Shared processes resolve competition within and between episodic and semantic memory: Evidence from patients with LIFG lesions

2 Semantic cognition is supported by two interactive components: semantic 3 representations and mechanisms that regulate retrieval (cf. ‘semantic control’). 4 Neuropsychological studies have revealed a clear dissociation between semantic and episodic 5 memory. This study explores if the same dissociation holds for control processes that act on 6 episodic and semantic memory, or whether both types of long-term memory are supported by 7 the same executive mechanisms. We addressed this question in a case-series of semantic 8 aphasic patients who had difficulty retrieving both verbal and non-verbal conceptual 9 information in an appropriate fashion following infarcts to left inferior frontal gyrus (LIFG). 10 We observed parallel deficits in semantic and episodic memory: (i) the patients’ difficulties 11 extended beyond verbal materials to include picture tasks in both domains; (ii) both types of 12 retrieval benefitted from cues designed to reduce the need for internal constraint; (iii) there 13 was little impairment of both semantic and episodic tasks when control demands were 14 minimised; (iv) there were similar effects of distractors across tasks. Episodic retrieval was 15 highly susceptible to false memories elicited by semantically-related distractors, and 16 confidence was inappropriately high in these circumstances. Semantic judgements were also 17 prone to contamination from recent events. These findings demonstrate that patients with 18 deregulated semantic cognition have comparable deficits in episodic retrieval. The results are 19 consistent with a role for LIFG in resolving competition within both episodic and semantic 20 memory, and also in biasing cognition towards task-relevant memory stores when episodic 21 and semantic representations do not promote the same response.


M A N U S C R I P T
A C C E P T E D ACCEPTED MANUSCRIPT 6 highlighted as 'lesion' (Seghier et al., 2008). A lesion map generated using this approach is 161 shown in Figure 1. In addition, we manually assessed lesions of individual patients by tracing 162 MRI scans onto standardized templates (Damasio & Damasio, 1989). All ten patients had 163 lesions affecting left posterior LIFG (see Figure 1B and Supplementary Table 2); in seven 164 cases, this damage extended to mid-to-anterior LIFG. Some lesions extended to inferior 165 parietal and/or posterior temporal regions, with less overlap between cases in these additional 166 regions. Three patients (P1, P3, P7) showed some degree of damage in the ATL. However, 167 ventral ATL, which has been implicated in conceptual representation across modalities   Associates test from WMS-III was impaired (see below).    The object use task (74 items), from Corbett et al., (2011), involved selecting an 252 object to accomplish a task (e.g., bash a nail into wood), with all items represented as 253 photographs. The target was either a canonical tool, normally used to complete the task (e.g.,

254
HAMMER), or an alternative non-canonical option (e.g., BRICK), presented among a set of five 255 unsuitable distractors. All patients were poorer at selecting non-canonical than canonical cut-off in the non-canonical condition, however this patient was impaired at the pictorial 261 version of the CCT.

262
The SA group showed strong sensitivity to all these control manipulations (Figure 2) 263 -i.e., more impaired comprehension of subordinate than dominant interpretations of 264 ambiguous words; sensitivity to cues and miscues; better comprehension with weak than 265 strong distractors and better retrieval of canonical than alternative object use. A composite 266 score reflecting each patient's deficits in semantic cognition was derived from the Camel and 267 Cactus Test and the three semantic control tasks described above using factor analysis.

268
Patients are ordered by this composite score in the graphs and tables below.

269
In the next section, we examined whether our participants with deregulated semantic         In each encoding block, eight word-pairs were presented consecutively on a screen instructions, the task was preceded by practice trials testing memory for four words pairs.

345
When the response was wrong, the correct answer was provided, and the practice procedure  In a subsequent experiment, we used the same task structure but eliminated semantic 352 links between the stimuli, using LSA scores of 0.5 or below [See Appendix Table 2 In a non-verbal episodic memory task, we presented black-and-white line drawings 358 of items during the training phase (mostly from Snodgrass & Vanderwart, 1980) and coloured 359 photographs of the same objects for recognition. These images were as dissimilar as possible 360 to prevent participants from relying on perceptual matching to identify the target. We again 361 manipulated semantic relatedness (related, unrelated) and episodic encoding strength (pairs 362 presented once or five times). Items on semantically-related trials were drawn from the same 363 semantic category (e.g., APPLE-ORANGE). Other aspects of the procedure followed the 364 description for Experiment 1 (see Fig. 4A for design and Appendix Table 3 for list of 365 stimuli).

367
Descriptive statistics are provided in Supplementary Table 5.  Figure 4E shows key results. are semantically-related -given the richness and distinctiveness of these stimuli.   .001], indicating that these participants were able to produce meaningful confidence ratings.

444
In Experiment 2, all probe-target pairs were semantically-unrelated; therefore, this  All participants were correct on both immediate and delayed recognition in at least two out of 483 three trials.

496
The target was presented alongside three distractors. On trials with episodic training, 497 these were the episodic distractor, a familiar distractor that was associated with a different   strongly-encoded yet irrelevant information, in both episodic and semantic tasks. This 597 necessity to constrain retrieval is reduced when the task provides strong cues to retrieval that 598 reduce competition and the need to internally shape retrieval.

599
Our findings have important implications for neuroscientific accounts of memory   Grey matter, white matter and CSF were segmented and changes from the healthy control brains were highlighted as 'lesion' using automated methods (Seghier, Ramlackhansingh, Crinion, Leff, & Price, 2008).
Colour bar indicates amount of overlap from 1 to 10 patients.

Semantic control network
Episodic network Overlap semantic + episodic