How many kinds of reasoning? Inference, probability, and natural language semantics

Adaptation in Predictive Prosodic Processing in Bilinguals

Native language listeners engage in predictive processing in many processing situations and adapt their predictive processing to the statistics of the input. In contrast, second language listeners engage in predictive processing in fewer processing situations. The current study uses eye-tracking data from two experiments in bilinguals’ native language (L1) and second language (L2) to explore their predictive processing based on contrastive pitch accent cues, and their adaptation in the face of prediction errors. The results of the first experiment show inhibition effects for unpredicted referents in both the L1 and the L2 that can be modeled with a Bayesian adaptation model, suggesting that bilinguals adapt their prediction in the face of prediction errors in a way that is compatible with the model. In contrast, the results of the second experiment, after a training phase that increased the predictive validity of the cue, show inhibition effects for unpredicted referents only in the L1, but not in the L2. In addition, the Bayesian adaptation model significantly predicts only the L1, but not the L2 data. The results are discussed with respect to adaptation to the statistical properties of the input.

Rational Adaptation in Lexical Prediction: The Influence of Prediction Strength

Recent studies indicate that the processing of an unexpected word is costly when the initial, disconfirmed prediction was strong. This penalty was suggested to stem from commitment to the strongly predicted word, requiring its inhibition when disconfirmed. Additional studies show that comprehenders rationally adapt their predictions in different situations. In the current study, we hypothesized that since the disconfirmation of strong predictions incurs costs, it would also trigger adaptation mechanisms influencing the processing of subsequent (potentially) strong predictions. In two experiments (in Hebrew and English), participants made speeded congruency judgments on two-word phrases in which the first word was either highly constraining (e.g., “climate,” which strongly predicts “change”) or not (e.g., “vegetable,” which does not have any highly probable completion). We manipulated the proportion of disconfirmed predictions in highly constraining contexts between participants. The results provide additional evidence of the costs associated with the disconfirmation of strong predictions. Moreover, they show a reduction in these costs when participants experience a high proportion of disconfirmed strong predictions throughout the experiment, indicating that participants adjust the strength of their predictions when strong prediction is discouraged. We formulate a Bayesian adaptation model whereby prediction failure cost is weighted by the participant’s belief (updated on each trial) about the likelihood of encountering the expected word, and show that it accounts for the trial-by-trial data.

A test of two processes: The effect of training on deductive and inductive reasoning

Dual-process theories posit that separate kinds of intuitive (Type 1) and reflective (Type 2) processes contribute to reasoning. Under this view, inductive judgments are more heavily influenced by Type 1 processing, and deductive judgments are more strongly influenced by Type 2 processing. Alternatively, single-process theories propose that both types of judgments are based on a common form of assessment. The competing accounts were respectively instantiated as two-dimensional and one-dimensional signal detection models, and their predictions were tested against specifically targeted novel data using signed difference analysis. In two experiments, participants evaluated valid and invalid arguments, under induction or deduction instructions. Arguments varied in believability and type of conditional argument structure. Additionally, we used logic training to strengthen Type 2 processing in deduction (Experiments 1 & 2) and belief training to strengthen Type 1 processing in induction (Experiment 2). The logic training successfully improved validity-discrimination, and differential effects on induction and deduction judgments were evident in Experiment 2. While such effects are consistent with popular dual-process accounts, crucially, a one-dimensional model successfully accounted for the results. We also demonstrate that the one-dimensional model is psychologically interpretable, with the model parameters varying sensibly across conditions. We argue that single-process accounts have been prematurely discounted, and formal modeling approaches are important for theoretical progress in the reasoning field.

Do modals identify better models? A comparison of signal detection and probabilistic models of inductive reasoning

The nature of the relationship between deductive and inductive reasoning is a hotly debated topic. A key question is whether there is a single dimension of evidence underlying both deductive and inductive judgments. Following Rips (2001), Rotello and Heit (2009) and Heit and Rotello (2010) implemented one- and two-dimensional models grounded in signal detection theory to assess predictions for receiver operating characteristic data (ROCs), and concluded in favor of the two-dimensional model. Recently, Lassiter and Goodman (2015) proposed a different type of one-dimensional model, the Probability Threshold Model (PTM), that they concluded offered a good account of data collected over a range of decision modals (e.g., How likely, possible, or necessary is the argument conclusion?). Here, we apply the PTM and the signal detection models to ROCs from 3 large experiments in which participants made judgments about arguments varying in terms of modals introduced by Lassiter and Goodman (2015). Two independent variables that are theoretically important for the study of inductive reasoning, namely premise-conclusion similarity (as utilized in Heit & Rotello, 2010) and number of premises (as utilized in Rotello & Heit, 2009), are also varied in Experiments 1 and 2, respectively. In all cases, the PTM provides the poorest fit both quantitatively and qualitatively; the two-dimensional signal detection model is preferred.

Complex probability expressions & higher-order uncertainty: Compositional semantics, probabilistic pragmatics & experimental data

We present novel experimental data pertaining to the use and interpretation of simple probability expressions (such as possible or likely) and complex ones (such as possibly likely or certainly possible) in situations of higher-order uncertainty, i.e., where speakers may be uncertain about the probability of a chance event. The data is used to critically assess a probabilistic pragmatics model in the vein of Rational Speech Act approaches (e.g., Frank and Goodman, 2012; Franke and Jäger, 2016; Goodman and Frank, 2016). The model embeds a simple compositional threshold-semantics for probability expressions, following recent work in formal linguistics (Swanson, 2006; Yalcin, 2007, 2010; Lassiter, 2010, 2017; Moss, 2015).

Pareidolic and Uncomplex Technological Singularity

“Technological Singularity” (TS), “Accelerated Change” (AC), and Artificial General Intelligence (AGI) are frequent future/foresight studies’ themes. Rejecting the reductionist perspective on the evolution of science and technology, and based on patternicity (“the tendency to find patterns in meaningless noise”), a discussion about the perverse power of apophenia (“the tendency to perceive a connection or meaningful pattern between unrelated or random things (such as objects or ideas)”) and pereidolia (“the tendency to perceive a specific, often meaningful image in a random or ambiguous visual pattern”) in those studies is the starting point for two claims: the “accelerated change” is a future-related apophenia case, whereas AGI (and TS) are future-related pareidolia cases. A short presentation of research-focused social networks working to solve complex problems reveals the superiority of human networked minds over the hardware‒software systems and suggests the opportunity for a network-based study of TS (and AGI) from a complexity perspective. It could compensate for the weaknesses of approaches deployed from a linear and predictable perspective, in order to try to redesign our intelligent artifacts.

FN400 and sustained negativity reveal a premise monotonicity effect during semantic category-based induction

The premise monotonicity effect during category-based induction is a robust effect that occurs when generalization of a novel property shared by many cases is more likely than one shared by few cases. The timing of brain activity during this effect is unclear. Therefore, the event-related potentials (ERPs) underpinning this effect were measured by manipulating the premise sample size (single [S] vs. two [T]) in a semantic category-based induction task, with the conclusion categories either including the premise categories (congruent induction) or not (incongruent induction). The behavioral results replicated the premise monotonicity effect, and revealed that S arguments produced longer reaction times and more conservative response criteria than did T arguments. This suggests that the premise monotonicity effect was affected by both evidence accumulation speed and decision threshold. ERP results demonstrated that the premise monotonicity effect was reflected by two parameters during inductive decision: (1) S arguments elicited larger FN400 amplitudes than did T arguments under congruent induction, which was linked to reduced global similarity, decreased cognitive relevance, and attenuated conceptual fluency and (2) S arguments elicited larger sustained negativity (SN) in the 450-1050-ms time window than did T arguments, which is related to more inference-driven integration and interpretive processes. Our findings provide insight into the complex temporal course of the premise monotonicity effect during semantic category-based induction.

Inductive reasoning 2.0

Inductive reasoning entails using existing knowledge to make predictions about novel cases. The first part of this review summarizes key inductive phenomena and critically evaluates theories of induction. We highlight recent theoretical advances, with a special emphasis on the structured statistical approach, the importance of sampling assumptions in Bayesian models, and connectionist modeling. A number of new research directions in this field are identified including comparisons of inductive and deductive reasoning, the identification of common core processes in induction and memory tasks and induction involving category uncertainty. The implications of induction research for areas as diverse as complex decision-making and fear generalization are discussed. This article is categorized under: Psychology > Reasoning and Decision Making Psychology > Learning.

How do the hierarchical levels of premises affect category-based induction: diverging effects from the P300 and N400

Although a number of studies have explored the time course of category-based induction, little is known about how the hierarchical levels (superordinate, basic, subordinate) of premises affect category-based induction. The EEG data were recorded when nineteen healthy human participants were performing a simplified category-based induction task. The ERP results showed that: in the subordinate conclusion condition, the basic premise elicited a larger N400, versus the superordinate promise; in the basic conclusion condition, the superordinate promise elicited a larger P300 relative to both the basic premise and subordinate premise; in the superordinate conclusion condition, however, no difference was found between different promise. Furthermore, the process that reasoning from a higher level to a lower level evoked a larger P300, compared to it did in the reverse direction. The divergent evidence suggested that category-based induction at superordinate, basic, and subordinate levels might be affected by various factors, such as abstract level, direction, and distance between premise and conclusion, which yielded new insights into the neural underpinnings of category-based induction with different inductive strengths.

Probabilistic representation in syllogistic reasoning: A theory to integrate mental models and heuristics

This paper presents a new theory of syllogistic reasoning. The proposed model assumes there are probabilistic representations of given signature situations. Instead of conducting an exhaustive search, the model constructs an individual-based "logical" mental representation that expresses the most probable state of affairs, and derives a necessary conclusion that is not inconsistent with the model using heuristics based on informativeness. The model is a unification of previous influential models. Its descriptive validity has been evaluated against existing empirical data and two new experiments, and by qualitative analyses based on previous empirical findings, all of which supported the theory. The model's behavior is also consistent with findings in other areas, including working memory capacity. The results indicate that people assume the probabilities of all target events mentioned in a syllogism to be almost equal, which suggests links between syllogistic reasoning and other areas of cognition.

Imaging deductive reasoning and the new paradigm

There has been a great expansion of research into human reasoning at all of Marr’s explanatory levels. There is a tendency for this work to progress within a level largely ignoring the others which can lead to slippage between levels (Chater et al., 2003). It is argued that recent brain imaging research on deductive reasoning—implementational level—has largely ignored the new paradigm in reasoning—computational level (Over, 2009). Consequently, recent imaging results are reviewed with the focus on how they relate to the new paradigm. The imaging results are drawn primarily from a recent meta-analysis by Prado et al. (2011) but further imaging results are also reviewed where relevant. Three main observations are made. First, the main function of the core brain region identified is most likely elaborative, defeasible reasoning not deductive reasoning. Second, the subtraction methodology and the meta-analytic approach may remove all traces of content specific System 1 processes thought to underpin much human reasoning. Third, interpreting the function of the brain regions activated by a task depends on theories of the function that a task engages. When there are multiple interpretations of that function, interpreting what an active brain region is doing is not clear cut. It is concluded that there is a need to more tightly connect brain activation to function, which could be achieved using formalized computational level models and a parametric variation approach.