Date of Award
Doctor of Philosophy (PhD)
A novel word is rarely defined explicitly during the first encounter. With repeated exposure, a decontextualized meaning of the word is integrated into semantic memory. With the overarching goal of characterizing the functional neuroanatomy of semantic processing in young adults, we employed a contextual word learning paradigm, creating novel synonyms for common animal/artifact nouns that, along with additional real words, served as stimuli for the lexical-decision based functional MRI (fMRI) experiment. Young adults (n=28) were given two types of word learning training administered in multiple sessions spread out over three days. The first type of training provided perceptual form-only training to pseudoword (PW) stimuli using a PW-detection task. The second type of training assigned the meaning of common artifacts and animals to PWs using multiple sentences to allow contextual meaning acquisition, essentially creating novel synonyms. The underlying goals were twofold: 1) to test, using a behavioral semantic priming paradigm, the hypothesis that novel words acquired in adulthood get integrated into existing semantic networks (discussed in Chapter 2); and 2) to investigate the functional neuroanatomy of semantic processing in young adults, at the single word level, using the newly learned as well as previously known word stimuli as a conduit (discussed in Chapter 3).
As outlined in Chapter 2, in addition to the semantic priming test mentioned above, two additional behavioral tests were administered to assess word learning success. The first was a semantic memory test using a two-alternative sentence completion task. Participants demonstrated robust accuracy (~87%) in choosing the appropriate meaning-trained item to complete a novel sentence. Second, an old/new item recognition test was administered using both meaning and form trained stimuli (old) as well as novel foil PWs (new). Participants demonstrated: a) high discriminability between trained and novel PW stimuli. (d-prime=2.72); and b)faster reaction times and higher accuracy for meaning-trained items relative to perceptually-trained items, consistent with prior level-of-processing research. The results from the recognition and semantic memory tests confirmed that subjects could explicitly recognize trained items as well as demonstrate knowledge of the newly acquired synonymous meanings. Finally, using a lexical decision task, a semantic priming test assessed semantic integration using the novel trained items as primes for word targets that had no prior episodic association to the primes. Relative to perceptually trained primes, meaning-trained primes significantly facilitated lexical decision latencies for synonymous word targets. Taken together, the behavioral findings outlined above demonstrate that a contextual approach is effective in facilitating word learning in young adults. Words learned over a few experimental sessions were successfully retained in declarative memory, as demonstrated by behavioral performance in the semantic memory and recognition memory experiments. In addition, relative to perceptually-trained PWs, the newly meaning-trained PWs, when used as primes in a semantic priming test, facilitated lexical decisions for synonymous real words, with which the primes had no prior episodic association. The latter finding confirms our primary behavioral hypothesis that novel words acquired in adulthood are represented similarly, i.e. integrated in the same semantic memory representational network, as common words likely acquired early in the lifetime.
Chapter 3 outlines the findings from the fMRI experiment used to investigate the functional neuroanatomy of semantic processing using the newly learned as well as previously known words as stimuli in a lexical decision task. fMRI data were collected using a widely-spaced event-related design, allowing isolation of item-level hemodynamic responses. Two fMRI sessions were administered separated by 2-3 days, the 1st session conducted prior to, and the 2nd session following word-learning training. Using the same items as stimuli in the fMRI sessions conducted before and after behavioral training, facilitated a within-item analysis where each item effectively served as its own control. A set of stringent criteria, outlined below, were established a-priori describing characteristics expected from regions with a role in retrieving/processing meanings at the single word level. We expected a putative semantic processing region to exhibit: a) higher BOLD activity during the 1st fMRI session for real words relative to novel PWs; b) reduced BOLD activity for repeated real words presented in the 2nd fMRI session relative to levels seen in the 1st fMRI session; c) higher BOLD activity for meaning-trained PWs relative to novel PWs; d) higher BOLD activity for meaning-trained PWs relative to perceptually-trained PWs, e) higher BOLD activity for correctly identified meaning-trained PWs (hits) relative to their incorrect counterparts (misses). Given their previously documented associations with semantic processing, we expected to identify regions in left middle temporal gyrus (MTG) and left ventral inferior frontal gyrus (vIFG) to exhibit timecourses consistent with most of the semantic criteria outlined above.
Individual ANOVA contrasts, essentially targeting each of the criteria outlined above, were conducted at the voxelwise level. A fixed effects analysis based on 4 correct trial ANOVA contrasts (corresponding to criteria a-d, above) generated 81 regions of interest; and two individual error vs. correct trial ANOVA contrasts generated an additional 16 regions, for a total of 97 study-driven regions. Using region-level ANOVAs and qualitative timecourse examinations, the regions were probed for the presence of the effects outlined in the above criteria. To ensure a comprehensive analysis, additional regions were garnered from prior studies that have used a variety of tasks to target semantic processing. The literature-derived regions were subjected to similar ANOVAs and qualitative timecourse analysis as was conducted on the study-driven regions to examine if the regions exhibited effects outlined in the above criteria.
The above analysis resulted in three principal observations. First, we identified regions in the left parahippocamal gyrus (PHG) and left medial superior frontal cortex (mSFC) that, by satisfying essentially all the above criteria, demonstrated a role in semantic memory retrieval for recently acquired and previously known words. Second, despite strong expectations, regions in the left MTG and left vIFG failed to show activity in support of a role in semantic retrieval for the novel words. On the contrary, the profiles seen in the two said regions, namely a ‘word > novel PW’ and a word repetition suppression effect, were consistent with a role in semantic retrieval exclusively for the previously known words. The latter observation suggests that the novel words have yet to undergo adequate consolidation to engage, in addition to PHG and mSFC, canonical semantic regions such as left MTG.
Third, despite the potentially crucial distinctions noted in Chapter 3, left lateral/medial parietal regions implicated in episodic memory retrieval exhibited many similar properties as those outlined for PHG and mSFC above during retrieval of newly learned words. Crucially, instead of exhibiting repetition suppression for real words, as observed in PHG/mSFC, the parietal regions showed the opposite effect resembling the episodic ‘old>new’ retrieval success effect. The latter observation argues against a sematic role and in support of an episodic role consistent with previous literature. Taken together, these observations suggest that in addition to the role played by PHG/mSFC supporting semantic memory retrieval for the novel words, the parietal regions are also making significant contributions for memory retrieval of the novel words via complementary episodic processes.
Finally, using item-level timecourses derived from the 97 study-driven ROI, clustering algorithms were used to group regions with similar characteristics, with the goal of identifying a cluster corresponding to a putative semantic brain system. A number of clusters were identified containing regions with anatomical and functional correspondence to previously well-characterized systems. For instance, a cluster containing regions in left lateral parietal cortex, precuneus, and superior frontal cortex corresponding to a previously described episodic memory retrieval system (Nelson et al., 2010) was identified. Two additional clusters, corresponding to frontoparietal and cinguloopercular task control systems (Dosenbach et al., 2006, 2007) were also among the identified clusters. However, the clustering analysis did not identify a cluster of regions with semantic properties, such as PHG and mSFC noted above, that could potentially correspond with a semantic brain system.
The above outlined findings from the current study, juxtaposed with prior findings from the literature, were interpreted in the following manner. The two regions identified in the current study, i.e. left parahippocampal gyrus and medial superior frontal gyrus, constitute regions that are used for learning new words, and are also recruited during semantic retrieval of previously well-established meanings. In addition, the current results also suggest complementary episodic contributions to the word learning process from regions in left parietal/superior frontal cortex. The latter observation may imply strong episodic contributions to the observed behavioral semantic priming effects. A potential counter argument, i.e. in support of a semantic basis for the priming effects, is the shared recruitment, in a manner consistent with semantics, of PHG/mSFC by both novel and real word stimuli.
The left middle temporal gyrus, a region that the task-evoked and neuropsychological literature consistently associates with word-level semantic processing, was not recruited during memory retrieval of novel words, despite robust engagement by previously known word stimuli. Given their association with category-selective semantic deficits, as well as their role in conceptual/perceptual processing in healthy brains, the memory consolidation literature proposes regions in the lateral temporal lobes as potential neocortical loci for consolidated long-term memory. In the current setting, it is likely the case that the novel words have yet to be adequately consolidated to engage left MTG as did the previously known words.
Finally, the left vIFG exhibited similar characteristics as the left middle temporal gyrus, in that it was not recruited by the newly meaning trained stimuli, despite showing engagement by previously known words. Given that the region failed to appear in our primary contrasts, even those targeting real word stimuli, and its absence in other prior studies that have used similar lexical decision tasks as the current study, we have a slightly different interpretation for that region. The left vIFG is typically recruited in task settings that require controlled/strategic meaning retrieval, a process that may not be critical for adequate performance of the lexical decision task as employed in the current study.
Taken together, these findings suggest that a relatively small amount of word learning training is sufficient to create novel words that, in young adults, behaviorally resemble the semantic characteristics of well-known words. On the other hand, the fMRI findings, particularly the failure of the newly meaning-trained items to engage regions that are canonically responsive to single word meanings (e.g. middle temporal gyrus), may suggest a more protracted timecourse for the functional signature of novel words to resemble that of well-known words. That said, the fMRI findings identified brain regions (left PHG/mSFC) that, consistent with the memory consolidation literature, serve as the functional neuroanatomical “bridge” that connects the novel words to the eventual functional representational destination.
Chair and Committee
Bradley L Schlaggar
David Balota, Deanna Barch, Lori Markson, Steven Petersen
Nardos, Binyam, "Behavioral and fMRI-based Characterization of Cognitive Processes Supporting Learning and Retrieval of Memory for Words in Young Adults" (2015). Arts & Sciences Electronic Theses and Dissertations. 671.