• seminar

seminar

SEMINAR

This talk discusses some of the challenges we face when doing text analysis for Big Data. The talk gives a brief overview over the different types of applications we work with at Gavagai, and some of the technologies we use to deal with the Big Data challenge. We will in particular focus on distributional semantics, and the particular processing framework we favor at Gavagai – Random Indexing.

Date: 2015-03-26 10:30 - 12:00

Location: L308, Lennart Torstenssonsgatan 8

Permalink

SEMINAR

I will present the tools and frameworks we have built for easy and flexible manipulation and use of linguistic tree structures. A data format that makes no linguistic assumptions (e.g. phrased vs. dependencies) and tools to build, manipulate, search and visualise the trees. I will use the example of adding information about multiword expressions into The Prague Dependency Treebank as a short case study to illustrate the use of the tools.

Date: 2015-03-05 10:30 - 12:00

Location: L308, Lennart Torstenssonsgatan 8

Permalink

SEMINAR

In this talk, I propose the usage of graph databases as a means of facilitating the integration of multilingual lexical and constructional resources in the FrameNet domain. Graphs have the advantage of modeling relations between objects in the database in a more direct and intuitive way, without the need to create unnecessary tables, indexes and relations aimed exclusively at connecting, for example, language specific representations of a conceptual system (frame), or entities belonging to different classes, such as frames and constructions. Additionally, graph databases are easily linkable to other kinds of external data, such as ontologies and open data sources.

Tiago Timponi Torrent
Federal University of Juiz de Fora, Brasil, FrameNet Brasil

Date: 2015-02-09 13:15 - 15:00

Location: L308, Lennart Torstenssonsgatan 8

Permalink

SEMINAR

Date: 2015-01-22 10:30 - 12:00

Location: L308, Lennart Torstenssonsgatan 8

Permalink

SEMINAR

Yvonne Adesam, Gerlof Bouma and Richard Johansson: Defining the Eukalyptus forest – the Koala treebank of Swedish.

Richard Johansson and Luis Nieto Piña: Combining Relational and Distributional Knowledge for Word Sense Disambiguation.

Inari Listenmaa and Francis M. Tyers: Automatic Conversion of Colloquial Finnish to Standard Finnish.

Date: 2015-05-07 10:30 - 12:00

Location: L308, Lennart Torstenssonsgatan 8

Permalink

SEMINAR

The goal of this talk is twofold; first, to give an overview of the research concerning computer-assisted pronunciation training over the past decade or so. We will look into the most important features of such systems, the current research approaches followed and have also a glimpse of some of the most famous commercial systems. Second, the talk will be focused on automatic pronunciation error detection and the work that I have been carried out during the last four years. In short, I am going to present the perceptually motivated pronunciation error detection algorithms, which are motivated by the observation that almost all native speakers perceive, relatively easily, the acoustic characteristics of their own language when it is produced by speakers of the language. The methods are the result of combination of various research areas such as speech processing, auditory perception, phonetics, etc. The findings are promising and can help us understand how humans distinguish between various speech sounds. At the end of the talk, we will discuss how an ideal pronunciation error detection system could be benefited by different approaches to be used as a reference to determine current challenges and possible next steps in research efforts.

Date: 2015-02-19 10:30 - 12:00

Location: L308, Lennart Torstenssonsgatan 8

Permalink

SEMINAR

[CANCELLED!]

In this talk we will present a new Coreference Resolution system for Swedish, based on supervised machine learning methods (C4.5 and Latent Structured Perceptron) trained on the SUC-core dataset. Our method improves on state-of-the-art results for the data, achieving an average F1-score of 50.9 using the standard CoNLL 2012 metrics. We will discuss in more details the choice of features for training the algorithms as well as possible ways for improvement.

Fredrik Axelsson (Findwise) Svetoslav Marinov (Findwise) Fredrik Johansson (Chalmers) Devdatt Dubhashi (Chalmers)

Date: 2014-12-18 10:30 - 12:00

Location: L308, Lennart Torstenssonsgatan 8

Permalink

SEMINAR

We present a method of extracting functional semantic knowledge from corpora of descriptions of visual scenes. Such knowledge is required for interpretation and generation of spatial descriptions in tasks such as visual search. In the collostruction analysis (Dobnik and Kelleher, 2013) we estimate a log-likelihood ratio between a preposition and a target-landmark tuple which measures the strength of association between them. To determine their function-geometry bias we estimate the entropy of their target-landmark tokens. In (Dobnik and Kelleher, 2014) we identify semantic classes of target and landmark objects related by each preposition by abstracting over WordNet taxonomy. The inclusion of such knowledge in visual search should equip robots with a better, more human-like spatial cognition.

Simon Dobnik, Department of Philosophy, Linguistics, and Theory of Science, University of Gothenburg (joint work with John D. Kelleher, Dublin Institute of Technology, Ireland)

Dobnik, S. and Kelleher, J. (2014). Exploration of functional semantics of prepositions from corpora of descriptions of visual scenes. In Proceedings of the Third Workshop on Vision and Language, pages 33–37, Dublin, Ireland. Dublin City University and the Association for Computational Linguistics.

Dobnik, S. and Kelleher, J. D. (2013). Towards an automatic identification of functional and geometric spatial prepositions. In Proceedings of PRE-CogSsci 2013: Production of referring expressions - bridging the gap between cognitive and computational approaches to reference, pages 1–6, Berlin, Germany.

Date: 2015-01-29 10:30 - 12:00

Location: L307

Permalink

SEMINAR

Using an ostensible taste testing experiment, we investigated the effects of intrusive noises on people interacting in a kitchen setting. During the tastings, for half of the pairs, the experimenter started several noise making appliances (a kettle, a hairdryer, a hoover, a juicer, a strimmer and a reversing alarm) at random times – either predictably (i.e. making it obvious, by walking up to the appliances and pressing the on switch), or unpredictably – by remote control. To analyse the effects this intervention had on the interaction we analysed people’s facial expressions for how surprised they looked and how happy they looked when they heard the sounds, using automatic facial recognition software (SHORE). The results show three different effects of context on their responses. First, congruent noise sources i.e. those which are expected in context (e.g. a kettle) cause less disruption than incongruent noise sources (e.g. a strimmer). Second, noises that can be attributed a clear apparent cause are less disruptive than those that are apparently random. Third, there is an effect of social context in that if the cause of the noise is a person, the form of people’s responses depends strongly on the way in which the disruptive noise is introduced by that person.

Dr Christine Howes
Postdoctoral Researcher
http://flov.gu.se/english/contact/staff/chris-howes

Date: 2015-03-12 10:30 - 12:00

Location: L308, Lennart Torstenssonsgatan 8

Permalink

SEMINAR

The fundamental claim of this talk is that attention -- both visual and linguistic -- is an important overarching semantic category structuring visually situated dialogue. Based on this it is argued that computer systems attempting to model the evolving context of a visually situated dialogue should integrate models of visual and linguistic attention within their natural language processing framework. The talk will present a dialogue framework that integrates attention into the representations of the evoling context. Within this framework a model of perceptual attention underpins two of the core subtasks of NLP: reference resolution and the generation of referring expressions. There are three main components within the framework: a model of synthetic visual attention, a set of computational models of spatial term semantics, and a discourse model integrating visual and linguistic information.

John D. Kelleher Applied Intelligence Research Center, School of Computing, Dublin Institute of Technology

http://www.comp.dit.ie/jkelleher/

Date: 2014-12-11 10:30 - 12:00

Location: L308, Lennart Torstenssonsgatan 8

Permalink

X
Loading