Peter Norvig having a dinner.

The discussions ranged from science and technology to societal issues. We agreed that tools such as Google with the infrastructure of internet provide important means for positive future progress.

Visiting Jim March Seminar

2006-11-14T18:30:00.000+02:00

We visited Jim March Seminar on Monday, 13th of November. James G. March, Emeritus Professor at Stanford is best known for his research on organizations and organizational decision making. He conducts research, writes, and teaches on decision making, risk taking, information processing, and learning in organizations. In an issue of Harvard Business Review, it was mentioned that "Jim March is to organization theory what Miles Davis is to jazz".

In the seminar, we followed Sampo Tukiainen's presentation on "Cross-National Learning in Two Consecutive Finnish-Polish Power Plant Projects". An interesting and lively discussion was followed by the presentation. Jim March pointed out that people might easily perceive that problems in cross-national projects stems from nationality even though there may be important background issues related to the power structures. We discussed, for instance, how the labeling and classification may influence the perception of phenomena in general.

Later the same day, we visited SCANCOR offices and Wallenberg hall in which we will give a presentation on Friday.

In the evening, we met Don Steiny who is the president of the Institute for Social Network Analysis of the Economy. At Stanford he is a member of the Silicon Valley Network Analysis Project.

KULTA project visits California

2006-11-14T18:16:00.000+02:00

Tanja Kotro (National Consumer Research Center) and Timo Honkela visit California to give a number of presentations, to participate seminars and to meet people from different universities an companies. The first part of the visit is coordinated by Sari Stenfors from Stanford University. During the third week of the stay, Timo will visit University of California Santa Barbara, hosted by Prof. George Legrady.

SCAI 2006 and STeP 2006 Conferences

2006-10-30T12:05:00.000+02:00

Our laboratory was actively involved in two parallel conferences: SCAI 2006 (Scandinavian Conference on Articificial Intelligence) and the corresponding Finnish conference STeP 2006. Tapani Raiko as the chair of Finnish AI Society was the chair of the organizing committee of SCAI and Timo Honkela was the program chair. Information on both SCAI and STeP is extensively available in the web, including the proceedings.

Our group had several papers in the conferences:

"Analysis of Interdisciplinary Text Corpora" by Matti Pöllä, Timo Honkela, Henrik Bruun and Ann Russell

"Emergence of multilingual representations by independent component analysis using parallel corpora" by Jaakko J. Väyrynen and Tiina Lindh-Knuutila

"Describing Rich Content: Future Directions for the Semantic Web" by Timo Honkela and Matti Pöllä

Simulating processes of language emergence, communication and agent modeling" by Timo Honkela, Ville Könönen, Tiina Lindh-Knuutila, and Mari-Sanna Paukkeri

KULTA:Modeling Changing Needs of Consumers

2006-10-30T11:47:00.000+02:00

Our group coordinates a project called KULTA. The acronym comes from Finnish words that refer to consumers and needs. The acronym is a word in Finnish that means two things: gold and darling. Due to this nice ambiguity we have decided not to translate the acronym.

The project develops methods that can be used in understanding, conceptualizing and anticipating the changing needs of consumers. The conceptual models of the practice theory are applied to analyze changes in the consumer society.

These models are applied in the context of developing the business models of different kinds of companies. The data gained in this research and developing process are then analyzed and modeled using methods that are developed and applied in our laboratory.

The project was planned by Prof. Mika Pantzar (National Consumer Research Center), Dr. Tarja Knuuttila University of Helsinki, Department of Philosophy) and Dr. Timo Honkela who serves as the responsible director of the project. Dr. Tanja Kotro from the National Consumer Research Center is the project manager of KULTA.

A central motivation from our group's point of view is the possibility to deal with complex phenomena and data that deal with both individual and social levels of reality.

MedIEQ: Quality Labeling of Medical Web Sites

2006-10-30T11:30:00.000+02:00

Our group is involved in a EC funded project called MedIEQ. We were hosting a meeting of the project from 23rd to 24th of October here in Espoo. MedIEQ will deliver tools that crawl the Web to locate medical web sites in seven different European languages (Catalan, Czech, English, Finnish, German, Greek, and Spanish) in order to verify their content using a set of machine readable quality criteria. MedIEQ tools will monitor already labelled medical sites alerting labelling experts in case the sites' content is updated against the quality criteria, thus facilitating the work of medical quality labelling agencies (Web Medica Acreditada, Barcelona, and Agency for Quality in Medicine, Berlin). The project is coordinated by a Greek research center called Demokritos.

Our group concentrates on the use of adaptive and statistical methods to deal with "difficult content". By difficult, one refers to such content that is difficult to analyze using structured, e.g., rule-based method. One example is the task to detect whether a site is intended to be read by laypersons or medical professionals.

More information on the project and our objectives is provided by our site manager Matti Pöllä.

Cognitive Systems Research at Helsinki University of Technology

2006-10-30T09:22:00.000+02:00

This forum is centered around the research conducted at the Computational Cognitive Systems research group at the Adaptive Informatics Research Center of Helsinki University of Technology (TKK). This research is closely related to the other groups in the Adaptive Informatics Research Center, especially Multimodal Interfaces. There is also relevant related research conducted in other laboratories at TKK, for instance at the Laboratory of Computational Engineering (Computational Neuroscience) and the Brain Research Unit of the Low Temperature Laboratory. At University of Helsinki, related research is conducted, e.g., at the Neuroinformatics Group at HIIT Basic Research Unit.

Cognitive Systems

2006-10-30T09:11:00.000+02:00

This is a forum related to cognitive systems. Cognitive systems combine perception, action, reasoning, learning and communication. This area of computer-based modeling draws upon biological, cognitive and social system approaches to understanding cognition.