Past research

Embodiment of gestures of the head

In these researches we studied if and under what conditions there is an effect of compatibility between the gestures of consent and dissent expressed with the head and the judgment relative to sentences that express objective truths-falsehoods, or subjective preferences.      


To bring clarity to the data of experience and reduce environmental complexity, we make use of mental processes that can register the information we receive from outside, within organized constructs. These constructs, categories, serve the valuable function of guiding our behavior, enabling us to categorize new information based on what we have already acquired and to integrate new knowledge with what we have encountered and processed in the past, thus making interactions with the world easier. Categorization is, therefore, that complex cognitive phenomenon in which both processes of acquiring and using knowledge come into play.

In our lab, we studied how the information available during the categorization process is used when adopting an analytical mode of representation, that is, one that considers individual features separately, or a holistic one, which instead represents examples as a whole. We have studied analytic and holistic categorization modes by introducing a paradigm in which biased examples (in which irrelevant features are made salient) are presented at first and then gradually eliminated.

In this study, we introduced a new experimental paradigm, called Active Feature Composition, which replaces the traditional classification task with a task in which one is required to actively "construct" the examples to be tested by combining their features. This is a production task, in which participants do not have to classify new exemplars but are actively involved in producing items that they believe belong to the learned categories. The procedure is as follows: at the end of the training, after observing a set of particular instances of one or more categories, participants are asked to select features from a set containing all possible dimensions of the features of the learned categories, to combine them together to obtain the complete example. Once the combination is complete, each example is to be assigned a categorical label, thus indicating the category to which it belongs.                        

Greco A., Moretti S. (2017) Use of evidence in a categorization task: analytic and holistic processing modesCognitive Processing, 18, 4, 431-446. DOI: 10.1007/s10339-017-0829-2

Action representation

Representations for concepts and actions are increasingly being considered as tightly connected. The study of embodied cognition is assuming more and more theoretical significance and a recent trend has emerged that questions the idea of a mental concept representation made by the combination of abstract, arbitrary, and amodal symbols, as the first cognitivist stance had posited.

New perspectives stress the role that internal simulation of concrete action, with its nonarbitrary aspects, has for comprehension of abstract representations. Action representation can be considered as componential if sequences of actions can systematically be combined and recombined according to represented rules. This componential view of action representation may be opposed to a holistic one, that considers representations for actions as global, procedural, implicit and not analytical, i.e. not based upon single features.

This line of research pursued in our Lab is aimed at developing a new paradigm for testing different aspects of representations for motor concepts and their connection with verbal representations. In our paradigm the source of empirical data is both experimental and simulative, since we use the same conditions with human subjects and neural networks. Neural networks are a good tool for investigating such questions, because they can naturally implement non-symbolic or analog representations.


Greco A., Caneva C. (2010) Compositional symbol grounding for motor patternsFrontiers in Neurorobotics, 4, 111. doi: 10.3389/fnbot.2010.00111 (PDF file)

A. Greco, C. Caneva, From actions to symbols and back: are there action symbol systems (PDF file). InProceedings of XXVII Annual Conference of the Cognitive Science Society, July 21-23, Stresa, 2005.

Diagrammatic reasoning

This line of research is pursued in our Lab with several aims. The most important investigations are aimed at studying:

1) how verbal descriptions of situations, expressing different properties and relations, are converted into pictorial representations

2) the effectiveness of diagrams in education and other communication and problem solving fields

3) how diagrammatic instructions for assembling objects are created and used

Symbol grounding and compositionality

The classical computational conception of meaning has been challenged by the idea that symbols must be grounded on sensorimotor processes. A difficult question arises from the fact that grounding representations (GREPs) cannot be symbolic themselves but, in order to support compositionality, should work as primitives. This implies that they should be precisely identifiable and strictly connected with discriminable perceptual features. Ideally, each representation should correspond to a single discriminable feature.

The present study was aimed at exploring whether feature discrimination is a fundamental requisite for grounding compositional symbols. We studied this problem by using Integral stimuli, composed of two interacting and not separable features. Such stimuli were selected in Experiment1 as pictures whose component features are easily or barely discriminable (Separable or Integral) on the basis of psychological distance metrics (City-block or Euclidean) computed from similarity judgments. In Experiment 2, either each feature was associated with one word of a two-word expression, or the whole stimulus with a single word. In Experiment 3 the procedure was reversed and words or expressions were associated with whole pictures or separate features.

Results support the hypothesis that single words are best grounded by Integral stimuli and composite expressions by Separable stimuli, where a strict association of single words with discriminated features is possible.

Presuppositions and context

Filippo Domaneschi (Department of Philosophy – University of Genoa)

Presupposition is a widely discussed topic in the analytic tradition since the beginning: Frege, in Über Sinn und Bedeutung claims that the use of a singular term presupposes the existence of the individual denoted. Since this Fregean stance, analytic scholars have given the following definition: a sentence p semantically presupposes a sentence q if we need the truth of q in order to treat p as endowed with sense, that is either as true or as false. Only in the fifties ordinary language philosophy recovered a new concept of presupposition. Starting with Strawson and especially with Austin the concept of presupposition is linked no more on necessary conditions for truth conditional evaluation of a sentence, but as a necessary condition for the felicity or correctness of a speech act.

With Stalnaker analytic philosophy abandoned the notion of semantic presupposition to treat presupposition as a propositional attitude. The focus of the theory passes from the semantic level of sentences to the pragmatic level of utterances, therefore including the ‘cognitive context’ of the speakers background of beliefs. In Stalnaker’s view the common ground of a conversation at a particular time is the set of propositions that participants in that conversation at that time mutually believe as accepted as true and that, for that reason, they take for granted. Hence, an assertion of a sentence p is appropriate only if at the time t the common ground includes the presupposition q required by p, namely q is believed as accepted as true by the speakers. According to Gauker, Stalnaker’s cognitive account is not satisfactory in order to explain the phenomenon of presupposition. He claims that presuppositions cannot be defined in terms of speakers beliefs on the common ground. Rather, it should be taken into account what he calls the ‘objective context’, that is the set of objectively relevant propositional elements – the propositional context – that speakers ought to share in order to evaluate the appropriateness of utterances so as to reach the goal of a conversation.

The main topics of this research are the following:

(i) to give a systematic analysis of problems and terminology of the recent literature on presupposition.

(ii) to compare and provide an evaluation of the cognitive and the objective account of presuppositions.

(iii) to develop an original definition for the notion of ‘context’ that is able to face the objections connected with the received accounts.

(iv) to build a theoretical assessment of the number of ways in which speakers may take for granted a presupposition (that is an unsolved problem in the contemporary debate).

(v) to explore the relation between the ground of presupposition and the notion of ‘goal of a conversation’.

(vi) finally, we aim to work out some tests on cognitive reactions on the role of presupposition in discourse understanding, in order to verify the psychological plausibility of the claims given in (iii) and (iv) and to verify possible alternatives.

Analytic and holistic processing

According to two of the main current approaches to categorization, categories might be constructed either by abstraction of defining features (Smith, Shoben, & Rips, 1974), or by comparison of new cases with previousexamples (Brooks, 1978; Nosofsky, 1986). According to a new trend in literature such approaches may be considered as not necessarily in contradiction, and various possibilities have been explored that attenuate the differences between them (e.g. Erickson M.A., Kruschke J.K., 1998).

One possibility is that these competing approaches refer to different strategies people use in categorization, namely analytic or non-analytic strategies. As the name suggests, analytic strategy involves stimulus analysis (attentional focus, feature scanning, identification of relevant/irrelevant features, rule extraction, etc.) whereas non-analytic strategy relies on memory of previous cases considered as a whole. The analytic strategy (AS) involves detection of relevant features in encountered cases, which are subsequently used to classify new cases by evaluating if they possess or not such features. AS has been related to explicit processes and to the classical theory of categorization. By contrast, non-analytic strategy (NAS) simply involves that positive examples be represented integrally, without a specific representation for features or rules, and new cases be compared with such stored representations. NAS has been related to implicit, unconscious, automatic processes and to exemplar or similarity theories of categorization.
A first problem for NAS is that, if features or rules are not explicitly represented, it is not clear how abstraction could be accomplished. Abstraction is one of the basic processes in categorization; a common claim is that there is no categorization without abstraction. In substance, to abstract means “to neglect” or ignore some aspects of the stimulus, which are not important or relevant, where relevant aspects are ones that are part of the defining criteria of a category (features and rules). It can be asked, then, how would it be possible to know what is relevant and what is not, without some form of stimulus analysis. A possible alternative explanation is calling into play a sort of perceptual abstraction, mainly based on analog or similarity-based comparison, that leads to the same result as if analysis had been accomplished, but without an explicit process of feature identification.
A second problem for NAS is that, still assuming no explicit features/rules representation, it is also not clear how information coming from negative examples could be exploited. In fact, the detection of relevant features normally relies on the comparison between positive and negative examples. In principle, in order to learn a concept, it would be possible to analyze only positive examples, but the inadequacies of a confirmatory approach are well known (since Wason, 1960). In free sorting, or in unsupervised categorization tasks, negative examples may play a little role, but in supervised categorization (where association between examples and their category is given to subjects) abstraction is certainly speeded and guided if “appropriate” negative examples are available. The role of comparison between positive and negative examples is then fundamental in categorization. This comparison would be an unintelligible process without postulating some form of analysis. In NAS, however, it is not clear what kind of information negative examples can give (if any at all), because when this strategy is used – by definition – differences are not analyzed.
These two main questions (is there a perceptual abstraction? should differences between negative and positive examples be explicitly represented?) can be clarified in a task where the impact (number and evidence) and format of irrelevant features (pictorial vs. symbolic) is manipulated, jointly with the kind of information available from negative examples.
The aim of the present research is to explore categorization in conditions where the detection of relevant features, and then AS, is more difficult, by manipulating both the task difficulty and the kind of information available from negative examples.
We shall consider three conditions that seem to prevent detection of relevant features and so increase task difficulty:
1) When relevant features cannot easily be discovered. One condition when this happens is when positive examples have many salient but not relevant features (SNRF). In this case, confusability between positive and negative cases is bigger. This increases task difficulty especially with more general rules, where only a few relevant features are to be detected, and more specific examples are frequently encountered (e.g., when learning the abstract category “animal”, the relevant features are few and relatively nonsalient, whereas many differences between cases must be ignored). Sometimes relevant features may be evident, but very often, particularly with natural categories, irrelevant features are more salient than relevant ones (e.g. moving objects may be categorized by children as “animals”). When many irrelevant attributes are present, performance is biased towards a very specific category (e.g. a child that saw only long-fur dogs might not include short-fur dogs in the “dog” category). In this condition, negative examples play a more crucial role in eliminating irrelevant features.
2) When features cannot easily be composed into a symbolic rule, such as when the rule is too complex. In particular, this often happens with pictorial features, that cannot easily be described using manipulable symbols. In principle, of course, the difference between pictorial and symbolic stimuli is not so clear-cut: pictures may act as symbols and many symbols are pictures (e.g. in written language). However, in most cases a picture cannot be simply translated into a symbolic formula.
3) When the attentional focus is not directed towards the part of the stimulus that includes relevant features.
In the present research such three conditions were controlled in a task where a simple but very general (then difficult) rule was chosen, different sources of confusability (pictorial vs. symbolic) were added, and the subjects’ attentional focus was splitted and monitored.
The general aim of the experiment was to test how irrelevant features affect category learning and strategies in use of evidence. In this experiment, the rule was very general and difficult to detect (presence of a single letter in a 4-letter string, at any position), and SNRF were the presence of one or two additional letters at fixed positions, and/or the visibility of a particular pictorial pattern. The rationale of manipulation was to produce an initial bias towards very specific features, by presenting – at the very first training phase – positive examples with many SNRF and negative examples completely free from SNRF, in order to assess the power of negative cases in correcting such bias in subsequent phases.
In particular, the hypothesis was tested that the format (pictorial vs. literal) of irrelevant features affects the capability of eliminating them. If analytic strategy is more effective in eliminating irrelevant features, then it should be more difficult to eliminate pictorial SNRF, because they cannot be easily coded as explicit, discrete features.


Subitizing is the ability to estimate the number of elements without having to count them. It is a process that plays an important role during learning and in carrying out arithmetic operations. We studied how this ability is related to automatic attention processes.  

Applied research


This research is described in the Italian page:

Biases and error prevention in medical reasoning                        
Hand-off management                        

Last update 27 June 2023