Back to this year's colloquia: 2017-2018
Note: Earlier years back to 96'-97' will soon be included, gradually.
Academic Year 2016-2017
back to top
Dealing with Out-of-Vocabulary (OOV) Problem in Natural Language Processing
by Burcu Can (Computer Engineering, Hacettepe University)
on 26th of May
Abstract Out-of-Vocabulary (OOV) problem is inevitable in many natural language processing applications. Morphology is one of the major factors that poses the OOV problem by leading to a large number of different word forms. The problem is even more severe in agglutinative languages such as Turkish. We deal with OOV problem with morphological segmentation that is the task of splitting words into their smallest units called morphemes. For example, the word *Türkçeleştiremediklerimizden *is split as *Türk+çe+leş+tir+e+me+dik+ler+imiz+den. *We also tackle the OOV problem in part-of-speech tagging (PoS tagging) by a joint task that involves stemming. Thus, stems are tagged rather than words in order to reduce the sparsity in the corpus.
A Tractarian Method for Computational Ontology
by Aziz F. Zambak (Philosophy, METU)
on 12th of May
Abstract Tractatus Logico-Philosophicus, written by Ludwig Wittgenstein, is one of the most influential book in the history of philosophy. In my presentation, I will describe the general metaphysical principles of Tractatus and discuss how can we apply these principles to Knowledge Representation and Formal Ontologies.
Adaptive Control of Behaviour: Complex Supervision or Simple Retrieval?
by Nart Bedin Atalay (Psychology, TOBB University of Economics and Technology)
on 5th of May
Abstract Cognitive control is the ability to adjust behaviour to meet task demands under a changing context. Incoming information is suppressed or processed in light of the current goal and the environment. There is an ongoing debate on whether the adaptive control of behaviour is governed by a supervisory mechanism or by retrieval of past events. I will describe our studies and discuss implications to resolving this debate.
Who is to blame when a cat fears a dog?: Implicit Causality Biases in Turkish Psych-Verbs
by Duygu Özge (Foreign Language Education, METU)
on 21st of April
Abstract Different verb types have different causality biases. In action verbs, for instance, it is always the subject who is responsible for the action. For the sentence 'John broke the vase’, the subject (John) is the causer. However, the causality information is not as straightforward – but it is implicit – for some event types. Implicit causality refers to the tendency to assign a specific participant of the event as the cause of the resulting action, situation, or emotion. I focus on the implicit causality biases for Turkish psychological verbs (psych-verbs) in this talk. Psych-verbs (e.g., frighten, fear, love) tend to reflect irregular intuitions about the cause of the triggered emotion. For a sentence like ‘The cat frightens the dog’, people tend to hold the subject (cat) responsible for the feeling whereas the cause is not as clear in sentences like ‘The cat fears the dog’. This irregularity about which participant receives the causal role raises some interesting issues concerning (i) the widely held assumption that subjects are more prominent, (ii) the ranking of event participants as possible antecedents of ambiguous anaphors, and (iii) the nature of syntax-semantics mapping during language acquisition. I will discuss these issues in light of our recent pronoun resolution study.
Early Symptoms of Autism Spectrum Disorders and Eye Tracking Technology
by Selda Özdemir (Special Education, Gazi University)
on 14th of April
Abstract Autism spectrum disorders (ASD) is a developmental disorder that appears in the first three years of life and affects social emotional development and social communication skills of children. Children with ASD often have difficulty in making eye contact, initiating social interactions, maintaining conversations, displaying interactive social play behaviors, understanding others’ intentions and emotions, adjusting to transitions, and displaying verbal and nonverbal social interaction behaviors. Due to these social communication problems, children with the disorder appear to be isolated and obvious to the social world and such problems deteriorate these children’s parent-child interactions, peer relations and social adjustment throughout childhood and adulthood.
Implementation of early intervention programs that are designed to address social communication problems of children with ASD are very important in changing the tragic development of the disorder. In order to begin early intervention efforts as soon as possible, assessment of ASD risks in the first three years of life is crucial. This lecture will begin with a review of early symptoms of ASD and will continue by discussing a research project on examining eye tracking patterns of children with ASD supported by SCRTC-EU(COST). We believe that eye tracking technology now make it possible to improve the precision of existing assessments for ASD. Second, an early intervention program “Parent Mediated Joint Attention Early Intervention Program” will be introduced in the lecture. An important aspect of this program is that it promotes joint attention skills of young children with ASD and/or high risk for ASD and provide them with necessary early developmental experiences through enriched social exchanges. We think that this improvement is necessary to address early developmental and learning needs of children with ASD.
Öğrenmede Elektronik Okuma: Çok-katmanlı Etkileşim ve Öğretimsel İleti Tasarımı
by Serkan Şendağ (Computer and Teaching Technologies, Mersin University)
on 8th of April
Talk in Turkish
Abstract Öğrenmek için okuma, temelinde bir materyalden öğrenme etkinliğidir. Bu tür etkinliklerin özellikle bir hedef kitlesi vardır. Bu hedef kitlenin farklı beceri, bilişsel, duyuşsal ve hazırbulunuşluk düzeyleri vardır. Bunlara okuyucuların farklı sosyo-kültürel çevrede yaşamasının eklenmesiyle okuyucularda farklı beklentilerin oluşması doğal bir hal almaktadır. Bu durumda hedef kitlenin gereksinimlerini karşılarken gerekli içeriği etkili bir şekilde aktarmak bir durumdan diğerine değişen bağlamsal bir problem çözme süreci haline gelmektedir. Oysa ders, çalışma ve alıştırma kitapları başta olmak üzere öğretimsel içerikler standart bir anlayışa göre sunulmaktadır. Bu bağlamda elektronik ortamda sunulan öğrenme içerikleri zengin bir etkileşim ortamı sağlayarak okuma içeriğinin bireyselleştirilmesine olanak tanıma potansiyeline sahiptir. Güncel bilişim teknolojileri araçları yazılı metnin yanı sıra, ses, durağan resim, hareketli görüntü, iki ve üç boyutlu animasyonlar, sanal ve artırılmış gerçeklik uygulamaları ile çok katmanlı (multimodal) etkileşime olanak tanıyan bir öğrenme çevresi sunmaktadır. Böyle bir öğrenme çevresinden gerektiği gibi yararlanabilmek için öğretimsel tasarımın etkili bir şekilde yapılması gerekmektedir. Bu nedenle elektronik ortamda sunulan öğretimsel içeriğin tasarım özellikleri ile bu konudaki araştırma olanakları ve disiplinler arası çalışmanın önemi üzerinde durmak bu çalışmanın temel amacını oluşturmaktadır.
Tracking Eye Movements to Investigate Covert Processes in Language Comprehension
by Mustafa Seçkin (Neurology, İstanbul University)
on 31st of March
Abstract Progressive degeneration of language network in Alzheimer’s disease and related disorders may disrupt cognitive processes long before language impairment can be detected by clinicians. Detecting early impairment of language functions, in particular comprehension deficits, requires investigating covert processes responsible for linking words and sentences to their mental representations. However, traditional cognitive tests have low sensitivity for the diagnosis of pre-symptomatic impairment in language functions. Recently, we developed a new method where eye movements have been used to detect the impairment in linking written (or spoken) words to meaningful mental representations during a word-object matching paradigm. Additionally, we successfully integrated novel sentence-completion paradigms with eye tracking experiments to study the underlying mechanisms of sentence processing in language network. The preliminary findings suggest that tracking eye movements can also be used to reveal how language network decodes grammatical relations between the verb and its arguments and this process can be disrupted before overt grammatical errors can be observed in patients with neurodegenerative aphasia.
Symbol Emergence in Robots
by Emre Uğur (Computer Engineering, Boğaziçi University)
on 24th of March
Abstract In this talk, I will discuss whether, how and why symbols should emerge in intelligent systems such as humans and robots. For this, I introduce a learning framework that enables manipulator robots (with 7 dof. arms and multi-fingered hands) to progressively develop better and more abstract sensorimotor and cognitive skills through physical and social interactions with the environment. I will describe how behavior primitives, affordance categories, prediction and imitation mechanisms, and finally high-level symbolic operators can emerge in the continuous sensorimotor space of the robots through learning from such interactions. After giving a partial answer to 'how' question this way, I would like to discuss the 'whether' and 'why' questions with the audience.
Counter-Expectational Sense in Turkish and in General: Some Critical Data and Prospects for a Formal Treatment in a Layered-DRT
by Ceyhan Temürcü & Deniz Zeyrek (Cognitive Science, METU)
on 17th of March
Abstract Apparently, all human languages have strategies for challenging possible inferences from a preceding clause or discourse segment. This sense, which take part in the semantic ranges in Turkish *ama *or *fakat*, and of English *but*, has often been dubbed 'counter-expectational' (CE) and subsumed under the more general family of 'contrastive' senses. I this talk we will (i) provide critical data that reveals some semantic and pragmatic ingredients of the CE sense, (ii) propose a descriptive specification for this sense and (iii) provide prospects for a formal treatment of CE in a multi-layered version of DRT, which distinguishes temporal, epistemic, volitional, and illocutionary layers. In a nutshell, we will argue that CE denies a possible inference afforded by a counter-actual knowledge state.
Data we will consider will include cases where a possible inference from the first conjunct is explicitly challenged by the second conjunct (1), where it is indirectly challenged by the second conjunct (2-3), where there is apparently no (syntactic or discursive) first conjunct (4), where the first and second conjuncts belong to different interlocutors (5), and where the second conjunct denies a content which is directly expressed (rather then implied) by the first conjunct (6-8):
(1) Ahmet kısa boylu ama iyi basketbol oynuyor.
(2) Zor bir iş ama bence bir dene.
(3) Hava soğuk ama (neyse ki) paltom var.
(4) Ama sen sigara içiyorsun!
(5) I1. Ahmet evde / I2. Ama öğlen onu okulda gördüm.
(6) Ahmet burada olabilir, ama olmayabilir de.
(7) Arabayı sana verirdim ama sarhoşsun.
(8) Ahmet burada olabilirdi ama değil.
The proposal will be briefly compared to that of Winter and Rimon (1994), which is based on Veltman's (1986) data logic, and that of Toosarvandani (2014), which recruits Kratzer's (1981, 1991) analysis of modality.
Machine Learning of the Mind
by Fatoş T. Yarman Vural (Computer Engineering, METU)
on 10th of March
Abstract How does the information represented in the brain? We approach this question from the Artificial Intelligence perspective. Loosely speaking, we attempt to model the "Natural Intelligence" by using Artificial intelligence. We suggest a deep learning method to model the cognitive states of the brain, using functional magnetic resonance imaging data (fMRI). The suggested method assumes that the cognitive processes can be represented by a deep learning system, which is trained by the fMRI measurements.The most challenging problem of modeling such a learning system is the design of a model for the fMRI data, to extract the information about the underlying patterns of brain activity. In this study, we propose a new method called Mesh Learning, which learns the connectivity patterns of the brain. Then use these patterns to model and classify the fMRI data recorded during a set of cognitive tasks. The experimental results indicate that the suggested model has excellent performance compared to the state of the art statistical techniques.
Evolutionary Robotics and Swarm Robotics
by Ali Emre Turgut (Mechanical Engineering, METU)
on 3rd of March
Abstract Electronics and computers advanced considerably in the last two decades. As a consequence, we started to see more and more robots around us today than we see in the factories. Unmanned air vehicles and autonomous vacuum cleaners are two such examples. Main difference between these robots and the factory robots are that these robots have to tackle unstructured and unknown environments making it very challenging to design their controllers. There are different approaches to design the controllers. Evolutionary robotics is one such promising approach taking its inspiration from natural evolution. In the recent years, evolutionary algorithms have been applied successfully to different engineering problems including robotics. In this talk, I will make a brief introduction to evolutionary algorithms and evolutionary robotics. Then, I will briefly talk about swarm robotics and evolutionary swarm robotics field. The talk will end up with discussion about future directions.
L2 Influence on L1 for Turkish-English Late Bilinguals
by Özkan Kılıç (Cognitive Science, METU)
on 6th of January
Abstract The potential effect of acquisition of the second language (L2) on the first (L1) is less well-understood. One might assume that the L1, once acquired, would be stable and resistant to change. However, accumulating evidence indicates that acquisition of an L2 can, in some respects, influence use of the L1. One way of thinking about where L2 influence on L1 might arise is based on the fact that linguistic knowledge must be accessed in memory in order to be used. If the lexicon is stored in declarative memory but morphosyntactic and phonological knowledge are stored in procedural memory, then lexical performance may be most susceptible to L2 influence because it will be most subject to standard memory parameters such as frequency of retrieval and use. By this line of reasoning, morphosyntax and phonology, which involve more automatized procedures, should be less vulnerable to L2 influence. In this colloquium, we will talk about L2 influence on L1 for Turkish-English late bilinguals in terms of: pronoun usage, verb framing and phonotactics.
What Good is a Formal Discourse Representation?
by Umut Özge (Cognitive Science, METU)
on 23rd of December
Abstract Two phenomena are critical in holding a natural language discourse together. One is "anaphora resolution" as in inferring that the pronoun `they' in (2) refers to a bunch of delegates reported to have arrived in (1).
(1) A few delegates arrived.
(2) They registered.
The other is presupposition. Take sentence (3)*:
(3) It seems to me that the moment has come that the bombing of German cities simply for the sake of increasing the terror, though under other pretexts, should be reviewed.
The speaker of (3) already takes it for granted the content in (4):
(4) The German cities are being bombed for the sake of increasing the terror.
For instance, if you are not happy with the content in (4), responding `No, it hasn't' or `No, not really' to (3) would not help; you rather have to say something like `Wait a minute, who says that the bombing is for the sake of increasing the terror'. In technical terms, (3) is said to presuppose (4), and presuppositions survive under manipulations like negating the sentence or putting it into a question form. I will discuss how these two phenomena can be modeled under a formal discourse model, resulting in a unified treatment of anaphora and presupposition as one and the same phenomenon. Then, I will review some data from Turkish noun phrases which pose some interesting challenges to the model, together with sketches of possible solutions.
*Winston Churchill to Arthur Harris, Commander-in-Chief of RAF’s Bomber Command in 1945. Reported in Blitz, Bombing and Total War, Channel 4, January 15, 2005.
The Abysmal Splendor of Cognitive Neuroscience
by Didem Gökçay (Cognitive Science, METU)
on 16th of December
Abstract In this talk, a brief history of cognitive neuroscience will be presented from 1980's and on. Localization efforts regarding association of mental functions with specific brain structures will be explained. Breakthroughs brought by fMRI methodology will be summarized. Some tools and features of this domain will be presented and the hype and failings regarding cognitive neuroscience will be illustrated. The talk will conclude with a critique of new funding directions.
Sensitivity to Vowel Harmony in the First Year of Life: Implications for Theories of (Phonological) Learning
by Annette Hohenberger (Cognitive Science, METU)
on 9th of December
Abstract Infants’ language perception abilities develop dramatically in their first year of life. Starting out as a universal listener they become (neurally) committed to their native language (Kuhl, 2014). With behavioral methods – preferential listening and habituation – we studied the development of vowel harmony longitudinally in monolingual Turkish infants from 6-10-months of age. Infants at both ages can discriminate harmonic from disharmonic versions of the same word. However, 6-month-olds show discrimination only when they have been habituated with the disharmonic version first (e.g., zeybek-tan) and then hear the harmonic version in the test (e.g., zeybek-ten) whereas 10-month-olds can discriminate the versions in either order, indicating a more generalized discrimination ability. In terms of preference we found a characteristic shift from familiarity to novelty: 6-months-old infants preferred listening to harmonic words whereas as 10-month-olds preferred listening to disharmonic words. This shift indicates that infants first acquire the regular (harmonic) phonological pattern and filter out the irregular (disharmonic) pattern. Once the regular (familiar) pattern has been firmly established – which happens roughly between 6 and 10 months of age – they discern and appreciate the deviant (unfamiliar, novel) pattern. Such shifts have been observed in other cognitive domains as well, e.g., in vision.
The claim that early sensitivity to vowel harmony is based on language-specific auditory experience during the first year of life is supported by the negative results of van Kampen et al. (2008) who could not find any listening preference of 6-month-old German infants to vowel-harmonic vs. –disharmonic words. However, Mintz et al. (under review) could show that English-acquiring infants are able to segment words based on vowel (dis-)harmony cues after a short familiarization phase of only 50 sec. Obviously, studies using different experimental methodologies – discrimination, preference, segmentation – tap different processing mechanisms in early language development. In my talk I will try to synthesize these various findings and try to reconcile them with theories of early (phonological) learning.
Negation and Its Logical Representation(s)
by Ceyhan Temürcü (Cognitive Science, METU)
on 2nd of December
Abstract In this talk I will try to show that negation in natural languages comes in many different types, which extend beyond the classical internal-external (or, narrow- vs wide-scope) dichotomy. I will argue that a multi-modal logical representation is necessary to capture different types of negation, in terms of content, temporal, epistemic and volitional levels.
Multilingual Discourse Annotation: Annotating TED Talks
by Deniz Zeyrek (Cognitive Science, METU)
on 25th of November
Abstract 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 these basic concepts of discourse, I will talk about a recent initiative on annotating the transcripts of TED talks that involve several languages (Turkish, English, Portuguese, German and Russian). Annotation means adding informative information to texts and annotated texts are ultimately inputs to language technology applications. 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.
by Cem Bozşahin (Cognitive Science, METU)
on 18th of November
Abstract Verbs define thematic structure and argument structure. In one conception of grammar, they also define the basic word order of a language. We know that all possibilities of Subject (S), Object (O) and the verb (V) have been attested in the world's languages, and that there are actually 8 possibilities rather than 6, for a transitive clause. Therefore a theory of grammar must start with eight possibilities for the child for any language. I will show a micro-world experiment about how we can go from such indeterminacy to well-established basic word orders, by relating empirical probabilities to narrowly constrained internal categories, to the point of seeing almost switch-like behavior, without a switch. Implications for other cognitive processes will be discussed, such as planning. Its relation to Language of Thought Hypothesis is also discussed, time permitting.
Applications of Optical Brain Imaging in Human-Computer & Human-Human Interaction
by Murat Perit Çakır (Cognitive Science, METU)
on 11th of November
Abstract Recent advances in sensor design and data acquisition/analysis methods have opened up the possibility to conduct brain imaging studies in the field. The increasing portability of electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) systems allow researchers to study brain dynamics in various ecological settings (e.g. operating some machinery via an interface or engaging in some learning activity). The increasing availability of these systems have also brought the possibility of studying brain dynamics across participants in the context of a collaborative activity. This seminar will provide a brief overview of optical brain imaging methods and illustrate recent studies conducted at our lab to explore the use of fNIRS in domains such as ergonomics, communication and learning.
Group Eye Tracking (GET) Paradigm for Social Cognition Research
by Cengiz Acartürk (Cognitive Science, METU)
on 4th of November
Abstract Multi-user eye tracking (aka. group eye tracking GET) paradigm has opened new horizons for many disciplines ranging from cognitive sciences, education, economics and social psychology to human-computer interaction. In this talk, I will present two studies that have been conducted by using the GET paradigm. The first is a three-player game, where the participants played the game under competitive and elaborative conditions. The second study focuses on a decision-making task, where the participants make risky or conservative monetary choices. My goal is to share the findings, present the challenges in current research, and discuss the future of multi-user eye tracking paradigms.