Article summary: Neural organization of speech production: A lesion-based study of error patterns in connected speech

“Neural organization of speech production: A lesion-based study of error patterns in connected speech”

Recently published in Cortex, 2019


Authors: Brielle C Stark, Alexandra Basilakos, Gregory Hickok, Chris Rorden, Leonardo Bonilha, Julius Fridriksson

doi: 10.1016/j.cortex.2019.02.029

Free version:

This study involved 120 speakers with and without aphasia, all of which had a left hemisphere stroke and were at least six months into their recovery.

Paraphasias, which are what we call single-word errors (e.g. saying “cat” when you meant to say “dog”) are typical in aphasia. The brain areas that, when damaged by a stroke, result in production of these paraphasias, remains disputed. Further, the field has often evaluated paraphasias made during object naming, i.e. naming pictures of objects. However, this type of task is not particularly ‘real’ to life. We typically produce more fluent, connected speech. Therefore, to contrast paraphasias made during naming, we also looked at paraphasias made while people described what they saw in pictures (‘picture description’).

In this study, we focused on four different types of paraphasias:

  • Neologistic: made-up words (e.g. “nickleflubber”)

  • Phonemic: real word (e.g. “mat” instead of “cat”) or non-word (e.g. “mickle” instead of “pickle”)

  • Semantically related: a real word that is related to its intended target (e.g. “dog” instead of “cat”)

  • Semantically unrelated: a real word that is unrelated to its intended target (e.g. “tree” instead of “plate”)

The purpose of this study was twofold:

  1. Evaluate brain damage that results in paraphasias made during naming and connected speech tasks

  2. Evaluate the distribution of different types of paraphasias made during each task

Participants and the Task

We acquired data from 120 speakers with and without aphasia, all of whom had experienced a stroke at least six months prior to study participation. We call this the ‘chronic’ period of stroke.

Naming Task: Participants were asked to name 175 pictures of objects to the best of their ability. The specific task that we used was called the Philadelphia Naming Test. 57 people had data from this test.

Connected Speech Task: Participants were shown three different picture scenes and asked to describe each picture for two minutes. We combined their speaking data across pictures. 63 people had data from this test.

Task responses were transcribed and coded by graduate students in our lab. Paraphasias were coded by type (see above, e.g. phonemic) in the software called CLAN (by Brian MacWhinney, freely available).

We then looked at how many paraphasias, of each type, were produced during each task.

Behavior Results


Here, we compared the number of each type of paraphasias made during the naming and the connected speech tasks for the group who had BOTH assessments (36 people).

We found that proportionally more semantically related paraphasias were made during the naming task, whereas more semantically unrelated paraphasias were made during the connected speech task (see figure, right). There was not a significant difference in phonemic or neologistic paraphasias made between tasks.

Brain Results


We acquired MRI scans on all participants. We then used software (NiiStat) to complete ‘voxelwise lesion symptom mapping,’ which evaluates the relationship between brain damage and behavior. In this case, we evaluated lesions that were associated for each type of paraphasias (e.g. neologistic, phonemic, semantically related, semantically unrelated) for each task (naming, connected speech).

We found that neologistic paraphasias (green in above figure) were associated with damage to dorsal stream areas in connected speech, but we did not find a significant association between lesion and neologistic paraphasias during naming.

Most interestingly, we found that semantically related errors and semantically unrelated errors were associated with damage to the left hemisphere temporal lobes during both naming and connected speech (red and purple, above).


We then evaluated how similar the lesion location was for semantically related and semantically unrelated paraphasias across tasks.

To the right, you’ll note that semantically related paraphasias and semantically unrelated paraphasias were associated with damage along the temporal lobes. The damage related to semantically related paraphasias was slightly more anterior than damage associated with semantically unrelated paraphasias.


NEURAL Research Lab attends the 49th Clinical Aphasiology Conference

Brielle Stark and Manaswita Dutta (PhD student) made their way to the 49th Clinical Aphasiology Conference, hosted in lovely Whitefish, Montana. Manaswita was a NIDCD fellow, meaning that she took part in several career development opportunities during the course of the conference. Her abstract (see below), which was for a roundtable discussion, was one of the highest rated abstracts at the conference.

Poster Presentations:

Stark, BC. “A comparison of three discourse elicitation methods in aphasia and age-matched adults: implications for language assessment and outcome“ [paper is forthcoming in AJSLP]

Kristinsson, S; Yourganov, G; Xiao, F; Bonilha, L; Stark, BC; Rorden, C; Basilakos, A; Fridriksson, J. “Functional Brain Activation by BDNF Genotype in Chronic Aphasia .”

Roundtable Discussion:

Dutta, M; Murray, L and Stark, BC. “Standardizing assessment of spoken discourse in aphasia: Directions for future research .”

  • Abstract



den Ouden, D; Stark, BC; Fridriksson, J. “Lesion-symptom mapping of verbs and morphosyntax in picture descriptions.”