When AI can extract a complete cross-sensory translation structure from a single blogger review,
the starting point, cost, and possibilities of sensory analysis all change
In sensory-driven categories -- fragrance, skincare, food, beverages -- there is a structural problem that practitioners all recognize but rarely articulate explicitly:
Consumers can perceive, but cannot express.
In a sensory literacy study of over 3,000 Chinese female beauty consumers, we identified a pervasive disconnect: consumers possess rich perceptual discrimination (understanding) but lack the ability to translate perception into language (articulation). They can distinguish the texture difference between two face creams, but cannot tell you what the difference actually is. They know which perfume they prefer, but cannot explain "why" -- not because there is no reason, but because a translation bottleneck exists between the perceptual system and the linguistic system.
The traditional solution is the sensory panel -- training a group of sensory assessors to encode products using a standardized descriptive vocabulary. Combined with tools like SynthX for statistical analysis, one can generate sensory maps for products. This pathway is mature and effective, but it faces three hard constraints:
| Constraint | Implication |
|---|---|
| High cost | Building and maintaining a panel requires sustained investment in personnel and time, typically affordable only by large consumer goods companies |
| Pre-set vocabulary | The descriptive vocabulary used by the panel must be defined and trained in advance; new perceptual dimensions are difficult to capture within the existing framework |
| Not scalable | The number of products a panel can evaluate is constrained by physical conditions and cannot cover the full linguistic ecosystem of a category |
Meanwhile, the internet generates vast amounts of sensory descriptive content every day -- blogger reviews, consumer comments, social media discussions. This content is spontaneous, untrained, and full of personal style, yet it contains rich cross-sensory translation information. The question is: How do you extract structured sensory mappings from unstructured natural language?
This is the problem that the Synesthetic Mapping Atlas aims to solve.
Before introducing the method, it is necessary to define its boundaries. "Using AI to analyze consumer content" is already being done widely, but the approaches diverge significantly.
A fragrance recommendation engine. Givaudan's Myrissi already maps emotion to fragrance notes, with the goal of recommendation. Our goal is not recommendation, but understanding how perception is transmitted through language.
A synesthetic mapping atlas. It documents how the perceptual system, when confronted with sensory stimuli, uses language to complete cross-channel translation -- from olfaction to touch, from taste to space, from temperature to emotion.
A blogger analysis tool. Bloggers are the data source, not the research subject. We extract not their opinions or preferences, but the cross-sensory translation actions they perform in language.
A structured atlas of category-level sensory language. It can answer: How is consumer perception organized in this category? Which sensory channels are activated first? Which layers of meaning are systematically skipped?
A fragrance note database. Fragrantica records product ingredients and note classifications. We record what the perceptual system does when a person encounters those ingredients.
An extensible foundational method. Fragrance is the current validation category, but the same extraction logic applies to any sensory-driven category -- the texture language of face creams, the flavor descriptions of food, the sensory communication of spatial design.
The core distinction lies in one term: atomic unit. The atomic unit of traditional content analysis is "a single review" or "a keyword." Our atomic unit is a single bridge action -- a complete translation that departs from one sensory channel, passes through a linguistic mechanism, and arrives at another sensory channel or meaning layer.
What does a bridge action look like? Consider a concrete example.
A blogger describing Acqua di Parma's Jasmine Tonka wrote:
[Now this jasmine flower is suddenly suspended in the rising steam of hot black tea]
"Suspended" specifies a highly precise physical state: not rising (the flower does not drift away), not sinking (the flower does not dissolve into the tea), held at the middle layer of the steam. This is not merely an image -- it is a physical metaphor for a fragrance that is "just right."
Traditional text analysis would tag this sentence as "positive sentiment" or extract keywords like "jasmine," "black tea," and "steam." Synesthetic mapping extraction does something fundamentally different: it identifies that this sentence completes a cross-channel translation from olfaction, through spatial-physical state, to emotional shape, and records the specific mechanism of translation -- the word "suspended."
Consider another example, from a different blogger describing Beast Youth Longjing Jasmine:
[Warm, restrained, gentle, and refined]
Same category, different blogger, entirely different bridge vocabulary -- yet the target channels are consistent: both depart from olfaction, pass through tactile or temperature vocabulary, and arrive at an emotional or personality dimension.
In the tea-fragrance category, we extracted 829 such bridge actions from 22 blogger reviews. Each one is annotated with the source sensory channel, the bridge mechanism (specific vocabulary and linguistic strategy), the target meaning layer, the source blogger, and the specific product.
When these bridge actions are aggregated, catalogued, and tallied, the sensory translation structure of the category begins to emerge. It is no longer the vague signal of "consumers think tea fragrance smells nice," but rather a structure that can answer precisely: Which sensory channels does the olfactory signal of tea fragrance get translated into? Through which vocabulary mechanisms? At what frequency?
The bridge action is the atomic unit. On top of the atomic unit, the Synesthetic Mapping Atlas produces four types of structured category insights. Each one addresses a question that traditional methods struggle to answer.
With 829 bridge actions catalogued by sensory channel, the synesthetic translation structure of the tea-fragrance category becomes clear:
| Target Channel | Bridge Actions | Share | Typical Bridge Words |
|---|---|---|---|
| Touch | 141 | 17% | Silky, dense, crisp-cool, sheer |
| Temperature perception | 83 | 10% | Warm-smooth, cool-clean, toasted, icy |
| Emotional shape | 71 | 9% | Restrained, quiet, translucent, unobtrusive |
| Motion-time | 42 | 5% | Suspended, diffusing, settling, flowing |
| Vision | 31 | 4% | Transparent, golden, smoky-haze |
| Gustatory synesthesia | 31 | 4% | Lingering sweetness, astringent, clean-sweet |
A comparative finding: touch and temperature together account for 49% -- nearly half of all synesthetic translations point toward somatosensory channels. This means that when people describe tea fragrance, the most natural response is not "what color does it look like," but "what texture does it feel like."
However, the "typical bridge words" in the table above are merely abbreviations. Observe the granularity of the bridge inventory in actual analysis -- within the same channel (touch), different bloggers invent entirely different translation vocabularies:
[Juice-burst]
木法沙 describing a citrus tea fragrance. "Juice-burst" converts olfaction into the oral tactile experience of biting through citrus peel -- not a slow diffusion, but an instantaneous rupture. This word appears only once in the entire bridge inventory, a highly personal translation invention.
[A composed sense of coolness]
小空试香水 describing a double-mint tea fragrance. "Composed" is a personality word; "cool" is a temperature word. Two words from entirely different perceptual channels are compressed together, giving the coolness a sense of personal temperament: not a sharp, aggressive chill, but an unhurried, self-possessed one.
[Crushed mugwort / The scent after snapping a stem / An unripe peach hanging heavy on the branch]
香香学姐Luna describing the Documents series. She uses no adjectives -- instead, she translates olfaction into the physical action that produces the scent. Not "a fresh mugwort aroma," but "the moment of crushing mugwort." This is an entirely different bridge strategy: it translates not the attribute of a scent, but the source event of the scent.
Three examples, three entirely different translation strategies. This is precisely where the value of the bridge inventory lies: it records not "consumers think it smells good," but how the perceptual system invents cross-channel translations.
The bridge inventory is not merely an analytical output -- it is a material library for synesthetic design. Brand creative teams can directly search it for "validated bridge pathways": if you want consumers to perceive "restraint," you do not need to imagine from scratch what language to use -- the inventory tells you that "suspended," "negative space," and "neither sinking nor floating" are bridge words already validated as effective in this category. The distance from analysis to creative execution is a single table lookup.
Bridge actions answer "how is it translated." The density terrain map answers a more macroscopic question: What structure does this category's narrative operate on?
We decomposed the narrative of each product into six structural layers -- Anchoring (A), Sensory Translation (B), Scene (C), Personality (D), Emotion (E), Logic (F) -- and assigned a density rating to each layer (thick / present / thin / absent). The density patterns of 139 tea-fragrance products were projected into PCA space (the six-layer density coding uses ordinal scores, which suit PCA dimensionality reduction), forming a category-level narrative terrain map. For analyzing bloggers' synesthetic translation channel preferences, we used correspondence analysis (CA) -- because bridge action data is fundamentally frequency-count data. CA is purpose-built for contingency tables, and the blogger-level CA explained 88.9% of information in the first two dimensions, far exceeding PCA's 36.4% on the same data.
This map reveals three things:
First, the category's "foundation" and "superstructure." Layer B (sensory translation) and Layer F (logic bridge) are almost never absent -- they are the infrastructure of category narrative. Layer C (scene) and Layer D (personality) have absence rates of 40% and 56%, respectively -- they are the luxuries of narrative: when present, they significantly enhance depth, but most products choose to skip them.
Observe the specific density differences -- the same blogger describing two products from the same brand, yet with radically different density patterns:
All six layers thick -- 闪闪好香 describing Beast Youth Mo (Longjing tea fragrance)
A-thick (Longjing + jasmine, clearly anchored tea varieties) → B-thick (rich and mellow, warm and restrained, multi-dimensional tactile translation) → C-thick (misty Jiangnan tea mountains, invoking classical Chinese poetic imagery) → D-thick ("Chinese old money," the poised temperament of a cultural figure) → E-thick ("body and mind cleansed," purification-stillness emotion) → F-thick (Chinese aesthetic cultural mapping)
This is one of the narratively densest cases among all 139 products. All six layers thick means the blogger committed maximum narrative resources -- because this fragrance carried her core argument about "what Chinese aesthetics is."
Functional summary -- 木法沙 describing Kilian Imperial Tea
A-thin (only "jasmine tea" as a single anchor) → B-present (icy + clean-sweet) → C-thin → D-absent → E-present (clean-sweet, no pressure) → F-thick ("colleagues and classmates will all compliment you" -- social proof replacing sensory persuasion)
Physical description is minimalist, yet Layer F is the thickest -- the blogger uses quantified social validation ("even perfume skeptics will say it's good") to replace sensory exposition. This is an entirely different narrative strategy: winning not through sensory depth, but through social proof.
Scene replacing sensory -- 木法沙 describing FABRI Garden Tea Party
A-thin → B-absent (zero sensory bridge words) → C-thick ("European aristocratic ladies gathering in a small garden, eating pastries and drinking English afternoon tea") → D-absent → E-present → F-present
This is an extreme case: Layer B is completely absent -- the blogger used no cross-sensory bridge words at all, directly substituting a complete scene picture for the olfactory description. When a scene is specific enough, sensory translation can be bypassed entirely. The density terrain map captures this anomalous "scene replacing sensory" pattern precisely.
Second, the category's "default pathway." From 137 traceable narrative chains, the most common pathway is A→B→E: identify the tea variety, translate into tactile sensation, land on emotion. Scene and personality are systematically skipped. This is not missing information -- it is simply the category's current default grammar.
Third, the category's "blank territory." The theoretical combinations of six layers at four density levels total 4,096. The 139 products cover only 98 (2.4%). 97.6% of narrative space is blank. Some of this is structurally prohibited (for example, "thick emotion + zero logic" never appears), while some represents strategic white space not yet claimed by any brand.
The logic of the density terrain map does not depend on the tea-fragrance category itself. It depends on the assumption that "narrative can be decomposed into structural layers, and each layer exhibits density variation." This means any sensory-driven category can generate its own terrain map. What layers does face cream texture narrative consist of? What pathway does coffee flavor narrative follow? Where does spatial fragrance perception communication break down? Same framework, different category inputs, different terrains.
This is the most surprising of the four outputs.
Traditional methods for discovering consumers' "default expectations" about a category typically require focus groups or large-scale surveys -- directly asking consumers "what do you think a tea fragrance should be like." The problem is that default assumptions are "default" precisely because they do not need to be stated. Direct questioning often yields post-hoc rationalizations rather than the underlying frameworks that truly drive perception.
The Synesthetic Mapping Atlas offers a pathway that circumvents this problem: reverse-engineering default assumptions from negation structures.
The logic is straightforward: when a blogger says "this tea fragrance is not intense," "intense" is the state she assumes the audience expects by default. The negated term is the category's default assumption. Aggregate all negations, and what you obtain is the negative image of category expectations -- no focus groups needed, no questions asked. It is simply waiting in the natural language to be extracted.
In the tea-fragrance category, we extracted 164 negation structures. The results are highly concentrated:
The default consumer expectation for tea fragrance -- or more precisely, the expectation that the tea-fragrance category is systematically negating -- is that "perfume should be strong." 45% of negations are saying the same thing: tea fragrance is not thick, not heavy, not dense, not cloying, not sweet, not aggressive. And the alternative converges on a clear aesthetic direction: light, transparent, clean, breathable.
Observe how reverse inference from negation structure works in practice:
[Many tea-fragrance perfumes on the market don't smell authentic]
EROS闻香识人 setting up the review of Riding the Wave. What is negated here is "default category quality" -- the blogger assumes the audience expects that tea-fragrance perfumes generally lack authenticity. This negation automatically exposes the category's trust deficit. The subsequent description of Riding the Wave using "fiber-like texture" and "authenticity" builds an exception within this negation framework. No focus group needed -- the blogger's natural language has already performed a consumer expectation diagnosis for you.
[Won't tire of smelling it]
imiss香氛实验室 describing Guanxia tea fragrance. "Won't tire of it" reveals the default category assumption that tea fragrance may become tedious, losing its freshness after habituation. The blogger then deploys "long-termism" to counter this assumption -- positioning longevity as the core selling point. The negation structure exposes not only the assumption, but also the consumer's purchase anxiety.
[Eau de parfum does not equal aggressive intensity]
妙人十三 describing Guanxia's Eau de Parfum Collector's Edition. This negation breaks a deeper equation: high concentration = aggression. The blogger substitutes "quiet abundance" for this equation -- demonstrating that concentration and restraint can coexist. What the negation structure reverse-engineers is not just a product-level expectation, but the boundary at which a category rule ("strong = aggressive") fails.
This finding itself may not be new to the tea-fragrance industry. But the way it was discovered is new: it was not obtained through questioning, but automatically extracted from negation structures in natural language. This means the same method can be applied to any category -- you do not need to know what questions to ask; you only need sufficient natural language text. The negation structures will tell you what consumers in that category are negating, and therefore what they are expecting.
The first three outputs are all structured -- they organize unstructured language into forms that can be coded, counted, and compared. But during the extraction process, we found that some bridge actions, even after classification, still carry a residue -- the classification label can capture what function was performed, but cannot capture why it has power.
We recorded these instances and call them meaning excess.
Consider an example. A blogger, describing the structural support of a fragrance, used the expression "skeletal support." In the bridge inventory, it is tagged as "structural sense -- bone metaphor." But "skeletal" in Chinese aesthetic discourse carries a deeper connotation: it denotes a beauty that emanates from within, independent of surface ornamentation. To say a fragrance has "skeletal quality" is to say that its merit derives not from concentration or exuberance, but from internal architecture. "Bone metaphor" captures the linguistic mechanism, but it cannot capture the value judgment embedded in the Chinese aesthetic tradition -- that the internal matters more than the external.
Consider two more instances of meaning excess:
[Passionate, yet with negative space]
EROS闻香识人 describing the guaiacwood base of Jasmine Tonka. In the bridge inventory, this is tagged as "intensity contrast." But "negative space" (liu bai) in the Chinese aesthetic tradition is an independent aesthetic category -- it originates from ink-wash painting, where "the unpainted part is also part of the painting." To say a fragrance is "passionate with negative space" is to say the foundation is full but the expression chooses restraint -- an aesthetic-ethical judgment about "how to present," far exceeding the functional classification of "intensity contrast." The classification framework captures the what, but not the why.
[A smoky, mildly intoxicated feeling]
小空试香水 describing an aged tea fragrance. In the bridge inventory, this is tagged as "temperature + texture." But "mildly intoxicated" extends the olfactory experience into a state of consciousness -- a slight tipsiness, a blurring of perceptual boundaries. This is neither a temperature nor a texture; it is a shift in cognitive state. The classification framework can handle translations between sensory channels, but when the translation's endpoint is "state of consciousness," it exceeds the boundaries of sensory-dimension classification. This excess points to a dimension not yet covered by the framework: the direct impact of sensory experience on consciousness.
Twenty-one meaning excess instances were automatically grouped into 9 types:
| Type | Count | Meaning |
|---|---|---|
| Cultural semantic layer | 8 | Additional value judgments carried by vocabulary within the Chinese aesthetic tradition |
| Negative-space synesthesia | 3 | Conveying perception through "absence," "incompleteness," or "suspension" |
| Physiological trigger | 3 | Directly evoking body memory (salivation, throat resonance, skin sensation) |
| Temporal structure | 2 | Using the time dimension to redefine the meaning of sensory experience |
| Other | 5 | Epistemological cancellation, state-renewal, restoration of authenticity, etc. |
Meaning excess is not a failure of analysis -- it is the classification framework's self-diagnostic mechanism. When a particular language pattern repeatedly produces excess, it indicates that the current classification system has a blind spot requiring expansion. In our practice, the 8 instances of cultural semantic layer excess already triggered a schema update -- the bridge action structure definition gained a new "cultural_semantic_load" field. An analytical framework that evolves during operation is itself one of the method's design objectives.
The entire workflow, from a single blogger review to a structured category atlas, proceeds through six steps. The full pipeline is completed with AI collaboration, with clearly defined points of human intervention.
The entire pipeline has been validated on the tea-fragrance category. When a new blogger review is added, the time from input to bridge cataloguing is approximately 30 minutes (including human verification). The workflow can be directly reused for new categories -- only the category-specific parameters in the extraction rules need to be replaced.
This is not a replacement relationship. The sensory panel answers "what are this product's sensory attributes" -- a product-level question. The Synesthetic Mapping Atlas answers "how is consumer perception organized and transmitted through language" -- a category-level and cognitive-level question. The intersection of the two is this: when you need to design the communication language for a product, you need both the product's own sensory attributes (what the panel tells you) and the sensory channels through which consumers in this category habitually receive sensory information (what the atlas tells you).
One comparison deserves separate attention. In recent years, several attempts have emerged to use AI to map images to fragrance formulations (such as MIT's Anemoia Device), typically following the path: image → visual description → concept terms → matched ingredients → formula. This is a concept lookup approach -- the image contains a flower, flower = floral accord, formulate floral.
The Synesthetic Mapping Atlas takes a different route:
Image contains a flower → flower = floral → formulate floral
Flat mapping from concept to formula (flat lookup)
The flower in the image is suspended in mist → "suspended" bridges to "restrained, not excessive" emotional shape → match a white-floral + tea narrative pattern with skeletal support
Structural matching from synesthetic dimension to emotional shape to narrative pattern
The difference is this: concept lookup processes "what is in the image," while structural matching processes "in what state the elements exist, and what perceptual structure that state corresponds to." The former operates at the content level; the latter operates at the synesthetic structure level. The bridge inventory is precisely the infrastructure for the latter -- without a validated library of bridge pathways, structural matching has no foundation to work from.
The density terrain map can tell you where your product (or a competitor's) breaks down in the six-layer narrative structure -- is it that Layer B vocabulary fails to match consumer perceptual channels? That Layer C scenes lack cultural localization? Or that Layer D personality narrative has no supply at all? Different breakpoint locations correspond to entirely different repair strategies, and the traditional recommendation to "strengthen brand equity" cannot offer this level of precision.
The traditional product development path is: define concept → compose fragrance → test → launch. Synesthetic mapping offers a reverse pathway: start from the target emotional shape, look up the sensory channels and bridge mechanisms that lead to this shape in the bridge inventory, then reverse-engineer what sensory attributes the product needs to activate that pathway. Not "what scent do consumers like," but "what sensory structure, through what synesthetic pathway, arrives at what emotional shape."
The 829 bridge actions constitute an extracted and catalogued "synesthetic translation blueprint library." When writing product copy, you do not need to start from scratch wondering "what adjectives to use" -- you can first determine which synesthetic channel you want to follow (touch? temperature? space?), then look up validated bridge words for that channel in the bridge inventory. From "inspiration-driven copy" to "structured synesthetic design" represents an upgrade in working methodology.
Bridge actions are reusable across categories. The bridge pattern "using temperature words to translate olfaction into emotion" may appear in tea fragrance, but it may also appear in skincare texture descriptions, food flavor narratives, or even the sensory communication of spatial design. Once bridge inventories have been established for multiple categories, cross-category comparison of synesthetic structures becomes possible -- which synesthetic pathways are universal structures of human perception? Which are category-specific?
The Synesthetic Mapping Atlas has completed full-pipeline validation on the tea-fragrance category:
| Completed | Scale |
|---|---|
| Bridge action extraction and cataloguing | 829 entries, with synesthetic and rhetorical type classification |
| Six-layer density coding | 139 products, 98 density signatures |
| Negation structure extraction | 164 entries, 5 sub-dimensions automatically classified |
| Meaning excess capture | 21 entries, 9 types automatically labeled, 3 schema updates triggered |
| Statistical analysis | 11 independent analyses, including CA (synesthetic dimensions), PCA (density structure), non-parametric tests |
| Pipeline standardization | 6-step pipeline, new content can be extracted in 30 minutes |
The current priorities are in two directions:
Vertical expansion: Continue adding samples within the tea-fragrance category. The current 139 products cover 98 density signatures -- as new review content is added, atlas coverage will continue to improve, and blank areas will be further validated as either "structural prohibitions" or "possibilities not yet discovered."
Horizontal transfer: Apply the same methodology to a second category. Selection criteria for the candidate category are: sensory-driven, possessing rich expert-level natural language content, and having a category narrative structure that is not yet well understood. Skincare texture, coffee flavor, and spatial fragrance are all potential directions.