Cognitive Science Colloquia

Cognitive Science Colloquia are held* throughout the year and consist of seminars/talks given by invited speakers from various universities, as well as seminars by the METU staff or our own PhD students. The Colloquia aim to share information on cognitive science research, as well as promote academic research and collaboration. The seminars are open to anyone interested and provide an opportunity for our students to keep up with the recent research conducted in the field, allowing them to observe how research topics develop through academic exchange.

* Unless it is announced otherwise, seminars take place every Wednesday at 12:40, in the Informatics Institute (room II-03), METU.

For past years' colloquia, please visit here.

NEXT IN THE SERIES     >     Probabilistic Models that Combine Knowledge and Data for Learning and Decision Making

by Barbaros Yet (Industrial Engineering, Hacettepe Uni.)
on 21st of May, Tuesday, 12:00
in METU Informatics Institute, Ural Akbulut Lecture Hall

Abstract Despite recent promising results of purely data-driven machine learning approaches, their success were mainly focussed on certain domains where large and accurately labelled datasets can be prepared. Moreover, these knowledge-independent approaches often face challenges in generalizing their inferences to different problems with similar properties. Human experts, however, are often able to discover key relationships from noisy and smaller datasets, and successfully apply them to similar problems. We focus on a probabilistic modelling approach that combines knowledge and data for learning and reasoning. Our approach uses Bayesian probabilistic graphical models that represent the current theory and understanding of the problem in terms of a causal structure, and analyses available data based on this theory. These models also enable us to make complex probabilistic and causal inferences about the problem domain. This presentation will focus on our methodological work on such models, and illustrate it with case studies of legal, clinical and development decisions.

This Year (2018-2019)

Spring Semester

Open Science and Replicability
by Mine Mısırlısoy (Psychology, METU)
on 15th of May

Abstract Science is about testing hypotheses, in an effort to create general rules from results that can be replicated later; and using those rules to make further testable predictions. Open science is the collective action of making science more accessible and more transparent for all; including both professionals and amateurs. The main goal of the so called Open Science Movement is to make science more robust and replicable. In this talk, I will talk about the recent Replication Crisis, focusing specifically on the problematic practices that gave rise to this methods crisis in Psychological Science and Neuroscience. After a brief historical overview, I will talk about the benchmark events that gave rise to the replication crisis, and later the open science movement. I will put forward the Questionable Research Practices that, unfortunately, most scientists use either due to lack of knowledge and/or motivation, and give information on how to detect QRPs as users of knowledge and how to avoid them as creators of knowledge. Lastly, I will talk about ways to improve (psychological) science; such as, preregistration, registered replications, open methods, and open data.

Dopaminergic modulation of cross-modal attention in Drosophila
by Özlem Çevik (Psychology, TOBB Uni.)
on 8th of May

Abstract Translational models of attention and its disorders have been debated on the grounds that non-human animals cannot be fully validated as models of higher order cognitive abilities or disabilities of humans. Nonetheless, in recent years, there has been a new research trend to study attention and its disorders in animals with simpler brains, including fish and insects. Grounded on mounting evidence for the homology of molecular mechanisms as well as the similarity of underlying neural circuit principles, this new approach defines attention on the basis of its functional properties. Although researchers have long depicted elaborate experimental models of gain modulation in the peripheral nervous system of insects, they had a concessive reluctance to present peripheral gain modulation as a model of centrally-gated vertebrate attention. Behavioral tests of insect attention that have so far been used mainly depend on selective responding to a subset of stimuli or stimulus properties that impact the same sensory modality. We developed a behavioural test that involves competition between different sensory modalities each of which is associated with a different response system, such that behavioural choice potentially engages the activity of central neuropils involved in cross-modal stimulus association and/or action selection.

Speed Estimation as a Multistage and a Multisensory Process
by Hulusi Kafalıgönul (UMRAM, Bilkent Uni.)
on 24th of April

Abstract A growing body of evidence shows that the information provided by other modalities can affect visual motion perception. For instance, it is well known that the timing of brief static sounds alters different aspects of visual motion such as speed estimation. Yet, little is known about the principles and cortical processes underlying these effects of auditory timing. In this seminar talk, I will mainly talk about our progress in understanding the neural substrates involved in such multisensory nature of motion and speed estimation. In particular, I will present our recent findings on an audiovisual motion paradigm combined with a speed discrimination task. At the final part, I will discuss these findings from a broader perspective and provide their specific implications for audiovisual temporal processing and multisensory integration.

Catastrophic Risk Modeling and its Pricing in Insurance
by Sevtap Kestel (Actuarial Sciences, METU)
on 17th of April

Abstract Increasing prevalence of natural catastrophes due to population growth and global climate change propelled the insurance market to create sophisticated models and financial devices to offer protection against calamitous events such hurricanes, droughts, landslides, earthquakes and others. Uncertainty in the occurrence of these events and the financial losses that they may cause requires quantification of the risk and its financial impact. It is crucial to understand how catastrophe models are developed with repect to their underlying factors. Focused on earthquake, the techniques which enables us to measure the risk and its pricing in insurance and reinsurance markets are presented with examples.

The truth is not always that easy to remember: The effects of generating lies on memory predictions and actual memory performance
by Miri Besken (Psychology, Bilkent Uni.)
on 10th of April

Abstract Generating lies takes a longer time, requires more effort and is frequently considered to be less fluent, as compared to telling the truth. Clearly, more effortful and less fluent tasks typically increase memory performance and sometimes decrease memory predictions, but this claim has not been investigated within the realm of lie generation paradigms. In this talk, I will discuss the effects of generating lies on memory predictions and actual memory performance through three separate lie-generation manipulations. These manipulations involve lying about general knowledge questions, lying about personal semantic information and lying about episodic incidents. This line of research reveals that generating lies also increases actual memory performance and decreases one’s memory predictions about those items, just like other effortful and disfluent manipulations. However, the type of task used, the level of motivation to lie and the type of memory test used may also moderate the effects of lying on both memory predictions and actual memory performance.

Modeling Hereditary and Behavioral Patterns of Facial Expressions
by Hamdi Dibeklioğlu (Computer Engineering, Bilkent Uni.)
on 3rd of April

Abstract Over the last two decades, computational analysis of facial expressions has been a very active area of research. Today, following the recent dramatic advances in the fields of machine learning and computer vision, we are able to model subtle and complex patterns of facial responses in a reliable manner. Our findings indicate that facial expressions do not only reveal instant emotional state of individuals but also display hereditary and behavioral characteristics. This talk will focus on deep architectures to model such patterns for various tasks such as recognizing preferences, assessing psychopathology, and predicting how your future child's expressions look like.

I feel cared for therefore I am well: The influence of close relationships on health and happiness
by Emre Selçuk (Psychology, METU)
on 20th of March

Abstract It has long been demonstrated that social relationships, especially those of attachment and mating variety, affects health and well-being. But a key question remains: How? Recent work in our lab aims to address this question using multiple methods including experimental, longitudinal, observational, and daily experience approaches and focusing on processes at multiple levels of analysis ranging from behavior to affect to physiology. The present talk will showcase studies from this research program documenting how partner responsiveness—the extent to which individuals believe that their partner cares for, understands, and appreciates them—is linked to mechanisms leading to health and well-being in the long run. I will discuss evidence that partner responsiveness buffers against stress reactivity, moderates the extent to which individuals benefit from social support, promotes health behaviors, and (possibly) alters long-term functioning of stress regulation systems, all of which influence later physical health and psychological well-being. Altogether the findings offer a window into the mechanisms of how relationships get to affect personal well-being.

Anlam-Biçim Eşleme Problemi Açısından Türkçede Çok Sözcüklü İfadeler
by Özkan Aslan (Cognitive Science, METU)
on 13th of March

Abstract Çok Sözcüklü İfadelerin (ÇSİ) tanınması problemi, hem Varlık İsmi Tanıma, hem de mecazi kullanım tanıma problemlerinden farklıdır. Doğal Dil İşleme Alanında ÇSİ'lerle ilgili genel anlayış, bunların işleme sırasında ayrıştırma yapmadan kalıp halinde alınması gereken ifadeler olduklarıdır. Ancak ÇSİ'ler özel deyim olarak çok farklı üretkenlikte ve sözdizimsel karmaşıklıkta karşımıza çıkabilmektedir. Biz ÇSİ-deyimlerin, aslında yakından bakıldığında istisnai değil bileşimsel ayrıştırmanın özel kategorilerle yapılmış hali olduklarını iddia ediyoruz. Bunu, grameri sözcük bilgisine indirgeyen kuramlarla modellediğimizde, uzun varlık isimlerinden ve mecazdan farklı kullanımlarının da öğrenilebilir olduğunu düşünmekteyiz. Bu bağlamda Combinatory Categorial Grammar (CCG) kuramını ve CCG modelleme, gramer öğrenme ve ayrıştırma yazılımlarını kullanıyoruz.

Academic Writing: An Overview
by Deniz Saydam & Esin Savul (Academic Writing Center, METU)
on 6th of March

Abstract This presentation aims to give an overview of the basic structure of academic writing, the ways to avoid plagiarism, and the citation methods (quoting, paraphrasing and summarizing) and to make some general suggestions on academic writing.

Expanding Existing Wordnets to New Languages
by Gönenç Ercan (Informatics, Hacettepe Uni.)
on 27th of February

Abstract Wordnet is a resource providing lexical knowledge in combination with a network of word meanings and their relationships. Word senses are organized using relationship types motivated by early psycholinguistic studies. Wordnet as a natural language resource accommodates a large body of research, yet it is still not available for many languages. Many projects have been and are still being proposed to build Wordnets for different languages. Even a translation based approach is reported to be an arduous work requiring experts to spend days on the task. As more resources including computational power became available, automated Wordnet construction methods emerged. While most methods combine several resources to translate Princeton Wordnet to a target language, we hypothesize that when words and their translations are observed for a large number of languages, it is possible to construct Wordnets for all the languages in the translation graph without requiring any additional resource. In this talk, we will discuss our work on building a multilingual Wordnet and the built resource which includes Wordnets for more than 40 languages.

Basic Tenets of Cognitive Science
by Ceyhan Temürcü (Cognitive Science, METU)
on 20th of February

Abstract N/A

Fall Semester

Type of expression and type of reference in a layered DRT
by Ceyhan Temürcü (Cognitive Science, METU)
on 4th of January

Abstract In this talk I will describe A-DRT, which is a multi-layered DRT and use it (i) to distinguish type of reference (explicit vs. implicit reference) from type of expression (presupposition vs. new information) (ii) to evaluate approaches that identify semantics with truth-conditional interpretation, and (iii) contribute to our understanding of presupposition and inference.

Multi-word expressions, meaning, and compositionality
by Cem Bozşahin (Cognitive Science, METU)
on 21st of December

Abstract Multi-word expressions, verb-particle constructions, idiomatically combining phrases, and phrasal idioms have something in common: not all of their elements contribute to the argument structure of the predicate implicated by the expression. Radically lexicalized theories of grammar that avoid string-, term-, logical form-, and tree-writing, and categorial grammars that avoid wrap operation, make predictions about the categories involved in verb-particles and phrasal idioms. They may require singleton types, which can only substitute for one value, not just for one kind of value. These types are asymmetric: they can be arguments only. They also narrowly constrain the kind of semantic value that can correspond to such syntactic categories. Idiomatically combining phrases do not subcategorize for singleton types, and they exploit another locally computable and compositional property of a correspondence, that every syntactic expression can project its head word. Such MWEs can be seen as empirically realized categorial possibilities rather than lacuna in a theory of lexicalizable syntactic categories.

Lexical Inference in Turkish Emphatic Reduplication
by Özkan Kılıç (Cognitive Science, METU)
on 7th of December

Abstract Turkish Emphatic Reduplication (TER) occurs in adjectives and adverbs to accentuate their meanings (e.g., beyaz --> bembeyaz). TER is viewed as a purely phonological operation, and often used as an example in Optimality Theory and Morphological Doubling Theory. Despite the theory and the phonological rules for the selection of linker type in TER, variations do occur. For example, ci-r-il-ciplak, ci-s-ciplak, ci-r-ciplak, ci-p-ciplak and ci-m-ciplak will show up in web searches. In this study, 42 university students are asked to reduplicate 19 adjectives, 19 nouns, 19 verbs and 19 pseudowords of the from CVCCVC. Their answers, eye-tracking data and reaction times are analyzed. The results indicate that there is a lexical interference in TER.

Visual perception of actions: An interdisciplinary work between cognitive neuroscience and social robotics
by Burcu Ayşen Ürgen (Psychology, Bilkent University)
on 23rd of November

Abstract Successfully recognizing the actions of others is of utmost importance for the survival of many species. For humans, action perception is considered to support important higher order social skills, such as communication, intention understanding and empathy, some of which may be uniquely human. Over the last two decades, neurophysiological and neuroimaging studies in primates have identified a network of regions in occipito-temporal, parietal and premotor cortex that are associated with perception of actions, known as the Action Observation Network. Despite a growing body of literature, the functional properties of this network remain largely unknown. We take a multi-modal, interdisciplinary, and computational approach to characterize the functional properties of this network in humans. To this end, we 1) collaborated with a robotics lab to vary various aspects of actions including visual appearance and movement kinematics of the agents, 2) used a wide range of brain measurement modalities (fMRI and EEG) together with state-of-the-art analytical techniques including representational similarity analysis, computer vision, and clustering to investigate the neural processing while human subjects performed action perception tasks. While our findings improve our understanding of the Action Observation Network, the interdisciplinary work with robotics also allows us to address questions regarding human factors in artificial agent design in social robotics and human-robot interaction such as uncanny valley, which is concerned with what kind of agents we should design so that humans can easily accept them as social partners.

Introducing METU Language and Cognitive Development Lab
by Duygu Özge (Cognitive Science, METU)
on 16th of November

Abstract I will give an overview of some major research questions in the field of language/cognitive development and their significance for cognitive science, followed by a summary of research projects we run in our emerging METU Language and Cognitive Development Lab.

Object detection through search with a foveated visual system
by Emre Akbaş (Computer Engineering, METU)
on 2nd of November

Abstract In this talk, I will present a foveated object detector (FOD) as a biologically-inspired alternative to the sliding window (SW) approach which is the dominant method of search in computer vision object detection. Similar to the human visual system, the FOD has higher resolution at the fovea and lower resolution at the visual periphery. Consequently, more computational resources are allocated at the fovea and relatively fewer at the periphery. The FOD processes the entire scene, uses retino-specific object detection classifiers to guide eye movements, aligns its fovea with regions of interest in the input image and integrates observations across multiple fixations. Our approach combines object detectors from computer vision with a recent model of peripheral pooling regions found at the V1 layer of the human visual system. We assessed various eye movement strategies on the PASCAL VOC 2007 dataset and show that the FOD performs on par with the SW detector while bringing significant computational cost savings.

Perspectives on the processing debate in reading
by Cengiz Acartürk (Cognitive Science, METU)
on 19th of October

Abstract Reading has been subject to intermittent research periods since the 1920s. Several computations models have been proposed since past decade, which have resulted in an intense debate on the interaction between oculomotor processes, lexical access and visual attention. From the perspective of general visual cognition, reading provides a prolific test bed -compared to object perception- due to its relatively systematic structure on the stimuli side. On the other hand, there is no consensus on the processes that guide reading recently. This talk will present the experience of the Reading Research Group in the Cognitive Science Program on reading in Turkish, since the past four years. I will present an instance of how theoretical frameworks drive data collection and how they lead to novel experimental paradigms through a set of examples in the reading debate.

Discourse annotation: A multi-lingual approach
by Deniz Zeyrek (Cognitive Science, METU)
on 12th of October

Abstract Linguistic annotation means adding informative information to texts and annotated texts are ultimately inputs to language technology applications. Linguistic annotation of language resources has become a science, encompassing methods of annotation for the design of the corpus, annotation creation, physical format considerations, evaluation, etc. Discourse is a unit above the sentence level and can be analyzed in terms of several kinds of patterning including but not limited to discourse relations. Discourse relations are a level in discourse associated with the semantic relations (contrast, condition, expansion, etc.) that hold among text segments (clauses or groups of clauses). After introducing the basic concepts of linguistic annotation, I will talk about a recent initiative on annotating the transcripts of TED talks that involve several languages (Turkish, English, Portuguese, German and Russian). Our initiative involves annotating discourse relations across texts. I will explain the annotation procedure and describe the corpus along with implications on our understanding of discourse.