Semantic feature analysis is a structured therapy technique used mainly to help people who have lost the ability to find and say words, most commonly after a stroke. The approach works by guiding a person to describe features of a target word they cannot retrieve, such as what it looks like, what category it belongs to, what it is used for, and where you find it. A systematic review of 21 studies found that naming of practiced items improved for about 82% of participants, with the technique producing positive outcomes across a wide range of treatment variations.
How a Typical Session Works
Imagine you had a stroke and now struggle to name everyday objects. Your speech-language pathologist places a picture of an apple in front of you and asks you to name it. You cannot. Instead of just telling you the word, the clinician walks you through a chart with prompts: What group does it belong to? (fruit) What does it look like? (round, red or green) What do you do with it? (eat it) Where do you find it? (grocery store, orchard) What does it remind you of? (pie, teacher’s desk). By generating these features aloud, you are activating a web of related meanings in your brain. Often, the target word surfaces on its own during or after this process. Even when it does not, the exercise strengthens the connections that support retrieval the next time.
The method was originally designed around nouns and picture naming, but clinicians have since adapted it for verbs and for use within connected speech. When applied to action words, the feature prompts shift accordingly: Who does this action? What tools are involved? Where does it happen? One early study of a person with anomic aphasia found that applying this feature-based approach to action names led to improved accuracy that held up six weeks after treatment ended.
Why Generating Features Matters
The therapy rests on the idea that words are stored in the brain as bundles of features rather than as isolated entries. When you think “dog,” you are simultaneously activating fur, barking, four legs, pet, and dozens of other properties. Damage from a stroke can weaken these connections without erasing them entirely. By systematically activating features during therapy, you are strengthening links throughout the network, making the target word easier to reach.
Recent research has bolstered this account by looking at what happens within individual therapy trials. A 2025 study found converging evidence that the act of generating features is itself a key ingredient in the treatment’s success, with both the number and the quality of features predicting better outcomes. In other words, the more features a person generates, and the more meaningfully those features relate to the target, the more likely naming improves.
What Happens in the Brain During Treatment
Neuroimaging studies have given clinicians a window into the brain changes that accompany recovery through this therapy. One study using brain scanning before and after treatment in people with chronic, severe word-finding difficulty found that both hemispheres contributed to naming, and that behavioral improvement was associated not with large new areas lighting up but with a tightening of the existing neural network supporting word retrieval.
Other imaging work has pointed to more specific changes. A study tracking recovery from word-finding problems after semantic feature analysis found that the therapy appeared to trigger recruitment of an alternative neural pathway involving the left precentral gyrus and the left inferior parietal lobule, brain regions associated with the mirror neuron system. Research comparing semantic and phonological treatments has also shown that activity in the left caudate nucleus before treatment predicted greater success specifically for items treated with semantic feature analysis, while different brain areas predicted success for phonological approaches. These findings suggest that the therapy is not simply exercising language in a generic way but engaging particular circuits tied to meaning-based processing.
How Well Does the Improvement Carry Over
One of the biggest questions with any word-finding therapy is whether gains extend beyond the exact words practiced in the clinic. If you practice naming 20 pictures of kitchen items, can you also name things you never practiced? The evidence here is mixed but generally encouraging for semantic feature analysis.
A comparison between semantic feature analysis and a phonological approach found that the semantic group achieved higher accuracy on untrained items, with broader improvements across semantic categories, while the phonological group improved mainly for words related by sound. This makes intuitive sense: if the therapy strengthens an entire network of meaning, other words that share features with the practiced items should become easier to access too.
Typicality turns out to play a role in how well this generalization works. When researchers trained the features of atypical items within a category (say, penguin rather than robin for the category “bird”), participants were significantly more likely to correctly name untrained typical items over time. The logic is that atypical items share some features with the category while also having distinctive features, and training those connections radiates benefit outward through the category.
Generalization to connected speech is a somewhat different picture. One study found that participants did increase the number of target words they used in discourse after treatment but did not increase overall lexical diversity. That means people got better at using the specific words they needed, but their speech did not become richer in general vocabulary. For someone struggling to describe their symptoms to a doctor or order food at a restaurant, though, retrieving the right word at the right moment is exactly the goal.
Dosage and Intensity
How much therapy is enough? Clinicians and researchers have been trying to pin this down. A study looking at practice-related predictors found that the total number of treatment trials mattered more than total treatment time or the rate at which trials were completed. Spending more hours in the clinic did not automatically translate to better outcomes if those hours involved fewer actual naming attempts. This is a useful finding for therapists designing efficient treatment schedules: packing in more practice opportunities within a session appears more important than simply extending session length.
There is also emerging evidence that intensive delivery, meaning more sessions packed into a shorter period, may accelerate progress. A randomized pilot study comparing standard and intensive schedules found that both groups improved on naming tests, but the rate of improvement was greater for the group receiving intensive treatment. Scores remained significantly better than baseline even at follow-up, suggesting the gains were not just a temporary spike from frequent practice.
How It Compares to Phonological Approaches
Semantic feature analysis is not the only game in town for treating word-finding difficulties. Phonological component analysis, which prompts the person to think about the sounds in a word rather than its meaning, is the most common alternative. Studies directly comparing the two reveal an interesting pattern.
In one head-to-head comparison with eight participants, seven showed significant improvements on items treated phonologically, while four showed significant improvements on items treated semantically. The semantic therapy was not helpful for participants whose underlying problem was primarily semantic, which is a somewhat counterintuitive finding. The phonological approach, by contrast, helped most participants regardless of the nature of their deficit. Research in Persian-speaking patients with aphasia echoed this broad finding: both approaches improved naming, but the treatment that matched the person’s specific underlying deficit was more effective.
These comparisons point to something important for clinical decision-making. Semantic feature analysis is not universally superior. Its strength lies in strengthening meaning-based connections, and it shines for people whose sound-level processing is relatively intact but whose access to word meanings is disrupted. For others, a phonological approach or a combination of both may work better. This is why thorough assessment of the nature of the word-finding problem matters before choosing a treatment path.
Who Responds Best
Not everyone benefits equally, and researchers have been working to identify what predicts a good response. One study examining cognitive functions found that verbal recognition memory and visuospatial memory were both associated with the size of early treatment gains, with stronger cognitive performers showing larger improvements. Visuospatial recall also showed some connection to whether gains generalized to untrained words. This makes sense: therapy requires learning, and learning depends on memory systems that may themselves be impaired after a stroke.
For bilingual individuals, the picture adds another layer. A study of bilingual people with aphasia who received semantic feature-based treatment found significant improvement over time in both languages, even when therapy was conducted in only one. However, people who received therapy in their first language improved more in that language. Higher pre-treatment naming scores also predicted greater response, which is a pattern seen across many aphasia treatments: the more language you have to work with going in, the more you tend to gain.
Using the Approach in Neurodegenerative Conditions
Most of the research on semantic feature analysis has focused on stroke-related aphasia, but clinicians have also tried it for people with primary progressive aphasia, a condition in which language abilities gradually deteriorate due to neurodegenerative disease rather than a single event like a stroke. The therapeutic challenge here is different: you are not trying to rebuild connections damaged all at once but rather trying to slow or temporarily reverse an ongoing decline.
Results have been cautiously positive. In a case report of a person with the semantic variant of primary progressive aphasia, intensive therapy using semantic feature analysis improved naming of trained items dramatically, from about 40% to 88% accuracy, though untrained items improved only modestly. The broader benefit appeared in a definition task, where access to semantic features improved for both trained and untrained items. That suggests the therapy strengthened the meaning network even if the naming gains themselves did not generalize widely.
A pilot study combining semantic feature analysis with a multimodal communication program for a person with primary progressive aphasia found that while changes in word retrieval and switching between communication modes were minimal, listeners were better able to understand the participant’s communication attempts after treatment. In a degenerative condition where eventually all words will fail, helping a person communicate more effectively through whatever means available is a meaningful outcome.
Combining Therapy with Brain Stimulation
One of the more intriguing developments is pairing semantic feature analysis with transcranial direct current stimulation, a technique that delivers a weak electrical current to the scalp to modulate brain activity. The idea is that gently increasing the excitability of language-related brain regions during therapy might amplify the therapeutic benefit.
In a study of five patients with progressive naming impairment, a two-week course of semantic feature training combined with brain stimulation produced advantages for trained over untrained items, with trained items also showing a shallower rate of decline over six months compared to untrained items that continued their progressive slide. Patients tolerated the stimulation well.
A longitudinal case report of a person with the semantic variant of primary progressive aphasia found that after combined semantic feature analysis and brain stimulation, about 80% of trained words could be named correctly two weeks after each stimulation phase. Untrained words also improved significantly compared to baseline. These are small studies, and the field is far from establishing brain stimulation as a standard add-on, but the early results are encouraging enough that larger trials are underway.
Technology-Assisted and Remote Delivery
Access to speech-language pathology services is a real barrier for many people with aphasia, whether because of geography, mobility limitations, or cost. This has driven interest in self-administered versions of semantic feature analysis delivered through tablets and other technology.
A case-series study tested a tablet-based treatment that participants used at home four times per week for four weeks. All participants showed significant improvement on trained words, and the gains were maintained two months after treatment ended. The self-administered format preserved the core structure of the therapy while removing the need for a clinician to be present at every session. This is promising for extending the reach of treatment, though clinicians still play a role in setting up the program, selecting target words, and monitoring progress.
Delivery format also matters in group versus individual settings. A study comparing treatment fidelity in both formats found that while fidelity was high in both cases, individual sessions achieved higher fidelity than group sessions (about 95% versus 87%). Group delivery still worked, but the more complex dynamics of a group setting made it slightly harder to follow the protocol precisely. For clinicians balancing caseloads and efficiency, this suggests that group delivery is viable but requires extra attention to keeping the structured prompting on track.
Applications in Education
Semantic feature analysis did not originate as a classroom tool, but its underlying logic, connecting a concept to its defining features to deepen understanding, translates naturally to vocabulary instruction. A study of school-aged Portuguese children tested the strategy in inclusive classroom settings and found that students in the experimental groups significantly outperformed peers who received regular instruction, with large effect sizes in both cycles of the study. The researchers also noted that the strategy was easy to teach, inexpensive, and flexible enough to work in any classroom.
This educational application shares the same core mechanism as the clinical version: organizing knowledge around features makes concepts stickier and more retrievable. For children building vocabulary rather than recovering lost words, the benefits extend to concept knowledge and reading comprehension. The feature chart becomes a thinking tool rather than a therapeutic device, but the cognitive process is fundamentally the same.
The Richer-Than-Expected Semantic Network
The effectiveness of semantic feature analysis depends partly on how rich and interconnected our mental representations of concepts really are. Research into semantic feature norms, the databases that catalog which features people associate with which concepts, has long suggested that these networks are dense and overlapping. A recent study introduced an approach that augmented human-generated feature norms with responses from large language models, then validated the results against human judgments. The resulting dataset showed much higher feature density and overlap among concepts than previous human-only datasets, and it outperformed both older norm datasets and word-embedding models in predicting people’s judgments of how similar two concepts are.
This finding has implications for therapy design. If human conceptual knowledge is richer than previous norm databases captured, then there are more potential connections for therapy to activate than clinicians might assume. It also opens the door to using AI tools to help identify the most productive features to target in treatment, selecting features that are densely connected to many other concepts rather than features that are distinctive but isolated.