Research in our lab seeks to understand the basic mechanisms of cognitive and perceptual processing, particularly in the domain of spoken language. We focus on spoken language because it lies at the intersection of basic perceptual and higher-level cognitive processing – drawing on both and elucidating the core principles that span all of cognitive function. Our research follows three core tenets:
1. Show me the model. A verbal account can be taken to predict (or not predict) just about any observed pattern, especially when the account posits complex interactions between multiple processes. We use computational models to concretely instantiate our verbal theories so they can be tested and used to make new predictions.
2. The same and not the same. No two people have exactly the same cognitive system. How individuals differ provides unique insights into cognitive processes. We study individual differences in their many and varied forms: differences among typical college students, differences due to normal development and aging, differences induced by neurological damage such as stroke, and atypical development such as autism spectrum disorders. (This tenet’s title is borrowed from Dr. Roald Hoffmann’s wonderful book, we hope he doesn’t mind.)3. Methods for the madness. When the only tool you have is a hammer, every problem looks like a nail. We try to improve and expand the set of tools available to cognitive neuroscientists and neuropsychologists. This includes statistical methods for analysis of time course data (GCA), a large web-based database to facilitate the use of cognitive neuropsychological methods in the study of language and cognition (MAPPD), and many tutorials.
Current projects apply these tenets to several inter-related research projects:
There are two kinds of things we know about (object) concepts: their features, which tell us taxonomic information about the object (that an apple is a fruit, that it is the same kind of thing as a pear or a peach, that it is round and edible, etc.), and the events or situations in which they participate, which tell us thematic information about the object (that apples grow on trees and ripen in the fall, that they can be baked into pies or made into cider, that they sometimes have worms inside, etc.). Our initial studies in this area found that feature-based similarity predicts the degree of activation during word recognition (Mirman & Magnuson, 2009b) and that distinctive semantic features (ones that are relatively unusual) play a particularly important role in processing of word meanings (Mirman & Magnuson, 2009a). We then found that taxonomic and thematic semantic knowledge have different time courses of activation (Kalenine, Mirman, Middleton, & Buxbaum, 2012), that individuals exhibit consistent differences in strength of taxonomic vs. thematic knowledge (Mirman & Graziano, 2012a). We also found evidence that these two kinds of knowledge are neuroanatomically distinct, with the anterior temporal lobe playing a particularly important role in taxonomic semantics and the temporo-parietal cortex playing a particularly important role in thematic semantics (Schwartz et al., 2011; Mirman & Graziano, 2012b). Our recent systematic review brought together behavioral, computational, and neuroscience evidence that taxonomic and thematic semantic systems are distinct and proposed neuro-computational principles that may drive this distinction (Mirman, Landrigan, & Britt, in press). We are delving deeper into this distinction to try to understand its functional basis and the reason for the apparent neuroanatomic specialization.
Identifying relationships between location of brain damage and cognitive deficits is a foundational method in cognitive neuroscience, tracing its history at least to the behavioral neurologists of the mid-19th century. Advances in neuroimaging technology have made possible finer-grained analyses at the level of individual voxels. In recent work, we combined voxel-based lesion-symptom mapping with factor analysis of extensive behavioral assessments to identify two major divisions within the language system: meaning vs. form and recognition vs. production, and their instantiation in the brain. Phonological form deficits were associated with lesions in peri-Sylvian regions, whereas semantic production and recognition deficits were associated with damage to the left anterior temporal lobe and white matter connectivity with frontal cortex, respectively (Mirman et al., 2015a). We also replicated these results using a multivariate lesion-symptom mapping technique based on support-vector regression (Mirman et al., 2015b). We are continuing to develop lesion-symptom mapping methods and to use them to answer basic and applied research questions on the neural bases of language and language disorders.
In our early research on word learning in neurologically intact young adults, we found that statistical segmentation facilitates word learning in behavioral experiments (Mirman, Magnuson, Graf Estes, & Dixon, 2008) and used a computational model to explain this effect and the trade-off between frequency and transitional probability statistics (Mirman, Graf Estes, & Magnuson, 2010). More recently, we have found that verbal short-term memory, but not word processing abilities, predicted ability to learn novel words in a group of individuals with aphasia (Peñaloza et al., 2016). We are continuing to study novel word learning in aphasia as a possible predictor of response to speech therapy treatments. And we are examining other behavioral, neurological, and biological predictors of recovery and the possibility that non-invasive brain stimulation can be used to measure or promote plasticity to enhance the effects of therapy.
Competition and cooperation among co-activated representations, and the role of cognitive control in spoken languageOne of the most widely-held basic principles of cognitive processing is that during a task (recognition, categorization, memory retrieval, etc.) multiple related or similar representations are activated in parallel. But, once activated, do these representations compete or cooperate? Both facilitative and inhibitory effects have been shown, but researchers have tended to focus on just one kind of effect at a time. We both facilitative effects both facilitative and inhibitory effects within the same task (Mirman & Magnuson, 2008; Mirman, 2011), and developed a simple interactive activation and competition model, which revealed that these reversals arise from a single computational principle: strongly active neighbors have a net inhibitory effect and weakly active neighbors have a net facilitative effect (Chen & Mirman, 2012). More than just fitting the behavioral data, the model also made a novel and counter-intuitive prediction that preview duration should induce a U-shaped pattern of neighborhood effects by modulating degree of neighbor activation, which we subsequently tested and confirmed (Chen & Mirman, 2015).
Real-world language use depends on other cognitive functions, particularly memory and cognitive control. We have found that at least some deficits in spoken word recognition in aphasia may be attributable to deficits in response selection (Mirman, Yee, Blumstein, & Magnuson, 2011) and that participants with aphasia have particular difficulty categorizing objects based on a single dimension (Lupyan & Mirman, 2013). We have also examined effects of item repetition (Mirman, Britt, & Chen, 2013), distinguished lexical and semantic competition (Britt, Ferrara, & Mirman, 2016), and investigated the role of frontal regions in resolving competition (Mirman & Graziano, 2013; Nozari, Mirman, & Thompson-Schill, 2016). These issues are related to the broad distinction between "storage" deficits, in which the lexical-semantic representations are thought to be impaired, and "access" deficits, in which the representations are thought to be intact but access to them is impaired, possibly due to deficits in related functions of memory and cognitive control. We reviewed the behavioral phenomena associated with access deficits in aphasia, the main theoretical perspectives on these deficits, and identified important open questions and promising future directions (Mirman & Britt, 2014).
Mirman, McClelland, & Holt, 2006) and delaying recognition of context-inappropriate stimuli (Mirman, McClelland, & Holt, 2005). Top-down effects are also modulated by attention – the more you attend to the context, the stronger its effect (Mirman, McClelland, Holt, & Magnuson, 2008). Such effects of context are a general property of cognition: an ambiguous color looks more brown when presented in context of “chocolate” (Kubat, Mirman, & Roy, 2009). This work was part of a larger effort to solidify interactive processing as a core principle of cognition and we have reviewed the empirical and computational evidence for interactive processing in several articles (McClelland, Mirman, & Holt, 2006; Mirman, 2008; McClelland, Mirman, Bolger, & Khaitan, 2014). A truly interactive system is not well-described by components (such as separate processing levels for words and speech sounds), so instead of focusing on components, we should focus on the dynamics that describe the interactions. We have used tools from statistical physics, computational biology, and other domains of “complexity science” to study cognition, perception, and action (Dixon, Holden, Mirman, & Stephen, 2012), particularly in the domain of eye movements (Stephen, Mirman, Magnuson, & Dixon, 2009; Stephen & Mirman, 2010; Kelty-Stephen & Mirman, 2013) and with a special interest in individual differences (Mirman, Irwin, & Stephen, 2012).