BMe Research Grant

Ferenc Kemény

E-mail address

Home page

Phone: +36 1 463 3525

BMe Research Grant - 2010

3rd Prize

Psychology PhD School

Department of Cognitive Sciences

Supervisor: Ágnes Lukács, PhD

Modality- and Domain Independence in Probabilistic Categorization

Introducing the research area

The proposed research deals with the modality- and domain independence of category learning. We modified the Weather Prediction task, a well-known probabilistic category learning paradigm. This way we may be able to find out whether probabilistic categorization is specific to a given domain or modality or whether it is a robust, domain-general learning mechanism. In the research participants face different versions of the same task. The versions are structurally identical, only the set of stimuli differs – it may consist of verbal or non-verbal stimuli presented visually or auditorily. The major research question is whether or not different conditions lead to the same level of category learning.

Brief introduction of the research place

In 2001 the Centre for Cognitive Sciences was founded at the Budapest University of Technology and Economics. The centre was transformed into a department in 2004. It works together with the HAS-BME Research Group. The joint department soon became one of the most well-known places of cognitive psychology education. The same intellectual density may only be found at the HAS Research Institute for Psychology – collaborators of the Institute also teach and supervise students of the Department.

History and context of the research

During our everyday life we perceive a number of different objects. These objects may differ from or be similar to each other by different extents. However, the unique features of the different objects are frequently irrelevant, or even present a handicap in information processing. These unique features – which differentiate one object from others – require a huge processing capacity that we simple cannot provide. When children are playing with cards of cars, their goal is to find the car with the highest top speed. When their parents are seeking a car for the family, they have a number of different features to pay attention to. This may lead to a differentiation between two cars of the same make and same type. Both of them have to form their specific categories that best code either the way the opponent can be defeated (children's category) or the type of car the family requires (parents' category).

Frequently, criterion for category membership is not obvious choice. An example taken from everyday life refers to the category of birds. Birds have wings, lay eggs and fly. However, the ostrich is a bird, too, and still cannot fly. Similarly, reptiles lay eggs and bats fly. The situation where cues do not completely predict category membership is called probabilistic categorization. The research deals with probabilistic categorization, and more closely with one of the most popular paradigms known as Weather Prediction task.

Aim of the research

Over the past decades, category learning has become one of the most popular research areas within the cognitive sciences. Quite often, category learning brought different disciplines together. A number of neuropsychological, cognitive neuroscience or developmental papers have been published on neural background of categorization, categorization patterns shown by different clinical groups or the role of awareness in category learning.

Research in the neural background of categorization led many researchers to the conclusion that representation of categories are located in different areas of the brain (Warrington, 1975).The question arises whether categories that are possibly localized in different areas of the brain can be acquired with the same or with different learning mechanisms. Earlier researches did not cover this issue.

One of the frequently used research settings is Probabilistic Categorization. This form of learning has been associated with the family of implicit/procedural learning mechanisms. This means that learning is incidental, content acquired is complex, and is hard to verbalize. Earlier research acknowledges a number of similar tasks, like Statistical Learning, Artificial Grammar Learning or Serial Reaction Time Task. Research in connection with these tasks showed that the underlying mechanisms are robust and lack modality or domain constraints. In general, earlier literature left open whether probabilistic categorization is expected to be domain and modality dependent or robust.


The task used in the current research is the Weather Prediction task (Knowlton, Squire, & Gluck, 1994). In the task, participants face 1, 2 or 3 out of 4 different cues. The role of the participants is to decide whether it will be rain or sunshine. Each cue has a preset value that leads to the different outcomes. Cue1 predicts sunshine in 90% of cases, Cue2 in 70%, Cue3 in 30% and Cue4 in 10%. In all other cases, cues predict rain. Participants know neither the probabilities nor the structure of the task - they are merely instructed to guess what the weather will be like. They receive immediate feedback after each answer. The experiment consists of four blocks of 50 items.

Performance may be characterized by two variables: one is the proportion of correct answers, and the other is the strategy used in solving the task. Earlier research suggests that proportion of correct answers may not differ significantly from chance level (50%) in the first block, but may reach as high as 80% in the fourth.

While the amount of correct answers may reflect accuracy of categorization, strategy fitting explains how categorization takes place. Earlier research (Gluck, Shohamy, & Myers, 2002) identified three strategies: in One–Cue strategy the solver focuses on a specific cue, and whenever the cue appears they answer consistently, whereas in the absence of the cue random answer is given. Singleton strategy means that consistent response is given if only one cue is present, whereas in case of combinations random answer can be expected. Multi–Cue strategy leads to the highest performance; basically, it means the summation of the observed predictive values and answering according to the sum.

In the original task, cues were geometric forms or tarot cards, i.e. cues have been presented visually. If probabilistic category learning is independent of modalities and domains, then performance and strategy use is not expected to change if either modality or domain is modified.

Visual-nonverbal condition is the original Weather Prediction task, while in the verbal-visual task cues are CV syllables. Stimuli of the auditory verbal conditions are the same syllables as in the other verbal condition – though presentation is auditory –, while stimuli of the nonverbal auditory conditions are sounds like a knock, tweeting, quacking or the sound of a guitar.


Over the past few years we have conducted a number of experiments in connection with probabilistic categorization. Our most important result is that language impaired children show impaired probabilistic categorization performance (Kemény & Lukács, 2010). The results suggest that this basic skill participates in higher cognitive processes like language acquisition.

Language impairment is a developmental disability in which language is specifically affected while other domains remain intact (Pinket, 1999). This impairment has been a great example for the modular structure of mind and the autonomous functioning of language. A more recent theory, the Procedural Deficit Hypothesis (Ullman & Pierpont, 2005) suggests that language impairment is a result of the malfunctioning of a more general cognitive system, the procedural system.

Procedural and declarative systems are components of the representation-based dichotomy of memory. The declarative system codes factual information into clear, static representations, while the procedural system codes process-like information into fuzzy, dynamically changing representations. Examples for declarative representations may be dates, while an example for a procedural representation may be a skill like riding a bicycle or cognitive skills like counting or categorization – e.g. the Weather Prediction task. Our hypothesis was that language impaired children would show lower performance than their typically developing peers in the Weather Prediction task. Our hypothesis was justified, what's more, performance of children with language impairment did not differ from chance level throughout most of the task. Our results are in concert with the Procedural Deficit Hypothesis.

In our other experiments studying probabilistic categorization, we wanted to answer whether categorization performance is enhanced if the arbitrary link between cues and outcomes (geometric forms>weather) are transformed into a transparent link (Kemény & Lukács, 2009). In the original Weather Prediction task, participants based their guesses regarding weather on geometric forms. In such a case, link between cues and outcomes is arbitrary – one cannot be concluded from the other. In our experiments, we used stimuli where the link between cues and outcomes were perceptually transparent: cues appeared on segments of pictures, based on which participants had to decide whether a given picture shows a boy or a girl. To achieve this setting the cues must be ambiguous. Our hypothesis was that transparent link between cues and outcomes would enhance categorization. The results showed that this is only true at the beginning of the task and in later phases the arbitrary condition has the advantage. The reason might be that a strong perceptual link enhances single strategies like One-Cue or Singleton, but impairs the use of Multi-Cue strategy.

Expected impact and further research

The proposed research may help understand human mind in terms of modularization. Results may also contribute to better understand categorization and its link to other (implicit) processes, similarities and differences.

In the proposed research a visual nonverbal task is adapted to visual and auditory modalities and verbal and nonverbal domains. What so far has not been paid attention to is the advantages and disadvantages that are rooted in the different modalities themselves. Also, can we expect the same performance if cues are presented simultaneously or sequentially? Will the difference between sequential and simultaneous presentation lead to the same result in both auditory and visual domains? These questions need further studies.

Publications, references, links

Publications related to the research.

  • Kemény F., Keresztes A. (2008). A nem tudatos ismeretszerzés idegrendszeri alapjai. Pedagógusképzés, 6, 25–42.(The neurological bases of non-conscious acquisition of information)
  • Kemény F., Lukács Á. (2009). The Role of Transparency in Probabilistic Category Learning. Learning and Perception, 1, Supplement 1, pp26.
  • Kemény F., Lukács Á. (2010). Impaired procedural learning in language impairment: results from probabilistic categorization. Journal of Clinical and Experimental Neuropsychology, 32(3), 249–258.
  • Kemény F., Lukács Á. (2007). Probabilistic Category Learning in Language Impairment. Poster presented at the First Language Acquisition Summer-school, 3–7 September, 2008, Basel
  • Kemény F., Lukács Á. (2009). Probabilistic Category Learning in Children and Adults: Sunshine or Chocolate? Poster presented at 15th ESCoP, 29 August – 1 September 2007., Marseille
  • Kemény F., Lukács Á. (2009). The Role of Transparency in Probabilistic Category Learning. Lecture presented at the 16th ESCoP, 2–5 September, 2009., Cracow
  • Kemény F., Lukács Á. (2009). The Role of Transparency in Probabilistic Category Learning. Poster presented at DUCOG II., 22–24 May, 2009., Dubrovnik
  • Kemény F., Lukács Á. (2010). Az ingermodalitások hatása a probabilisztikus kategóriatanulásra. Lecture presented at XVIII. MAKOG, 25–26 January, 2010., Budapest




  • Gluck, M. A., Shohamy, D., & Myers, C. (2002). How do people solve the "weather prediction" task?: Individual variability in strategies for probabilistic category learning. Learning & Memory, 9(6), 408–418.
  • Kemény, F., & Lukács, Á. (2009). The Role of Transparency in Probabilistic Category Learning. Learning and Perception, 1(Supplement 1), 26.
  • Kemény, F., & Lukács, Á. (2010). Impaired procedural learning in language impairment: Results from probabilistic categorization. Journal of Clinical and Experimental Neuropsychology, 32(3), 249–258.
  • Knowlton, B. J., Squire, L. R., & Gluck, M. A. (1994). Probabilistic classification learning in amnesia. Learning & Memory, 1, 106–120.  
  • Pinker, S. (1999). A nyelvi ösztön. Budapest: Typotex.
  • Pléh, C. (1999). A lélektan története. Budapest: Osiris
  • Ullman, M. T., & Pierpont, E. I. (2005). Specific language impairment is not specific to language: the procedural deficit hypothesis. Cortex, 41, 399–433.
  • Warrington, E. K. (1975). The selective impairment of semantic memory. Quarterly Journal of Experimental Psychology, 27(4), 635–657.