Project 4

There are two overarching goals of Project 4. The first goal is to use theoretical and quantitative models of language processes and neural organization to improve aphasia diagnosis, treatment, and outcome prediction, which contributes directly to the clinical goals of the C-STAR program. The second goal is to use data generated by C-STAR to improve our models of the neurobiology of language, which then feeds back into the first goal. Under the current grant, we developed a new quantitative model of naming using a multinomial processing tree approach. The model has allowed us to increase our computational resolution and therefore diagnostic specificity of naming dysfunction: whereas typical clinical assessment provides a single measure of naming ability (% correct), our MPT model can pinpoint which sub-abilities (semantic, lexical-semantic, lexical-phonological, phonological/phonetic) are causing the dysfunction and which improve or not during treatment and recovery. We have also shown that these sub-abilities map onto different neural networks, which not only pushes forward our understanding of the neural basis of language but also provides additional validation for our MPT model. 

We have also broadened our theoretical scope during the current grant period by developing a new model of the cortical organization of syntax.  The model proposes a syntactic hub in the posterior middle temporal gyrus region, which codes hierarchical relations between morphemes and sits at the nexus of lexical-phonological networks in the superior temporal gyrus and semantic networks in the anterior temporal lobe and angular gyrus. This syntactic hub also interfaces with frontal networks, which play a key role in morpheme sequencing. Thus, the model breaks from traditional approaches by proposing overlapping but non-identical “syntactic” computation during sentence comprehension (posterior networks) versus production (fronto-posterior networks). This production-comprehension asymmetry fits the framework of our model of word comprehension and naming, providing a coherent model of the entire language network.   

This progress during the existing grant enables us to develop the project in two directions in the present proposal. One direction broadens the target for theoretical and experimental development to include not only word production but also sentence production, which will lay the groundwork for clinical research that includes a more complete picture of aphasia. In another direction, we make contact with recent clinical findings from the C-STAR project showing that extra-lesional findings such as connectivity and small vessel white matter disease explain some of the variance in aphasia severity and recovery potential. Our models provide a means to measure the computational implications of such extra-lesional findings, which will improve our functional anatomic maps and perhaps diagnostic/prognostic precision. 

Aim 1: Extend our quantitative and neural model of word production. Our previous work focused on building quantitative models of naming and using them to map the underlying neural networks. Here we propose to extend our MPT model to include sub-phonemic features to map the speech motor control hierarchy at the phonological versus sub-phonological levels. This will be clinically relevant to distinguishing “linguistic” from “motor” phonological disorders, which we expect will map to distinct hierarchically organized sensorimotor circuits. We also propose to extend our MPT model to integrate data from other word-level production tasks. We have evidence that different tasks differentially tax the underlying network and if integrated will paint a more complete picture of a patient’s impairment. 

Aim 2: Extend our theoretical scope to include syntax. A complete model of language organization and disruption must include syntax. We have developed and published a new functional-anatomical model of syntactic processing that we propose to test and develop. We will test specific predictions of the model using existing data to map distinct syntactic abilities (those underlying agrammatism versus paragrammatism) and develop new syntactic expressive and receptive tasks as a first step toward the development of a quantitative model of syntactic competence. This work will set the stage for the next generation of aphasia therapies targeting morphosyntax. 

Aim 3: Use quantitative models for treatment outcome evaluation and prediction. Clinical resources are limited and so knowing who is likely to benefit from treatment is invaluable to optimizing resource allocation. Standard metrics for evaluating treatment response rely on simple summary measures, such as percent correct (%C) on a given test. Yet, it is well-known that even simple tasks such as picture naming are complex and multi-stage, meaning that equivalent %C scores could have very different latent computational sources. Our MPT model was developed to measure latent abilities and therefore provides a better measurement of the state of system pre- and post-treatment. We have begun using MPT parameters as a metric for evaluating recovery and predicting response to treatment, finding that it out performs standard metrics. We propose to develop and test this new line of investigation.