Semantic Memory, Traumatic Brain Injury, and the Iceberg Effect: What Deficits May Lie Below the Surface?

Abstract

Purpose:

The purpose of this viewpoint was to advocate for increased study of semantic memory ability in traumatic brain injury (TBI).

Method:

We review modern conceptualizations of semantic memory and its neural correlates and discuss how common neuroanatomical and cognitive deficits in TBI place this population at an increased risk for semantic disruption. Building on discussions at the 2024 International Cognitive-Communication Disorders Conference, we offer possible explanations for how these disruptions may have been overlooked by our field and offer examples of how semantic memory has been studied in other populations as well as how this work may apply to TBI research.

Result:

Semantic memory is critical for academic, vocational, and interpersonal outcomes. Yet, little is known about semantic memory in TBI beyond naming ability. By examining only surface forms of semantic memory, we may be missing a deeper disruption in semantic structure.

Conclusion:

More in-depth examination of semantic memory promises to uncover underlying mechanisms of cognitive-communication disorders and new opportunities to develop more sensitive clinical measures of semantic memory impairment.

Conservative estimates suggest that, by adulthood, the average individual has acquired 12.5 million bits of information, the majority of which is lexical-semantic knowledge (Mollica & Piantadosi, 2019). Semantic memory, our knowledge of words, concepts, and general facts about the world, or these millions of bits of information, is the basis for nearly all human activity (Binder & Desai, 2011). We depend critically on semantic memory for communication and success in our academic, vocational, and interpersonal pursuits. Notably, academic, vocational, and interpersonal domains are all areas where individuals with traumatic brain injury (TBI) are at risk for negative outcomes (Ponsford et al., 1995; Rassovsky et al., 2015; Whiteneck et al., 2004; Willmott et al., 2014). Yet, in the area of TBI, we seldom examine semantic memory beyond confrontational naming ability. Consequently, we do not know if individuals with TBI have deeper, more extensive disruptions in semantic memory beyond naming or if such deficits are a source of these negative long-term outcomes.

At the 2024 International Cognitive-Communication Disorders Conference (ICCDC), we argued that common cognitive and neuroanatomical deficits put individuals with TBI at an increased risk for a constellation of semantic memory impairments (McCurdy & Duff, 2024). We invoked the “iceberg effect” metaphor to suggest that by examining only surface forms of semantic memory (e.g., naming), we may be missing a deeper disruption in semantic structure in TBI. There was considerable discussion about the importance of semantic memory for a range of postinjury outcomes, how semantic memory is measured in other disciplines, and the advantages of more systematic investigation of semantic memory for characterizing cognitive-communication impairments and functional outcomes in TBI.

In the rest of this viewpoint, we revisit and build on the discussions at the 2024 ICCDC to define semantic memory and illustrate some examples of how the depth and richness of semantic memory has been considered across fields. We highlight the overlap between the neural and cognitive correlates of semantic memory and the common neuroanatomical and cognitive deficits in TBI. Finally, we describe a research program that revealed deficits in remote semantic knowledge by using approaches that examined the depth and richness of semantic memory in a population with acquired brain injury¹ (i.e., adults with hippocampal amnesia), for whom semantic memory status had long been declared intact, in part, based on performance on confrontational naming tasks. We believe such an approach in TBI may reveal similar, if not more extensive, disruptions in TBI given its more diffuse profile of neural and cognitive deficits.

Before we begin, we should acknowledge that the study of words and concepts has existed for centuries and that the study of semantic memory is a highly developed interdisciplinary area of scientific and clinical inquiry with many theories and approaches across development and clinical populations. A comprehensive review of semantic memory theories and methods is beyond the scope and goals of this article. For readers with broader interest in semantic memory, we recommend the following works: Barsalou (1999), Binder and Desai (2011), Duff et al. (2020), Kenett and Faust (2019), Kumar (2021), Ralph et al. (2017), Reilly et al. (2016), Rubin et al. (2014), and Yee and Thompson-Schill (2016).

Our focus here is to raise the possibility that we have missed disruptions in semantic memory in TBI and that such deficits may be linked to, and are an opportunity to improve, negative long-term outcomes in academic, vocational, and interpersonal spheres. Our goal is to highlight the lack of systematic investigation on the status of semantic memory in TBI despite risk factors and offer a preliminary road map for how we might begin to investigate the status of semantic memory in TBI beyond naming. We believe that the empirical study of semantic memory in TBI may advance the characterization of cognitive-communication impairment and provide new opportunities to improve long-term functional outcomes. We hope this viewpoint can extend the 2024 ICCDC discussion on semantic memory in TBI to our broader community of researchers and clinicians who work to improve the basic science and clinical practice of post-TBI outcomes.

What Is Semantic Memory?

Philosophers have studied concepts and meaning for centuries (e.g., Aristotle, Plato, Locke), but the modern empirical study of semantic memory, and its definition, is often linked to Endel Tulving (1927–2023). In a seminal work, Tulving proposed a distinction in memory between episodic memory (autobiographical knowledge of the events of one's life) and semantic memory (general world knowledge of words, concepts, and facts; Tulving, 1972). In his original proposal, Tulving (1972) defined semantic memory as

necessary for the use of language. It is a mental thesaurus, organized knowledge a person possesses about words and other verbal symbols, their meaning, and referents, about the relations among them, and about the rules, formulas, and algorithms for the manipulation of these symbols, concepts and relations. (p. 384)

Tulving's distinction between forms of memory served as the foundation for decades of theoretical and experimental work in psychology and cognitive neuroscience. Despite Tulving's broad definition of semantic memory, much of the subsequent work, including much of Tulving's own work on semantic memory in amnesia, focused narrowly on the one-to-one relation between a word and its meaning (see Duff et al., 2020, for a review). However, other disciplines (e.g., psycholinguistics, semiotics, cognitive science) have captured the depth and breadth of semantic memory, viewing it as a flexible, (re)constructive, relational, and multimodal system (Duff et al., 2020; Reilly et al., 2016; Rogers et al., 2004). From these disciplines comes a set of theories for conceptualizing the depth and breadth of semantic memory and sensitive methods for its empirical study. In the rest of this section, we highlight some of those conceptualizations and approaches.

Semantic memory is a system of relations. These relations, and the information they connect, are stored in a dynamic network that changes over time with experience as new relations are acquired and as existing relations are retrieved (Binder & Desai, 2011; Cohen & Eichenbaum, 1993; Ralph et al., 2017; Reilly et al., 2016; Rogers & McClelland, 2004; Yee & Thompson-Schill, 2016). There are relations within a word or concept referred to as its semantic richness, that is, the amount of information contained within or associated with a word or concept. The richness of a word or concept can influence its memorability (Hargreaves et al., 2012) and the speed and accuracy of behavioral responses (e.g., greater semantic richness is associated with faster and more accurate naming, lexical decision, and categorization; Grondin et al., 2009; Hargreaves & Pexman, 2012; Pexman et al., 2002, 2003). Richness can also be represented by the amount of sensory and perceptual features associated with a particular word or concept. For example, words that are higher in imageability (can readily generate a mental image) and concreteness (can be imagined with the senses) are typically processed more quickly; it is easier to retrieve the word “apple”—something that can be seen, touched, and tasted—than it is to retrieve the word “freedom”—a more abstract concept (Bennett et al., 2011).

Semantic memory contains knowledge about the relations among words including knowledge of words that have multiple meanings or senses (e.g., “bank” can mean a place where money is stored, the edge of a river, and a type of basketball shot), words that have similar meanings (e.g., synonyms), and words that are more or less likely to co-occur with other words (e.g., collocation) in the ambient language (Christiansen & Arnon, 2017; Günther et al., 2019; Jamieson et al., 2018; Landauer & Dumais, 1997). The nature of these relations can also reveal information about the organization of the semantic memory network. For example, words with denser semantic neighborhoods—or words that are associated with many different words or concepts—are processed more quickly in naming, lexical retrieval, and lexical decision tasks (e.g., it is easier to retrieve the word “nurse” after viewing the word “doctor” than it would be having just viewed the word “grass”; Hargreaves & Pexman, 2012; Jones et al., 2006; Rodd et al., 2002; Yap et al., 2012).

A hallmark feature of semantic memory is its compositionality. While semantic knowledge is represented as the relations among semantic elements (e.g., the relation between the orthographic form of a word and its meaning; a cactus is a desert plant), the individual representation of the elements is maintained. That is, these relations are not fused or unitized but rather can be retrieved independently and combined in familiar and novel ways. This compositionality supports a range of phenomena including the generativity of language as well as linguistic and cognitive creativity (Corballis, 2017; Mednick, 1962). For example, in conceptual combination, speakers leverage the relations among lexical items to create new concepts and meaning from words and concepts in pre-existing knowledge stores (e.g., cactus–carpet—these words can be processed individually or as an integrated concept, a “cactus carpet”; Keane et al., 2020; Lucas et al., 2019). Metaphor production and comprehension are other examples of compositionality. To generate and comprehend metaphors (e.g., “my job is a jail”), language users create or identify relations between the metaphor topic (“job”) and vehicle (“jail”), which requires the rapid processing of novel relations between seemingly disparate lexical items (Lakoff & Johnson, 1980). Compositionality supports creative uses of language to communicate ideas that go beyond the literal meanings of words themselves (e.g., word play, humor, puns, metaphor, sarcasm), combining and integrating multidimensional meaning to communicate ideas that might otherwise be inexpressible and to fill gaps in the lexicon by extending existing words to describe novel categories and concepts (Li et al., 2021).

Semantic memory is the basis for nearly all human activity (Binder & Desai, 2011). As these examples above highlight, semantic memory is critical for language production and comprehension and for conveying complex ideas where multiple relations and meanings must be generated and integrated to meet a range of everyday communicative goals. Semantic memory is also critical for creativity (Duff et al., 2013; Kenett & Beaty, 2023; Ovando-Tellez et al., 2022); the ability to recall the past and imagine our future (Irish & Piguet, 2013); and learning and reasoning by analogy, considered by some to be the core of cognition (Hofstadter, 2001). In these ways, we use semantic memory to create, represent, and extract meaning as we navigate our most fundamental interactions with the environment and each other (Duff et al., 2020; Reilly et al., 2016; Rogers et al., 2004).

Yet, we know very little about the integrity of the semantic memory system following TBI. Despite the depth, breadth, and richness of semantic memory, we seldom examine semantic memory beyond performance on confrontational naming tasks. Consequently, this lack of knowledge limits the full characterization of cognitive-communication impairments post-TBI. We are particularly interested in the possibility that semantic memory deficits may be linked to poor long-term outcomes in individuals with TBI. For example, semantic memory is critical for communication and success in our academic, vocational, and interpersonal pursuits, and these are all areas where individuals with TBI are at risk for negative outcomes (Ponsford et al., 2014; Rassovsky et al., 2015; Whiteneck et al., 2004; Willmott et al., 2014). Roozenbeek et al. noted that despite tremendous progress in reducing mortality after brain injury, there has been no parallel reduction in brain injury–related disability (Roozenbeek et al., 2013). This lack of progression in reducing brain injury–related disability necessitates new directions in basic and clinical TBI research to improve long-term outcomes. We believe examination of the status of semantic memory may offer such a novel step.

In the rest of this article, we argue for the increased study of semantic memory in TBI. In the next section, we will briefly review the neural and cognitive correlates of semantic memory and highlight the overlap with common patterns of neuroanatomical and cognitive deficits in TBI. We propose that the common sequelae of TBI put this population at an increased risk for a range of semantic memory impairments.

Common Sequelae of TBI Heighten the Risk of Semantic Memory Impairment

Neuroanatomical Deficits in TBI Predict Semantic Memory Impairment

Semantic memory relies on a network of distributed and interconnected neural areas working in concert to acquire, process, maintain, and retrieve semantic knowledge. Prominent theories suggest that semantic memory is organized by sensorimotor modality (e.g., auditory knowledge, motor knowledge, visual knowledge) in modality-specific brain regions (e.g., auditory, motor, visual, sensory, and association cortices; Barsalou, 1999; Binder & Desai, 2011; A. R. Damasio, 1989; Lakoff & Johnson, 1999; Pulvermüller, 1999, 2018). For example, the relational knowledge of a “hammer” can include auditory features (e.g., the sound it makes when it strikes a surface), motor features (e.g., movements associated with swinging or prying), and visual features (e.g., its shape and color). The neural representations of these features are stored in the sensory and motor areas associated with their initial processing. Association cortices support relations among language features such as the arbitrary relation between the label “hammer” and its meaning as well as its relation to, and co-occurrence with, other words (e.g., nail, mallet, gavel, sledge hammer, jack hammer, hammer of justice, Thor's hammer, to fight hammer and tongs). Thus, semantic memories are composed of individual elements from multiple modality-specific brain regions.

As noted above, a hallmark feature of semantic memory is its compositionality. The relational features of a hammer are shared with other semantic concepts and can be retrieved independently and combined in novel ways to create, update, and expand semantic memories with experience. For instance, the motor features of a hammer (e.g., swinging) are shared with the semantic representation of an axe, baseball bat, or flyswatter. Hammer can belong to a broader semantic category (e.g., tools) or be part of a schema (e.g., objects you would find in a hardware store). The demands on compositionality necessitate a reliance on white matter to connect across these gray matter structures (Conner et al., 2018; Gonzalez Alam et al., 2024; Han et al., 2013; Hula et al., 2020; Nugiel et al., 2016; Sierpowska et al., 2019).

While the neural correlates of semantic memory are distributed, there are key hubs. Damage to any one of them can produce a disruption in semantic memory impairment, ranging from mild to profound, depending on locus and severity of damage. The pathophysiology of TBI increases the potential for damage across various neural correlates of semantic memory. Indeed, immediate mechanisms of injury such as coup–contrecoup lesions and rotational shearing, secondary sequelae such as seizure activity, ischemia and hypoxia, and lingering tertiary degenerative effects give TBI a long pathological reach (Gentry et al., 1988; Palacios et al., 2013). In the rest of this section, we look more closely at the specific brain regions that contribute to semantic memory, and their functions, with special attention to those most frequently damaged following TBI.

The medial prefrontal cortex (mPFC) contributes to semantic memory by developing schema (general event knowledge, like what typically occurs at a birthday party), selecting context-appropriate memory representations, and integrating new information into the existing semantic memory structure (Preston & Eichenbaum, 2013; Schlichting & Preston, 2015). Focal damage to the mPFC has been shown to impair functions that draw on semantic memory representations, such as inferential reasoning (DeVito et al., 2010; Koscik & Tranel, 2012), attachment of emotional valence to words and concepts (Heberlein et al., 2008), and selection of appropriate semantic information from a set of competitors (Binder et al., 2009; Robinson et al., 1998). Frontal lobe damage is common in TBI, often resulting from motor vehicle accidents and falls, and is frequently associated with difficulties in attention, executive functions, and emotional regulation (Ylvisaker, 1992; Ylvisaker & Feeney, 1998). While executive and attentional processes have been implicated in naming task performance (Higby et al., 2019; Roelofs, 2007; Shao et al., 2012), the impact of frontal lobe damage from TBI on the broader semantic memory system remains underexplored.

The temporal pole and ventrolateral portions of the temporal cortical surface play important roles in supporting semantic memory processing (Binder et al., 2009; Vigneau et al., 2006; S. M. Wilson et al., 2018). Focal damage to the temporal pole produces deficits in object recognition and naming (H. Damasio et al., 2004; Grabowski et al., 2001; Tsapkini et al., 2011). Damage to the middle temporal gyrus is associated with semantic and language comprehension deficits (Dronkers et al., 2004; Hillis & Caramazza, 1991; Kertesz et al., 1993). In semantic dementia, progressive degeneration of the anterior and ventrolateral temporal lobes produces a gradual loss of semantic features, for example, losing the knowledge that bananas are yellow or that zebras have stripes (Jefferies & Lambon Ralph, 2006; Lambon Ralph et al., 2007). The temporal poles are vulnerable to lacerations and lesions following TBI due to bony protuberances of the cranial vault, but additionally along the ventral and lateral temporal surfaces. Given that temporal lobe damage across multiple disorders can yield semantic memory deficits, we have reason to suspect this would be the case following damage in TBI as well.

Within the medial temporal lobes, the hippocampus is another critical structure for semantic memory. The hippocampus binds the arbitrary relations between the elements of an experience (e.g., the relations between the phonological and orthographic forms of “cactus” and its physical referent and meaning) into durable representations and the flexible expression of representations across novel settings (i.e., generalization; Cohen & Eichenbaum, 1993; Eichenbaum & Cohen, 2001). Bilateral hippocampal damage produces deficits in creating new semantic memories (Gabrieli et al., 1988; Hamann & Squire, 1995; Postle & Corkin, 1998). Recent lesion studies also implicate the hippocampus in maintaining semantic memories over time (Klooster & Duff, 2015; Klooster et al., 2020), using semantic representations during language use (Hilverman et al., 2017), imagining the future (Irish, Eyre, et al., 2016), and using creative language (Duff et al., 2009, 2013; Keane et al., 2020; Warren et al., 2016). Damage to the hippocampus is a pervasive consequence of TBI. The hippocampus is susceptible to common sequelae of TBI including hypoxia, anoxia, and seizure activity (Palacios et al., 2011; Sharp et al., 2014; Tate & Bigler, 2000). The preponderance of evidence that the hippocampus plays a significant role in semantic memory and its documented susceptibility to the effects of TBI strongly indicate the possibility of semantic memory disruption.

The parietal lobe and, particularly, the angular gyrus are other significant contributors to the semantic memory network. The angular gyrus is a crossroads of multiple modal processing pathways, processing and combining information from these streams to support lexical semantics via high-level concept formation and integration (Binder & Desai, 2011; Binder et al., 2009; Geschwind, 1965; Kim et al., 2011; Price et al., 2015; Seghier, 2013; although see Humphreys et al., 2021). Lesions to the angular gyrus can produce several disruptions in semantic processing, including naming difficulties (Benson, 1979), transcortical sensory aphasia (Boatman et al., 2000; Kertesz et al., 1982), and difficulty understanding word meaning in sentences (Dronkers et al., 2004). While not as frequently and directly damaged as the frontal and temporal lobes, the parietal lobe is also subject to both the primary and secondary effects of TBI (Gentry et al., 1988; Irimia et al., 2017; Palacios et al., 2011; Rauchman et al., 2022), opening yet another cortical route to semantic memory impairment in TBI.

Finally, several white matter tracts have been identified in relaying semantic representations throughout the cortex, including the uncinate fasciculus, inferior fronto-occipital fasciculus, inferior longitudinal fasciculus, and corpus callosum. Gross damage to these pathways has been shown to produce a number of frank semantic impairments, such as difficulty with object recognition and naming, semantic substitutions (paraphasias) during language use, and difficulty identifying semantic relationships (Almairac et al., 2015; de Zubicaray et al., 2011; Gonzalez Alam et al., 2024; Mandonnet et al., 2007; Nugiel et al., 2016; Schulte & Müller-Oehring, 2010; Von Der Heide et al., 2013). TBI is considered a disorder of white matter connectivity. Its defining pathophysiology, traumatic axonal injury, is a mix of interrelated pathologies that unfold progressively over time, degrading or destroying white matter connections from initial injury to days, months, and years afterwards (Adams et al., 2017; Povlishock & Christman, 1995; Tomaiuolo et al., 2012). While all white matter tracts in the brain are vulnerable to micro- and macrostructural damage in TBI, long-range associative tracts and commissural fibers are particularly susceptible to its pathophysiology (Hayes et al., 2016). Diffusion tractography studies have identified damage in TBI to many of these major white matter tracts critical to semantic memory (Kinnunen et al., 2011; Kraus et al., 2007; Palacios et al., 2011, 2012).

In summary, TBI results in multiple pathologies that can affect nearly all white and gray matter structures in the brain. Semantic memory hubs in the frontal, temporal, and parietal lobes are especially vulnerable to the sequelae of TBI, and damage to these areas in other acquired and neurodegenerative brain injuries produces varying degrees of disruption in semantic memory. Taken together, the pathophysiological evidence suggests a significant chance of disruption to the semantic memory system following TBI. We next consider common cognitive disruptions in TBI that provide strong evidence to suspect some degree of semantic disruption.

Episodic Memory Deficits in TBI Predict Semantic Memory Deficits

Since Tulving's (1972) proposed distinction between episodic memory and semantic memory, considerable effort has been directed toward understanding their functional divisions and neuroanatomical correlates. The strongest interpretation of this distinction was that episodic memory and semantic memory were functionally and anatomically distinct and that one system could be impaired independent of the other (Kinsbourne & Wood, 1975). However, another proposal was that semantic memory and episodic memory are part of a unitary memory system (the declarative memory system), have shared neuroanatomical correlates, and are impaired in tandem (Cohen, 1984; Squire & Zola, 1996).

More than 50 years since Tulving proposed the episodic–semantic distinction, there are converging lines of research suggesting that, rather than discrete and isolated systems, episodic memory and semantic memory are interdependent and share largely overlapping neural substrates and operating characteristics (e.g., Cohen & Eichenbaum, 1993; Greenberg & Verfaellie, 2010; Irish & Vatansever, 2020; see Duff et al., 2020, for a review). Episodic memory deficits, secondary to damage to the medial temporal lobe and hippocampus, are a hallmark feature of several neurological disorders. Evidence from across multiple neurological disorders now suggests that deficits in episodic and semantic memory are often concomitant. The implication of this work is that the presence of episodic memory deficits is predictive of co-occurring deficits in semantic memory. In the rest of this section, we provided examples from the literature that point to the interdependence of episodic and semantic memory.

Recent research on Alzheimer's disease and the variant of frontotemporal dementia known as semantic dementia, two disorders once considered models for studying episodic and semantic memory separately early in disease progression, now provides evidence for their interdependence. Based on neuropsychological evidence, initial stages of Alzheimer's disease seemed to present with impaired episodic memory as the primary symptom, with other cognitive functions, including semantic memory, relatively or entirely spared (Almkvist, 1996; Greene et al., 1995; Hodges & Patterson, 2007; Perry et al., 2000; Perry & Hodges, 1999). This was further supported by histological and imaging studies showing early hippocampal pathology with a largely spared neocortex (Braak & Braak, 1991; Convit et al., 1993; McDonald et al., 2009). Conversely, semantic dementia seemed to present with the opposite deficit profile, where early stages of the disease showed semantic memory deficits in the face of seemingly spared episodic memory, supported by neuroimaging studies that identified early degeneration of the anterior temporal lobes in the context of a relatively intact hippocampal complex (Graham et al., 2000; Hodges & Patterson, 2007; Irish, Bunk, et al., 2016). Using more fine-grained behavioral (e.g., examination of episodic memory through creative future thinking or “prospection” tasks) and more powerful neuroimaging (e.g., hippocampal subfield analyses) techniques, deficits in semantic and episodic memory with a shared hippocampal pathology have been detected in semantic dementia and even in early stages of Alzheimer's disease (Hornberger & Piguet, 2012; Huang et al., 2020; Sabuncu et al., 2011; Verma & Howard, 2012; Vogel et al., 2005).

Perhaps the strongest evidence for the interdependence of episodic memory and semantic memory comes from patients with circumscribed hippocampal damage and profound episodic memory impairments. (Note that we will examine this population and their semantic memory deficits in more detail below.) The historical view was that hippocampal pathology affected the acquisition of new episodic memory, leaving remote episodic and remote semantic memory intact. Evidence at the time also suggested there were contexts in which new semantic memory could be intact following hippocampal damage. With more sensitive methods and controlled study designs, striking deficits are observed in the acquisition of both episodic and semantic memory in individuals with hippocampal amnesia (Gabrieli et al., 1988; Hamann & Squire, 1995; Postle & Corkin, 1998; Warren & Duff, 2019), suggesting that deficits in episodic memory and semantic memory are concomitant. Perhaps most striking is evidence suggesting that the hippocampus plays a long-term role in maintaining the detail and richness of remote episodic and semantic memory over time through reconsolidation processes (Grilli & Verfaellie, 2014; Klooster & Duff, 2015; Nadel & Moscovitch, 1997; Rosenbaum et al., 2008, 2009; St-Laurent et al., 2014; Verfaellie et al., 2014). Furthermore, deficits in semantic memory retrieval have been linked to impaired episodic memory (Greenberg et al., 2009; Sheldon et al., 2013), consistent with neuroimaging studies pointing to shared hippocampal involvement in the retrieval of semantic and episodic memory in noninjured individuals (Burianova & Grady, 2007; Burianova et al., 2010; L. Ryan et al., 2010).

Putting aside the specific definitions of episodic (autobiographical knowledge of one's life and daily events) and semantic (knowledge of words, concepts, and general facts about the word) memory for a moment, these forms of memory have a lot in common. Like semantic memory, episodic memory is a system of relations (Cohen & Eichenbaum, 1993; J. D. Ryan et al., 2000). An episodic memory contains the objects, people, and places of an experience and the temporal, spatial, and perceptual relations among them. Compositionality is a hallmark feature of episodic and semantic memory (Rubin & Cohen, 2017). Indeed, individuals not only can retrieve individual components of an episodic memory for recollection but can also combine and recombine episodic experiences to imagine the future or reimagine the past. Finally, there is significant overlap in the neural substrates of semantic and episodic memory (Binder et al., 2009; Irish & Vatansever, 2020). Like semantic memory, the hippocampus plays a critical role in the acquisition of new episodic memory, and the constituent features of an episode are stored in the sensory and motor areas associated with their initial processing.

Taken all together, the common neuropathological and cognitive sequelae of TBI put individuals with TBI at risk for disruptions in semantic memory. The neural correlates of semantic memory are particularly vulnerable to the pathophysiological impacts of TBI. The diffuse pattern of cognitive deficits common to TBI also puts this population at risk for semantic memory deficits. Of particular interest is recent work pointing to the interdependence of episodic and semantic memory. Given that episodic memory impairments are among the most reported deficits and the most frequent rehabilitation targets for individuals with TBI (Cicerone et al., 2011; Palacios et al., 2013; Vakil, 2005; B. A. Wilson, 1998), the evidence suggests that semantic memory deficits could be comorbid in TBI. Yet, despite these risk factors, semantic memory is underexamined in TBI.

Semantic Memory, TBI, and the Iceberg Effect

The iceberg effect is a commonly evoked metaphor to describe situations where the observable manifestations of a phenomenon (the visible tip of the iceberg above the water) mask the depth and complexity of the processes associated with it (the hidden body of the iceberg beneath the water's surface). As applied to semantic memory in TBI, we view the tip of the iceberg as representing the portion of semantic memory (the pairing of a word to its referent) that is typically examined in research and clinic through confrontation naming assessments, for example, the Boston Naming Test (BNT). Yet, as we have reviewed above, semantic memory has considerable depth and richness not captured by surface level assessments; it includes all the knowledge of the concepts that underlie the words we know and the relations among them. This vast memory store represents most of the iceberg that lies beneath the surface. In Figure 1, we depict this example with the word “book” at the top of the iceberg, while below the surface, we see the extensive amount of associated semantic information including the many features (pages, fiction, read, bookshelf) of the most common definition of “book,” a written or printed work. Other information below the surface not depicted here includes the multiple meanings and uses of the word “book” including to schedule something on a particular date (e.g., to book a flight), to run quickly (e.g., to book it), the Good Book (to refer to the Bible), a bookworm (a person who loves to read), to hit the books (to study), or to cook the books (to alter facts or figures dishonestly or illegally).

Figure 1.

The iceberg effect for the word “book.”

While the historical and practical reasons for the lack of systematic investigation of semantic memory in TBI are likely multifactorial, we believe there are several key reasons we have been stuck at the tip of the iceberg. First, although we have cataloged decades of widespread evidence of pathological risk factors for semantic memory deficits following TBI, it is important to note that much of the most compelling evidence for shared neural correlates and behavioral characteristics, and the subsequent parallel deficits in episodic and semantic memory, has only recently accumulated in a unified perspective (e.g., De Brigard et al., 2022; Duff et al., 2020; Renoult et al., 2019; although we also note this shared dependence was proposed decades earlier: Cohen, 1984; Cohen & Eichenbaum, 1993; Gabrieli et al., 1988). Even when strong scientific evidence exists, uptake can be slow. For example, the process of translation and implementation of evidence into clinical practice is estimated to take an average of 17 years (Green et al., 2009). As a field, we are likely just entering the integration phase in this arc where basic semantic memory research methodologies can be applied to the clinical research and clinical management of TBI. Second, there are issues with the sensitivity of existing clinical tools for examining semantic memory in TBI. The most common assessments used are confrontation naming batteries designed for detection of frank aphasia and whose psychometric properties have been questioned (i.e., the BNT; Harry & Crowe, 2014; Hawkins & Bender, 2002; Himmanen et al., 2003; Pedraza et al., 2009). This has made it difficult for clinicians and researchers to identify word retrieval difficulties commensurate with the lived experience of individuals with TBI (Elbourn et al., 2024; Ponsford et al., 2014; Ylvisaker, 1992). That is, individuals with TBI routinely perform well on standardized tests of naming despite reporting difficulty and embarrassment in their daily interactions. A potential consequence of not having tools to capture these lived experiences is that we have underestimated the significance of the impairments at the tip of the iceberg, leading us to focus on more visible deficits, rather than investigating semantic memory more deeply. Finally, we lack feasible experimental paradigms to study the breadth and depth of semantic memory in clinical populations. In the memory literature, there was an early explosion of interest in studying episodic memory that yielded theoretical and methodological developments that eventually translated to clinical populations. That same level of interest and study in semantic memory has lagged, resulting in fewer significant theoretical and methodological advances until relatively recently (see Duff et al., 2020, for a review).

We believe there are compelling reasons to suspect deep disruptions in semantic memory structure and processes in TBI and that this is the right time to systematically examine semantic memory in TBI. We note that deep disruption may not mean severe impairment. Clinicians working in mild TBI are familiar with individuals who have subtle deficits (or no deficits at all according to standardized assessments) yet experience significant difficulty integrating back into their work and relationships. By moving away from the metaphorical “tip of the iceberg,” we may uncover yet unidentified deficits, which could lead to improvement in detection, characterization, and eventual treatment of semantic disruptions in TBI. Indeed, semantic memory ability is critical for communication and academic, vocational, and interpersonal pursuits. If there are significant or even subtle deficits below the surface, semantic memory may provide a new and powerful means for improving long-term outcomes. In the next section, we describe findings from a research program that sought to move beyond the assessment of superficial semantic knowledge to examine the depth and breadth of semantic memory in a population with acquired brain injury, adults with hippocampal amnesia.

Traversing Leagues and Fathoms: Lessons From Beneath the Surface

The empirical study of semantic memory in individuals with hippocampal amnesia offers a parallel history to, and a compelling road map for, the study of semantic memory in TBI. Like our current state of knowledge in TBI, there was a time in the amnesia literature when there was clear and definitive evidence for deficits in episodic memory, but the status of semantic memory was unclear. The reader now knows that, today, we understand that this population has deficits both in the new learning of episodic and semantic memory and in the maintenance of remote episodic and semantic memory. To highlight the parallels between that literature and where we are now in our understanding of semantic memory in TBI, we return the reader to a specific turning point in our knowledge of remote semantic memory in amnesia. Specifically, we believe the remote semantic memory literature in amnesia offers a lesson in, and an approach to identifying, what deficits might lie beneath the surface when we investigate semantic memory more deeply.

When Tulving (1972) proposed a distinction in memory between episodic and semantic memory, a natural question among memory researchers turned to the status of semantic memory in patients with amnesia. The literature on new semantic learning in amnesia has been controversial. The debate stemmed from variable results of studies ranging from profound impairments to full preservation as well as various degrees of deficit and ability in between. Looking back at this work, when performance is evaluated against a common standard, new semantic learning in amnesia is impaired; there is not a single replicable example of new semantic learning in amnesia where the amount of information learned or the rate of learning is comparable to noninjured comparison participants (see Duff et al., 2020, for a review).

The literature on remote semantic memory was not controversial. Knowledge of words, concepts, and general facts about the world acquired prior to the onset of their amnesia and outside the effects of retrograde amnesia was judged to be fully preserved. However, most studies of remote semantic memory in amnesia examined only surface-level vocabulary or lexical information. Remote semantic memory was assessed with clinical tools commonly used to assess aphasia or dementia (e.g., BNT, Semantic Memory Test Battery; Kensinger et al., 2001; Schmolck et al., 2000) or experimental tasks whereby participants might be shown a picture of a lemon and asked to name it, asked to identify whether L-E-M-O-N or L-E-N-A-K are real words, or asked to match the label “lemon” to a short definition (e.g., a yellow citrus fruit; Gabrieli et al., 1988; Manns et al., 2003; Verfaellie et al., 2000). Individuals with hippocampal amnesia did not differ from noninjured comparison participants on these tasks, and their intact performance was taken as evidence that their remote semantic memory was not impaired.

Our lab and others (e.g., Blumenthal et al., 2017; Grilli & Verfaellie, 2014; Keane et al., 2020) have revisited the notion that remote semantic memory is intact in amnesia. Our work aimed to examine semantic memory more deeply by drawing on perspectives that consider the richness of semantic memory and as a dynamic network that changes over time and with experience. To do this, we used methods that move beyond surface-level pairings and, in doing so, found a range of disruptions in semantic memory in individuals with hippocampal amnesia. In the rest of this section, we highlight some examples of this work.

To make direct contact with previous naming studies, we asked individuals with amnesia and noninjured comparison participants to view color photographs of items and to provide a name for the picture (Hilverman & Duff, 2021). While individuals with amnesia perform well on measures such as the BNT, these tools have a limited number of items, and those items are skewed toward being highly familiar, frequent, and imageable (Himmanen et al., 2003). We used over a thousand items from the Bank of Standardized Stimuli database (Brodeur et al., 2010, 2014) that varied across a range of lexical features such as imageability, frequency, and familiarity. By using a wide range of image–word pairs, even subtle differences in naming may be detected. In contrast to previous studies of naming, we found that participants with amnesia were less likely than comparison participants to correctly name the objects and performance worsened in amnesia for words that were less familiar and more visually complex. Some of the errors that the participants with amnesia made were surprising. For example, one participant with amnesia was the only participant to not correctly name a pair of skis (referring to them as “two paint brushes”). Prior to the onset of his amnesia, he was an avid skier and worked at a ski resort. Another participant with amnesia was the only participant to not correctly name a syringe (referring to it as “thermometer”) despite having worked for almost 2 decades in a medical doctor's office, with one of her primary duties being stocking the supplies. Using a deeper range of stimuli to examine naming, we detected evidence for disruptions in naming following hippocampal amnesia. We also note that we can rule out a role for significant visual disturbance in the observed naming impairment as all participants with amnesia had normal or corrected-to-normal vision and performed within normal limits on standardized neuropsychological tests of visuospatial (e.g., Complex Figure Test copy score) and visual perceptual abilities (e.g., Judgment of Line Orientation).

To examine knowledge of the relations within a word, we assessed how much information was associated with highly familiar words that were likely acquired long before the onset of amnesia (Klooster & Duff, 2015). We used a word senses task (name all the senses of a word; e.g., lemon can be a fruit, a color, or a defective automobile) and a word features task (name all of the features of a word; e.g., lemon tastes sour, is native to Asia, and used in tea). Participants with amnesia performed significantly worse than noninjured comparison participants on both measures of remote word knowledge. For example, individuals with amnesia generated, on average, only half the number of features for common words (e.g., shirt) as the comparison group. This deficit was evident despite showing no differences from comparison participants on self-reported rates of familiarity of words used in the tasks (scoring familiarity with a particular word on a 9-point scale). The fact that the individuals with amnesia knew these words (i.e., had high familiarity ratings) suggests that they likely would have performed like comparison participants with traditional measures (e.g., naming) that only assess surface-level knowledge. Using tasks and measures that assess the depth and richness of semantic knowledge, individuals with amnesia are impaired, suggesting their remote semantic memory is impoverished.

This study also examined the relations among words or collocates (i.e., words that tend to co-occur with high statistical regularity in the language). Participants were asked to judge acceptable collocates, for example, to endorse that “sudden noise” was permissible in English while “sudden doctor” was not. Individuals with amnesia performed significantly worse than the comparison participants at making these judgments (Klooster & Duff, 2015). This suggests that individuals with amnesia have difficulty maintaining knowledge about the relations among words that interfere with making basic grammatical judgments about their language.

Examination of language production also revealed semantic disruptions. We analyzed the features of words used when participants described past and imagined events (Hilverman et al., 2017). Features of words reflect characteristics of what the word describes (e.g., a word's imageability measures the degree to which the word invokes an image in one's mind). Individuals with amnesia often produce fewer episodic details in their event descriptions (Race et al., 2011). Yet, the specific words used are not necessarily related to the number of episodic details. For example, an individual might say, “I was on a jet ski on a nice summer day and water was hitting my face as I went across the lake,” or “I was riding a jet ski on a bright summer day and water was spraying my face as I sped across the lake.” In both cases, the number of episodic details is the same, but the imageability and concreteness of the words used are greater in the second example. We found that participants with amnesia used words that were significantly less imageable than the comparison participants (Hilverman et al., 2017). Thus, even in seminaturalistic speaking contexts, individuals with amnesia used language that was semantically impoverished.

Taken all together, these findings, and those from other labs, challenged the historical view that remote semantic memory is intact in amnesia and that semantic memory becomes independent of the hippocampus over time. An important implication of this work is that there are likely more far-reaching disruptions in semantic memory than previously documented in populations where hippocampal pathology and episodic memory deficits are common, including in TBI. Yet, detecting these deficits in amnesia required a new approach and the application of new methods with increased sensitivity.

As we have discussed, common sequelae of TBI heighten the risk for disruptions in semantic memory. Given that TBI can cause diffuse neuropathology and widespread cognitive impairments, we suspect that semantic memory disruptions in TBI may be more extensive than what we have documented in individuals with hippocampal amnesia who have more focal neural and cognitive deficits. While there is considerable work to do to test this hypothesis, a couple of recent examples point to the promise of this approach in TBI.

We administered the Remote Associate Task (Mednick, 1962) to individuals with moderate–severe TBI and their noninjured peers (Rigon et al., 2018). Participants were shown compounds made up of three cue words related to each other (e.g., guy, rain, down) and were asked to identify a fourth word (solution) that was associated with all three cues (e.g., fall). The solution could be related with each of the three words by forming a compound word (e.g., rainfall, downfall) or a common phrase (e.g., fall guy, fall down) when combined with it, or because it had a close semantic association with it. Individuals with TBI were significantly less likely to produce a correct answer to the average compound item. In probability terms, individuals with TBI had a 52.62% probability of producing the correct answer compared to a 72.9% probability in the comparison group. Said another way, individuals with TBI were 2.45 times less likely to produce a correct response than their noninjured peers. While future work is needed to understand the mechanisms of disruption (e.g., semantic structure vs. semantic processes of search or selection) and their cognitive underpinnings (e.g., memory, attention), these findings suggest deficits in semantic memory in TBI.

In another study, we used a randomized within-participant crossover design to assess immediate encoding of new words and the consolidation of those words over time (Morrow et al., 2023). Unsurprisingly, participants with moderate–severe TBI exhibited a word learning deficit that began at encoding and persisted across time; individuals with TBI recalled fewer words at all time points. Critically, this deficit grew over the course of the week. The performance gap between groups was larger at the 1-week posttest than the immediate posttest, suggesting deficits in both encoding and consolidation of new words in TBI. If individuals with TBI have deficits in consolidation mechanisms, this raises the possibility of disruptions in the long-term maintenance and reconsolidation of semantic knowledge acquired long before the onset of their brain injury. This warrants further study as this interpretation predicts the gradual attrition of semantic knowledge over time and the potential for increasingly negative outcomes, particularly in academic and vocational spheres, as time postinjury increases. We note that participants with TBI in these two studies above also had normal or corrected-to-normal vision.

Given that TBI often produces a diffuse profile of neural and cognitive deficits, there are multiple possible mechanisms that when impaired could produce a semantic memory deficit in TBI. The studies of individuals with amnesia provide a critical comparison for future TBI studies in that the amnesia group only has the significant memory impairment and deficits in other cognitive domains (e.g., attention, executive function), if present, are not detected on standardized assessments. Our goal here was to encourage more behavioral study of semantic memory in TBI. If successful, the field would have sufficient evidence across tasks and samples to begin the work of delineating the underlying mechanism(s) of deficit, which would, in turn, inform intervention approaches.

Clinical Implications

In this viewpoint, we have advocated for increased study of semantic memory ability in TBI and proposed a future research agenda to examine the depth and breadth of semantic memory disruptions as well as its probable links to academic, vocational, and interpersonal outcomes. What we are proposing will take years. So, what can clinicians do now? In terms of assessment, it is important to note that strong performance on a standardized naming test does not preclude subtle or significant deficits in semantic memory beyond surface level associations. As a field, we need to develop more assessments with increased sensitivity and specificity to deficits common following TBI and with greater availability and accessibility to speech-language pathologists (Duff et al., 2002; Mitchell et al., 2024; Turkstra et al., 2005). This is certainly true for assessments of semantic memory beyond confrontational naming.

In the meantime, clinicians can gain information about the role that semantic memory plays in an individual's daily life and use that to prioritize treatment goals. For example, word play, including completing the New York Times Crossword Puzzle, Wordle, or Connections or engaging in verbal puns and banter with friends and family, is a highly prized skill and meaningful activity for many individuals with brain injury. For these individuals, targeting mild deficits in accessing and maintaining their deep stores of semantic knowledge can be a greater priority than treatment for severe impairments in other forms of cognition (see Hengst et al., 2010, for such a case). For other individuals, returning to school and work will be primary goals, and the ability to learn and maintain new words and concepts will be required for their success (Morrow & Duff, 2023).

In terms of treatment, a consequence of the clinical examination of surface forms of semantic memory (e.g., simple word–referent pairings, basic semantic associations) and historic views that all memories start out as episodic but that, over time, some become semantic through the process of semanticization or decontextualization (Meeter & Murre, 2004) is the assumption that semantic memories are context independent or decontextualized. However, successful use of semantic information is highly context dependent as appropriate representations must be retrieved, combined, and used based on the current demands of the specific social and physical context. Decontextualized and drill-based interventions have little effect on improving basic memory functions (Velikonja et al., 2014; Ylvisaker et al., 2003). Given that semantic memory and episodic memory share highly overlapping neural correlates and functional characteristics and that use of semantic memories is context dependent, there is a strong possibility that these types of interventions would do little to produce meaningful, functional changes in semantic memory ability. Rather, repeated, contextualized practice of the to-be-learned concept/information across time, such as distributed practice (Cepeda et al., 2006; Middleton et al., 2019), and repeated engagement in meaningful activities and in rich communicative environments (Hengst et al., 2010, 2019) seem likely candidates for effective semantic memory rehabilitation. More research in this area is warranted particularly around treatment dose recommendations. Recall that Morrow et al. found that individuals with TBI retrieved fewer words over the course of a 1-week word learning study, including words they had successfully retrieved earlier in the week (Morrow et al., 2023). This suggests an impairment in maintaining (or consolidating) semantic memory above and beyond the deficit in encoding. It is unclear if treatment strategies (e.g., number of trials or sessions) would differ for the acquisition of new semantic memory versus maintaining semantic knowledge from long before the onset of injury.

Semantic memory and episodic memory are part of a unified relational memory system, critically dependent on, but not limited to, the hippocampus. This raises the possibility that behavioral modifications and interventions targeting the structure and function of the hippocampus may have the potential to benefit both semantic and episodic memory in parallel. For example, the hippocampal relational memory system benefits from a range of lifestyle and noninvasive treatments such as neuromodulation (Wang et al., 2014), nutrition (Monti et al., 2014), sleep (Stickgold & Walker, 2005), and physical activity interventions (Erickson et al., 2011), which, when offered alone or in combination with behavioral treatments, hold tremendous promise for improving memory outcomes including in semantic memory.

Further research on semantic memory function in TBI may lead to improved detection and clinical treatment of this critical memory system. Characterization of semantic memory in TBI will be of clinical benefit regardless of its integrity, if largely intact, a strength that can be leveraged, and if disrupted, an area in which intervention has potentially wide-reaching benefits that will help advance academic, vocational, and interpersonal outcomes following TBI.

A Call to Explore

Our goal in this viewpoint was to raise the possibility that we have missed deep disruptions to semantic memory in TBI. We point to the increased risk of such deficits given the well-documented overlap between the neural and cognitive correlates of semantic memory and the common pathophysiological and cognitive deficits in TBI. We have highlighted the lack of systematic investigation around the status of semantic memory in TBI despite these risk factors and suggest that the historical literature and more recent approaches to studying semantic memory in amnesia offer a road map for how we might begin to investigate semantic memory more deeply in TBI. Specifically, methods with sensitivity to the depth and breadth of semantic memory and approaches that examine the consolidation of knowledge over time appear particularly promising. We believe that the empirical study of semantic memory in TBI will advance the characterization of cognitive-communication impairment and provide new opportunities to improve long-term functional outcomes. We hope others will join us in exploring the depths of semantic memory in TBI.

Data Availability Statement

The data sets generated and/or analyzed during the current study are available from the corresponding author on reasonable request.

Publisher Note: This article is part of the Special Issue: Select Papers From the Fourth International Cognitive-Communication Disorders Conference.

Funding Statement

This work was supported by National Institute on Deafness and Other Communication Disorders Grant R01 NIH DC017926 to M.C.D.