Paper deep dive
A Dynamic Atlas of Persian Poetic Symbolism: Families, Fields, and the Historical Rewiring of Meaning
Kourosh Shahnazari, Seyed Moein Ayyoubzadeh, Mohammadali Keshtparvar
Intelligence
Status: succeeded | Model: google/gemini-3.1-flash-lite-preview | Prompt: intel-v1 | Confidence: 94%
Last extracted: 4/2/2026, 11:51:31 PM
Summary
This paper presents a computational framework for analyzing Persian poetic symbolism by modeling symbols as 'families' rather than isolated tokens. Using a corpus of 129,451 poems, the authors construct a multi-layer graph to track the diachronic evolution of symbolic fields, revealing that while core symbols like 'Night' and 'Earth' remain stable, courtly vocabulary recedes over time while lyric-sensorial imagery (e.g., wine vessels, gardens) strengthens.
Entities (5)
Relation Signals (3)
Shab → persiststhrough → 11 Hijri-century bins
confidence 95% · some families remain widely distributed, especially Shab (Night)
Kherqe → gainsprominencein → Late Centuries
confidence 90% · Kherqe (Sufi Robe) gains late prominence
Khaaqaan → recedesin → Late Centuries
confidence 90% · Khaaqaan (Sovereign) ... are earlier-concentrated and then recede.
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:Persian poetry is often remembered through recurrent symbols before it is remembered through plot. Wine vessels, gardens, flames, sacred titles, bodily beauty, and courtly names return across centuries, yet computational work still tends to flatten this material into isolated words or broad document semantics. That misses a practical unit of organization in Persian poetics: related forms travel as families and gain force through recurring relations. Using a corpus of 129,451 poems, we consolidate recurrent forms into traceable families, separate imagistic material from sacred and courtly reference, and map their relations in a multi-layer graph. The symbolic core is relatively sparse, the referential component much denser, and the attachment zone between them selective rather than diffuse. Across 11 Hijri-century bins, some families remain widely distributed, especially Shab (Night), Ruz (Day), and Khaak (Earth). Wine vessels, garden space, flame, and lyric sound strengthen later, while prestige-coded and heroic-courtly vocabulary is weighted earlier. Century-specific graphs show change in arrangement as well as membership. Modularity rises, cross-scope linkage declines, courtly bridges weaken, and sacred bridges strengthen. Hub positions shift too: Kherqe (Sufi Robe) gains late prominence, Farkhondeh {Blessed} and Banafsheh (Violet) recede, and Saaghar (Wine Cup) stays central across the chronology. In this corpus, Persian symbolism appears less as a fixed repertory than as a long-lived system whose internal weights and connections change over time.
Tags
Links
- Source: https://arxiv.org/abs/2604.01467v1
- Canonical: https://arxiv.org/abs/2604.01467v1
Trouble viewing inline? Open PDF directly →
Full Text
57,424 characters extracted from source content.
Expand or collapse full text
1] University of Technology, , 2] University of Technology, , A Dynamic Atlas of Persian Poetic Symbolism: Families, Fields, and the Historical Rewiring of Meaning Moein [ [ Abstract Persian poetry is often remembered through recurrent symbols before it is remembered through plot. Wine vessels, gardens, flames, sacred titles, bodily beauty, and courtly names return across centuries, yet computational work still tends to flatten this material into isolated words or broad document semantics. That misses a practical unit of organization in Persian poetics: related forms travel as families and gain force through recurring relations. Using a corpus of 129,451 poems, we consolidate recurrent forms into traceable families, separate imagistic material from sacred and courtly reference, and map their relations in a multi-layer graph. The symbolic core is relatively sparse, the referential component much denser, and the attachment zone between them selective rather than diffuse. Across 11 Hijri-century bins, some families remain widely distributed, especially شب Shab (Night), روز Ruz (Day), and خاک Khaak (Earth). Wine vessels, garden space, flame, and lyric sound strengthen later, while prestige-coded and heroic-courtly vocabulary is weighted earlier. Century-specific graphs show change in arrangement as well as membership. Modularity rises, cross-scope linkage declines, courtly bridges weaken, and sacred bridges strengthen. Hub positions shift too: خرقه Kherqe (Sufi Robe) gains late prominence, فرخنده Farkhondeh (Blessed) and بنفشه Banafsheh (Violet) recede, and ساغر Saaghar (Wine Cup) stays central across the chronology. In this corpus, Persian symbolism appears less as a fixed repertory than as a long-lived system whose internal weights and connections change over time. keywords: Persian poetry, digital humanities, computational literary studies, symbolic networks, diachronic semantics, literary history 1 Introduction Persian poetry has long been read as a culture of recurrent images. A rose is rarely only a rose, a cup rarely only a vessel, and the night sky rarely a neutral backdrop. In lyric, mystical, and courtly registers alike, familiar symbolic materials return in changing constellations. Their force depends on relation: what they cluster with, what they attract, and what historical pressures shape their reuse [13, 15]. That problem sits awkwardly within the usual computational scales of literary study. Token-based methods are useful for lexical profiling, stylistic comparison, and retrieval, but they split apart symbolic units that readers often treat together. Document-level semantics can recover broader thematic structure, yet they blur the distinction between imagistic symbols, sacred names, courtly titles, and generic semantic association. For Persian poetry, the useful scale lies between those poles: not the isolated form, and not the whole poem, but a family of related symbolic forms whose literary work becomes legible in aggregate [8, 5, 18]. We model that middle scale through symbolic families. A family is a traceable grouping of closely related surface forms supported by contextual and relational evidence. Working at this level reduces token fragmentation without losing lexical accountability. It also keeps sacred and courtly reference visible without letting it determine the same topology as imagistic symbolism. The study asks three connected questions. Does family induction produce a cleaner symbolic inventory than token-only modeling? Does the system separate into imagistic, referential, and attachment structures? And across the mapped centuries, which families persist, which shift, and which change structural role inside the graph? Those questions matter both for Persian literary history and for computational work that needs a stable unit between word and document. 2 Related Work Computational literary studies has established the value of large-scale evidence for literary history through distant reading, macroanalysis, and corpus-level stylistics [8, 5, 18]. Network-oriented work has pushed the same conversation further by showing that literary organization often appears more clearly in relations than in counts alone, whether the units are characters, authors, or influence pathways [3, 16, 20]. Even so, the usual analytic units remain the token, the document, or the social actor. Symbolism poses a different problem. Its salient unit is often a culturally stabilized cluster of related forms. The gap is especially visible in Persian literary computation. Recent work has improved corpus building, era or style classification, and exploratory AI-assisted interpretation [9, 12, 10]. These studies make large-scale Persian analysis possible, but they usually model poems, poets, or lexical items rather than symbolic constellations. Persian literary scholarship has long treated imagery differently: as a historically sedimented system of figures whose meaning depends on reuse, adjacency, and convention rather than isolated lexical identity [13, 15]. The question is not whether Persian poetry can be processed computationally. It is how to represent symbols without either fragmenting them into raw tokens or dissolving them into document-level themes. This study also draws on diachronic semantics and dynamic network analysis. Work on semantic change has shown that historical meaning is often better captured by shifting relational profiles than by counts alone [4, 7]. Dynamic-network research offers parallel ways to think about change through tie formation and dissolution, latent-space trajectories, community evolution, and diachronic node embeddings [6, 14, 1, 11, 19, 2]. Literary applications move in the same direction. Tanaka’s phylogenetic modeling of waka and poet-network studies in Persian and Chinese traditions treat literary history as an evolving relational object rather than a static thematic inventory [17, 16, 20]. We extend that line of work by taking symbolic families, rather than single words or authors, as the basic analytic unit. 3 Corpus and Method 3.1 Corpus and chronology The analysis is based on a corpus of 129,451 Persian poems attributed to 216 poets. After structural parsing, the corpus contains 1,446,226 verses and 2,891,826 hemistichs. Chronological analysis uses poet-level century metadata covering the full corpus and grouped into 11 Hijri-century bins, with the earliest sparse material merged into a single bin spanning the third and fourth centuries. Verse coverage under this mapping is complete for the analyzed corpus: every poem inherits a century from an explicit local poet-to-century table rather than from a language-only inference procedure. The chronology is therefore coarse, but its coarseness is visible and controlled rather than hidden. Table 1: Corpus and symbolic-inventory summary. Measure Value Poems 129,451 Verses 1,446,226 Hemistichs 2,891,826 Poets 216 Century bins 11 Raw symbolic tokens consolidated 1,160 Canonical symbolic families 738 Families admitted to atlas construction 175 Unique families retained in the final atlas 157 3.2 Family induction and traceability The central methodological move is to treat the symbolic family, rather than the raw token, as the basic analytic unit. Families were built conservatively. Candidate merges were considered only for explicit surface relations such as clitic removal, prefix or suffix stripping, near-head extension, or synthetic heads supported by several strong variants. For each eligible merge, a composite score combined positive evidence from surface relation, static and occurrence-level embedding support, contextual overlap, cluster overlap, seed overlap, exemplar overlap, and head cleanliness, while subtracting penalties for fragment residue, function-word contamination, genericity noise, and proper-name inflation. In compact form, Smerge=∑rwrEr−∑qλqPq.S_merge= _rw_rE_r- _q _qP_q. Candidates were accepted only when the surface relation was eligible, the composite score cleared the staged threshold, and the merged head remained cleaner than its main variants. The staged schedule was conservative rather than expansive: strict early gates were relaxed only after a candidate had already passed the basic coherence checks, which is why the final inventory remains compact without becoming indiscriminate. Traceability and head selection were enforced separately. Every raw token maps to exactly one family, and the token-to-family assignments remain inspectable in the preserved member and diagnostics tables. The canonical head is the root form of the accepted assignment chain; when no observed head is adequate, a synthetic head is admitted only if multiple variants jointly clear the synthetic-family threshold. In practice this keeps a family philologically legible even when several surface forms are absorbed. At this scale, the model stays close to literary usage. A family can absorb orthographic variation, clitic fusion, and near-duplicate heads while remaining easy to inspect. It also keeps apart categories that a looser semantic grouping would blur. Sacred titles, courtly names, and imagistic families can all carry symbolic charge, but they do different work in the corpus. We discussed the grouping scheme with a specialist in Persian literary studies as a plausibility check rather than a formal annotation exercise. Figure 1 sketches the sequence from family recovery to graph construction and historical analysis. 1Corpus and chronology Poems are aligned to poet-level century bins. 2Family induction Forms merge only when several evidence layers indicate one symbolic actor. 3Layered atlas Imagery, reference, and bridges are modeled separately. 4Temporal profiles Century profiles are normalized, smoothed, and variant-checked. 5Dynamic graphs Century snapshots track hub drift, structural change, and bridge retargeting. 6Literary interpretation Interpretation compares persistent families with shifting fields. Figure 1: Analytical workflow from corpus alignment to dynamic graph analysis. 3.3 Atlas construction, edge weights, and community labels The graph uses three connected components. The symbolic core contains families whose main force is imagistic or figurative. A separate referential component contains sacred and courtly families whose names remain structurally important but should not determine the same topology. Cross-layer links record recurrent attachment between the two. Each retained family receives one primary residency in either the core or the referential component, and cross-layer participation is then tracked separately. Assignment depends on symbolic strength, genericity risk, referential pressure, and bridge-anchor behavior rather than a manual whitelist. Edges are likewise evidence-weighted rather than frequency-only. For each candidate pair (i,j)(i,j) in layer ℓ , the atlas combines multiview embedding support, weighted context overlap, cluster overlap, representative-context overlap, verse and document co-association, poet overlap, shared seed support, and pairwise local co-activation into a single layer-specific score, Sij(ℓ)=∑mwmAij(m)−γℓGij+ρℓRij.S_ij^( )= _mw_mA_ij^(m)- _ G_ij+ _ R_ij. Here GijG_ij is a genericity penalty applied in every component, while RijR_ij captures referential pressure: it is negative in the core, near-neutral in the referential component, and slightly positive in the bridge view, where attachment to sacred or courtly material is part of the target phenomenon. After scoring, each component applies a fixed threshold and then a top-k sparsification rule that becomes stricter as node genericity rises. The retained specification uses thresholds of 0.410.41, 0.360.36, and 0.380.38 for the core, referential, and bridge components, with top-k limits of 66, 44, and 44 before node-specific genericity adjustment. The result is sparse enough to read in print. These choices were fixed from construction diagnostics before the diachronic analysis. Community detection uses weighted greedy modularity on the core graph, both for the global network and for century-specific snapshots. Field labels come from overlap between community heads and anchor vocabularies for wine-tavern ritual, floral-vegetal, body-beloved, light-fire, water-sea, ascetic-mystical, and the two bridge domains. When overlap is weak and no seed anchor is present, the community remains in the mixed field. The labels are descriptive summaries of dominant membership, not stronger claims than the evidence supports. 3.4 Diachronic and dynamic analysis Historical analysis proceeds in two steps. First, family frequencies are tracked across century bins after normalization by verse volume. The main series is reported per 10,000 verses, with a three-bin rolling mean and a Gaussian smoothing pass with σ=1.0σ=1.0. To check poet concentration, we also compute poet-balanced weighting and top-poet-removal variants. Trend direction is evaluated from the smoothed trajectories with linear slope, Spearman monotonicity, and chi-square distribution-shift tests, then filtered through 100100 poet-bootstrap resamples. Families whose peak centuries or slope signs remain unstable under these checks are marked uncertain. Second, century-specific graphs are derived from the same backbone so that structural change can be read within a common symbolic space. Dynamic nodes are activated by century-specific family presence; dynamic edges are activated by century-restricted co-activation on existing backbone edges. From these graphs we track hub drift, community change, and bridge retargeting. The dynamic graph is rebuilt under five variants: loose, base, and strict activity thresholds, a poet-balanced weighting specification, and a high-confidence-only subset. In the base specification, families require at least 55 raw occurrences, 33 verse-presence events, and 22 edge co-activations per century; the loose and strict variants shift those cutoffs to 3/2/13/2/1 and 10/5/310/5/3. We foreground claims that survive across these variants and label weaker ones as threshold-sensitive. Appendix A lists the retained parameter values and robustness variants. 4 Results 4.1 Recovering symbolic families from lexical scatter A token-only view fragments recurring symbolic material into partially overlapping forms and lets prestige vocabulary, generic reuse, and surface variation compete with imagistic material on the same footing. Family induction reduces that scatter without giving up lexical traceability. Because most families remain small, the consolidation step is selective rather than blanket lemmatization. Table 2: Compact validation of family-level modeling. Diagnostic Value Summary Raw symbolic tokens → canonical families 1,160→7381,160→ 738 Family induction removes 36.4%36.4\% of analytic scatter while preserving token-level traceability. Median family size / single-member share 11 / 83.7%83.7\% Consolidation is conservative rather than indiscriminate; most forms are left untouched unless evidence is strong. Top-250 raw candidates → distinct families / clear fragment reduction 250→183250→ 183 / 98.65%98.65\% The same window loses 6767 duplicate raw positions and absorbs 7474 clear fragment-like forms into stronger heads, producing cleaner rankings. Seed-family recovery 18/1818/18 The manually seeded symbolic anchors are all recovered in the induced family inventory. Referential leakage into top 40 symbolic-core families 0 The induced core stays imagistic rather than being overrun by sacred or courtly name-like material. Table 2 shows the immediate effect of working at this scale. The inventory becomes smaller and cleaner, yet most families remain singletons. In the top-250 raw window, repeated surface variants stop taking multiple ranking slots, and sacred or courtly material stays out of the symbolic core. 4.2 Symbolic structure: imagery, reference, and attachment In the retained graph, the symbolic core is relatively sparse, the referential component much denser, and the attachment zone between them restricted rather than diffuse (Table 3; Figures 2A–2C). The unique retained inventory contains 157157 families assigned primarily to the core or referential component; bridge participation is an overlapping view on that same set. Imagistic symbols form selective neighborhoods. Sacred and courtly names cluster more tightly among themselves, and only some of them attach recurrently to the core. Table 3: Structural summary of the symbolic graph. Layer Nodes in layer Edges Avg. degree Density Role Imagistic core 127 223 3.51 0.028 Recurrent symbolic fields built from floral, bodily, ritual, elemental, and spatial imagery Referential layer 30 81 5.40 0.186 Sacred and courtly names treated as meaningful reference rather than folded into imagistic symbolism Bridge layer 63 143 4.54 0.073 Attachment zone linking referential materials to symbolic neighborhoods Note: The bridge row is a participation count rather than an additional disjoint node inventory. The unique atlas total in Table 1 reports the 157 retained families assigned primarily to the core or referential residencies; the bridge count records how many of those same retained families also participate in cross-layer attachment edges. Within the core, the dominant communities are recognizably literary. Figure 2A. Imagistic Core. The sparse symbolic core organizes recurrent families such as Saaghar (Wine Cup), Aatash (Fire), and Bahr (Sea) into legible imagistic neighborhoods. Figure 2B. Referential Layer. Sacred and courtly reference forms a denser cluster around titles such as Payghambar (Prophet) and royal designations such as Shaahanshaah (King of Kings). Figure 2C. Bridge Layer. Mediating families such as Eqlim (Realm), Farkhondeh (Blessed), and Dokhtar (Daughter) carry reference toward imagistic fields without collapsing the two layers. One cluster joins tavern, wine vessel, and ritual language; another gathers bodily beauty, hair, and fragrance; others organize around sea imagery, vegetal verticality, and light or fire. These are coordinated fields rather than isolated emblematic words. Figure 3 makes that visible at a glance. The referential component is not a residue left outside the core. Sacred titles and courtly names remain active and historically meaningful, but they behave differently from imagistic families. Their strongest recurring attachments concentrate on the mixed field, where prestige, devotion, and imagery meet through recurrent contact zones. 4.3 Stable anchors, rising families, and declining prestige vocabularies Century-level trajectories show a system that changes without losing its backbone. After smoothing and robustness filtering, 29 families are robustly rising, 9 robustly declining, and 4 stable across the full chronology. Continuity is present, but not as stasis. Persistent families remain in place while the surrounding fields are reweighted, and the bootstrap peak shares in Table 4 keep the remaining temporal uncertainty explicit. Figure 3. Community structure in the imagistic core. Colors identify the principal symbolic fields recovered from the atlas without further in-figure annotation. The rising families concentrate in a late-strengthening lyric ecology. پیمانه Peymaaneh (Wine Vessel) and صهبا Sahbaa (Wine) rise alongside families tied to gardens, flame, melody, and sensorial surface, including گلشن Golshan (Garden). This is not only a matter of raw frequency. The same late centuries also show tighter coordination among the wine-tavern, floral, and body-beloved fields. The strongest field-level coupling is between floral-vegetal and wine-tavern ritual (smoothed correlation =0.983=0.983), followed closely by body-beloved with wine-tavern ritual (0.9820.982) and body-beloved with light-fire (0.9790.979). The declining families point in another direction. خاقان Khaaqaan (Sovereign), فرخ Farrokh (Auspicious), باد Baad (Wind), and related items are earlier-concentrated and then recede. What declines here is not “reference” in general, but a particular prestige-coded register associated with courtly and heroic diction. The pattern does not justify a simplistic historical opposition between “courtly” and “mystical” poetry. Even so, the redistribution is clear enough to matter: earlier prestige vocabulary loses relative weight while later lyric-sensorial families become more central to the symbolic profile. Stable anchors push against overdramatic periodization. شب Shab (Night), روز Ruz (Day), and خاک Khaak (Earth) remain distributed across the chronology with high temporal entropy and low drift. They are not specialized motifs tied to one historical moment. They function more like durable coordinates around which other families rise or fall. Table 4: Representative family trajectories after temporal refinement. Peak-mode share reports the proportion of 100100 poet-bootstrap resamples that recover the same modal peak century. Family Trajectory Peak Peak-mode share Field Pattern Peymaaneh (Wine Vessel) Robust rising 12 76% Wine-tavern ritual Late strengthening of vessel imagery within lyric-tavern clusters Sahbaa (Wine) Robust rising 14 78% Wine-tavern ritual Sustained late growth of wine nomenclature beyond a single century spike Golshan (Garden) Robust rising 12 81% Floral-vegetal Expansion of garden space as a late symbolic environment Khaaqaan (Sovereign) Robust declining 6 63% Referential / courtly Recession of prestige-coded royal diction after early prominence Farrokh (Auspicious) Robust declining 5 93% Mixed field Weakening of auspicious courtly language inside the broader contact zone Baad (Wind) Robust declining 6 67% Mixed field Earlier-heavy mobility and force imagery loses weight historically Shab (Night) Stable timeless 12 51% Mixed field Persistent lyric anchor with broad distribution across the chronology Khaak (Earth) Stable timeless 11 90% Mixed field Durable ethical and metaphysical anchor rather than a period-bound image Figures 4A–4C show the contrast directly. The stable curves remain present across nearly the whole chronology, while the rising and declining curves track redistribution within the same repertoire. Figure 4A. Later-Strengthening Families. Peymaaneh (Wine Vessel) and Sahbaa (Wine) intensify most strongly in the later chronology. Figure 4B. Earlier-Concentrated Families. Khaaqaan (Sovereign) and Farrokh (Auspicious) retain their strongest weight in the earlier bins. Figure 4C. Stable Anchors. Shab (Night) and Khaak (Earth) remain broadly distributed across the mapped chronology. 4.4 Structural change across centuries The historical signal becomes clearer when the analysis shifts from family frequencies to network structure. Century-specific graphs show a system that stays recognizable while changing its internal arrangement. The late-minus-early deltas are modest, but they recur in the same direction often enough to keep. Core modularity rises in the base specification by +0.020+0.020, and the same direction appears in four of the five robustness variants. Cross-scope linkage declines by −0.032-0.032 in the base specification, again with the same direction in four of five variants. Degree centralization is essentially flat. What changes, then, is not the existence of the network but the separation of its fields. Hub drift makes that reorganization tangible. خرقه Kherqe (Sufi Robe) is the clearest rising hub, reaching its highest structural prominence late in the chronology. فرخنده Farkhondeh (Blessed), بنفشه Banafsheh (Violet), and related early-centered families are much stronger in earlier and middle centuries than they are late. ساغر Saaghar (Wine Cup) remains a stable hub throughout, and stable hubs such as باده Baadeh (Wine) and گل Gol (Rose) reinforce the same point: the symbolic repertoire persists even while the surrounding network changes. What shifts, then, is not simply which families exist, but which ones organize the symbolic space at a given historical moment. Table 5: Dynamic network changes with robustness and uncertainty summary. Metric Base change Uncertainty summary Variant support Interpretation Core modularity (late minus early) +0.020+0.020 Comparable-variant span +0.012+0.012 to +0.030+0.030; permutation upper-tail 0.140.14 Positive in 4/54/5 variants Small but fairly consistent separation among fields. Cross-link ratio (late minus early) −0.032-0.032 Comparable-variant span −0.032-0.032 to −0.019-0.019; permutation lower-tail 0.050.05 Negative in 4/54/5 variants Cross-scope links decline across most reconstructions. Courtly bridge strength (late minus early) −0.033-0.033 Variant span −0.033-0.033 to −0.007-0.007 Negative in 2/42/4, flat in 2/42/4 Courtly weakening is visible, though it depends more on threshold choice. Sacred bridge strength (late minus early) +0.049+0.049 Variant span +0.042+0.042 to +0.071+0.071 Positive in 4/44/4 variants Sacred-prophetic attachment strengthens late with the strongest support in this table. Degree centralization (late minus early) −0.001-0.001 Comparable-variant span −0.002-0.002 to −0.001-0.001; high-confidence subset +0.037+0.037 Flat in 4/54/5 variants Centralization contributes little to the main historical contrast. Community-level change points in the same direction. The wine-tavern ritual field is the most persistent major community, with a mean active-member ratio of 0.9900.990. Light-fire and water-sea are also highly stable. Floral-vegetal stays active but has the highest split pressure (0.3920.392), so it changes internally more than the other large lyric fields. Body-beloved shows stronger merge pressure toward floral-vegetal, which indicates recurrent traffic between bodily and garden imagery. Ascetic-mystical is smaller, but its strongest external pull is toward wine-tavern ritual. Figures 5A–5B and 6A–6B plot the same shifts. Modularity rises gradually, cross-scope linkage falls, and families do not move in parallel. Courtly bridges weaken overall, whereas sacred-prophetic bridges gain strength late while remaining centered on the mixed field. Figure 5A. Topological Evolution. Core modularity rises over time while cross-scope linkage declines and degree centralization remains comparatively flat. Shaded ribbons show the span across the loose, base, strict, and poet-balanced reconstructions. Figure 5B. Hub Drift. Kherqe (Sufi Robe) strengthens late, whereas Farkhondeh (Blessed) and Banafsheh (Violet) lose structural prominence; Saaghar (Wine Cup) remains comparatively stable. Figure 6A. Attachment Strength. The Sacred-Prophetic Bridge strengthens overall, whereas the Courtly Bridge weakens or remains threshold-sensitive. Figure 6B. Attachment Dispersion. The two bridge domains show uneven but differentiating patterns of target dispersion across the chronology. 4.5 Poet-level differentiation, clustering, and historical breadth We also recalculated the field profiles at poet level. The threshold-based core pool contains 3636 poets with at least 10,00010,000 verses and 1,0001,000 symbolic-family occurrences. The rule is coverage-driven rather than prestige-driven: it includes Saadi, Khaaqani, and Vahshi Bafqi under the same standard, while Hafez falls below the mapped verse floor and is excluded from both the core pool and the broader sensitivity check. Figure 7 orders the eligible poets by Ward clustering on standardized field profiles; Table 6 gives the strongest field-level concentrations, and Appendix Tables LABEL:tab:appendix_poet_profiles–LABEL:tab:appendix_poet_neighbors provide the full poet statistics. The poet space is structured rather than noisy. A three-cluster solution yields the clearest literary partition: a lyric-amplifying cluster with high wine, body, floral, and mixed-field rates; a sparse-anchor cluster with lower-than-average field intensity across the lyric system; and a broader mediating-balanced cluster that contains poets such as Sana’i, Khaaqani, Saadi, Mowlana, Vahshi Bafqi, and Bidel Dehlavi. The clusters are not reducible to century bins. The lyric-amplifying group stretches from the seventh to the thirteenth centuries; the sparse-anchor group joins fourth-, fifth-, sixth-, thirteenth-, and fourteenth-century poets; and the mediating cluster spans the fifth through thirteenth centuries. Chronology matters, but it does not exhaust poet-level symbolic differentiation. Similarity structure sharpens that point. In standardized field space, Khaaqani and Vahshi Bafqi remain comparatively central rather than isolated, whereas Saeb, Jahan Malek Khatun, and Ashofteh Shirazi are the clearest outliers. Neighbor relations also cross chronology: Saadi lies closest to Sana’i and Anvari, while Khaaqani sits near Mowlana rather than only beside adjacent sixth-century poets. Shared tradition survives as patterned resemblance, but poets amplify it unevenly and sometimes along noncontiguous historical lines. Figure 7. Clustered poet-field profiles. Eligible poets are ordered by Ward clustering on verse-normalized field rates; the side strip marks the three retained clusters, with gold for lyric-amplifying, teal for mediating-balanced, and plum for sparse-anchor profiles. Warmer cells indicate stronger-than-average concentration within a field, cooler cells weaker-than-average concentration. The poet comparison also helps with the century effect. On a broader sensitivity pool of 5858 poets with at least 8,0008,000 verses and 500500 symbolic-family occurrences, 24/2724/27 later poets lie above the earlier median for Wine-Tavern Ritual, 20/2720/27 for Floral-Vegetal, 21/2721/27 for Body-Beloved, and 25/2725/27 for Ascetic-Mystical. Those shifts persist after the single strongest late poet is removed: the late median remains above the earlier median in wine (580.6580.6 versus 292.6292.6), floral (679.8679.8 versus 458.0458.0), body (859.2859.2 versus 585.5585.5), and ascetic-mystical language (117.6117.6 versus 44.144.1). Saeb alone supplies 45.7%45.7\% of late wine-tavern occurrences, yet the field still rises across nearly the whole later pool. Table 6 shows the strongest concentrations inside that shared system. Saeb concentrates wine, body, and water imagery; Jahan Malek Khatun intensifies floral and mixed fields; Ahli Shirazi peaks in light-fire; and Ashofteh Shirazi in ascetic-mystical language. The repertoire is shared, but poets redistribute emphasis unevenly within it. Table 6: Leading poet-level field affinities in the threshold-based core poet pool. Rates report field occurrences per 10,00010,000 verses among mapped poets with at least 10,00010,000 verses and 1,0001,000 symbolic-family occurrences. Field Leading poet Rate / 10,00010,000 verses Characteristic family heads Wine-Tavern Ritual Saeb (11th c.) 3250.2 Mey (Wine); Baadeh (Wine); Sharaab (Wine) Ascetic-Mystical Ashofteh Shirazi (13th c.) 604.7 Meykhaaneh (Wine House); Meykadeh (Wine House); Moghaan (Magi) Body-Beloved Saeb (11th c.) 1624.6 Cheshm (Eye); Lab (Lip); La’l (Ruby) Floral-Vegetal Jahan Malek Khatun (8th c.) 1770.9 Sarv (Cypress); Gol (Rose); Qad (Stature) Light-Fire Ahli Shirazi (10th c.) 692.2 Aatash (Fire); Nur (Light); Zolmat (Darkness) Water-Sea Saeb (11th c.) 771.0 Daryaa (Sea); Mowj (Wave); Bahr (Sea) Mixed Field Jahan Malek Khatun (8th c.) 6457.3 Del (Heart); Eshq (Love); Shab (Night) 4.6 Literary context for the symbolic shifts The poet-level contrasts are easier to place once they are read back into literary history. Saadi and Mowlana occupy the broad mediating cluster rather than the most amplified lyric pole. That fits their historical role in consolidating the ghazal and didactic-mystical lyric as flexible symbolic forms rather than narrowly specialized repertoires. Saadi keeps bodily and floral material in measured relation to a stable mixed field. Mowlana leans more strongly toward light-fire and water-sea imagery, which suits an ecstatic symbolic economy in which motion, illumination, and dissolution matter more than courtly designation. Khaaqani points in a different direction. Although strongly associated with learned panegyric and courtly rhetoric, he remains symbolically central rather than sealed off in a prestige enclave. What declines over time is not rhetorical density as such, but the sustained weight of royal and prestige-coded symbolic heads. Courtly discourse loses relative force even while highly wrought poets continue to share the larger imagistic network. The later lyric shift is sharper. Saeb’s concentration in wine, body, and water fields matches the intensified imagistic chaining usually associated with later ghazal practice and, more specifically, with the dense figurative pressure of sabk-e hendi. Bidel belongs to the same later chronology without reproducing Saeb’s extremity: his profile stays more mediating, consistent with a more abstract and reflective late style. Jahan Malek Khatun and Ahli Shirazi show that amplification is field-specific rather than uniform, with one poet intensifying floral-mixed continuity and the other light-fire concentration. In this view, the rise of wine-tavern and adjacent lyric fields is not just an increase in sensual vocabulary. It marks a broader reorganization in which mystical metaphor, bodily lyricism, and imagistic density become more tightly coupled, while explicitly courtly symbolism recedes. 5 Discussion The main pattern is continuity with rearrangement. Night, day, and earth remain widely distributed, while later centuries bring stronger coordination among wine, garden, body, and light. Earlier prestige-coded and heroic vocabularies lose relative ground. The raw effect sizes are often modest, but the direction of change repeats across variants and across family-, field-, and graph-level summaries. Reference changes in a different way. Sacred figures and royal names remain symbolically consequential, but they do not behave like ordinary imagistic families. Keeping them in a separate component shows where they attach and where they do not. The mixed field matters here because it absorbs a large share of those recurring attachments. Methodologically, the family scale sits between token and document. It stays close to the lexicon while still supporting historical comparison. Persian symbolism is one case where that middle unit is useful, but the same logic could apply to motifs, ritual vocabularies, epithets, or recurring imagistic frames in other traditions. For Persian literary history, the result is fairly simple. The tradition does not reduce to a timeless storehouse of emblems, and it does not dissolve into a sequence of disconnected lexical fashions. Familiar resources remain available, but their relations are historically rearranged. 6 Limitations Several limitations shape these claims. First, the chronology is poet-level rather than poem-level, so internal career development, uncertain dating, and later textual transmission are compressed into coarse century bins. Second, the family inventory remains an interpretive construction even when it is strongly constrained by contextual evidence. Some border cases, especially in the mixed field, still admit alternative classification. Third, the dynamic graphs are derived from a common backbone. That improves comparability across centuries, but it also means that century-specific graphs are changing subsets of a shared symbolic space rather than independently discovered networks. Fourth, the null evidence remains modest: the chronology-order permutation baseline tests whether the observed directional claims are stronger than arbitrary reordering of the stored snapshots, but it is not a full generative null over poems or edges. Not every topological signal is equally strong. Modularity and cross-scope linkage move more consistently than centralization, and courtly bridge weakening is more threshold-sensitive than sacred bridge strengthening. 7 Conclusion Modeling symbolic families changes the view of Persian poetry at scale. Grouped forms produce a network that is sparse in its imagistic center, denser in reference, and shaped by recurring attachments between the two. Across centuries, some families stay widely distributed, others rise or fade, and the network shifts incrementally. Modularity increases, cross-scope links decline, sacred bridges strengthen, and hub positions move. The change is relational more than inventory-wide. Acknowledgements We thank the Ganjoor project for making broad digital access to Persian poetry possible and for enabling corpus-based work of this kind. We also acknowledge the Persian poetic tradition on which this study depends and approach that literary heritage with respect for its cultural importance. References Aggarwal and Subbian [2014] Aggarwal C, Subbian K (2014) Evolutionary network analysis: A survey. ACM Computing Surveys 47(1):10:1–10:36. 10.1145/2601412 Boccaletti et al [2014] Boccaletti S, Bianconi G, Criado R, et al (2014) The structure and dynamics of multilayer networks. Physics Reports 544(1):1–122. 10.1016/j.physrep.2014.07.001 Elson et al [2010] Elson DK, Dames N, McKeown KR (2010) Extracting social networks from literary fiction. In: Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics. Association for Computational Linguistics, Uppsala, Sweden, p 138–147 Hamilton et al [2016] Hamilton WL, Leskovec J, Jurafsky D (2016) Diachronic word embeddings reveal statistical laws of semantic change. In: Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Association for Computational Linguistics, Berlin, Germany, p 1489–1501, 10.18653/v1/P16-1141 Jockers [2013] Jockers ML (2013) Macroanalysis: Digital Methods and Literary History. University of Illinois Press, Urbana Krivitsky and Handcock [2014] Krivitsky PN, Handcock MS (2014) A separable model for dynamic networks. Journal of the Royal Statistical Society: Series B (Statistical Methodology) 76(1):29–46. 10.1111/rssb.12014 Momeni et al [2018] Momeni E, Karunasekera S, Goyal P, et al (2018) Modeling evolution of topics in large-scale temporal text corpora. Proceedings of the International AAAI Conference on Web and Social Media, 10.1609/icwsm.v12i1.15068 Moretti [2013] Moretti F (2013) Distant Reading. Verso, London Raji et al [2023] Raji S, Alikhani M, de Melo G, et al (2023) A corpus of persian literary text. Language Resources and Evaluation 10.1007/s10579-023-09689-6 Rasti Meymandi et al [2024] Rasti Meymandi A, Hosseini Z, Davari S, et al (2024) Opportunities for persian digital humanities research with artificial intelligence language models: Case study: Forough farrokhzad. arXiv preprint arXiv:2405.06760, 10.48550/arXiv.2405.06760 Rossetti and Cazabet [2018] Rossetti G, Cazabet R (2018) Community discovery in dynamic networks: a survey. ACM Computing Surveys 51(2):35:1–35:37. 10.1145/3172867 Ruma et al [2022] Ruma J, Akter S, Laboni J, et al (2022) A deep learning classification model for persian hafez poetry based on the poet’s era. Decision Analytics Journal 10.1016/j.dajour.2022.100111 Schimmel [1992] Schimmel A (1992) A Two-Colored Brocade: The Imagery of Persian Poetry. State University of New York Press, Albany Sewell and Chen [2015] Sewell DK, Chen Y (2015) Latent space models for dynamic networks. Journal of the American Statistical Association 110(512):1646–1657. 10.1080/01621459.2014.988214 Seyed-Gohrab [2011] Seyed-Gohrab A (ed) (2011) Metaphor and Imagery in Persian Poetry. Brill, Leiden Shahnazari et al [2025] Shahnazari K, Ayyoubzadeh SM, Fazli M, et al (2025) NAZM: Network analysis of zonal metrics in persian poetic tradition. arXiv preprint arXiv:2505.08052, 10.48550/arXiv.2505.08052 Tanaka [2025] Tanaka T (2025) Mean-reverting self-excitation drives evolution: Phylogenetic analysis of a literary genre, waka, with a neural language model. Humanities and Social Sciences Communications 10.1057/s41599-025-04714-1 Underwood [2019] Underwood T (2019) Distant Horizons: Digital Evidence and Literary Change. University of Chicago Press, Chicago Xu et al [2020] Xu J, Tao Y, Yan Y, et al (2020) Exploring evolution of dynamic networks via diachronic node embeddings. IEEE Transactions on Visualization and Computer Graphics 10.1109/TVCG.2018.2887230 Zhao et al [2022] Zhao J, Wei Y, Wu B (2022) Analysis of the social network and the evolution of the influence of ancient chinese poets. Social Science Computer Review 10.1177/08944393211028182 Appendix A Supplementary Appendix The supplementary materials provide compact support tables and a broader family inventory. They are meant to keep the symbolic inventory legible without repeating low-level variant detail. A.1 Scoring, uncertainty, and consultation summary Two points are worth stating directly. The unique atlas total reported in Table 1 is 157157 because graph gating assigns each retained family to one primary residency: 127127 in the imagistic core and 3030 in the referential layer. The bridge layer is an overlapping participation view on that same set, not an additional disjoint inventory. We also discussed the family-level framing with a specialist in Persian literary studies as a plausibility check rather than a large formal annotation campaign. Table 7: Compact scoring summary for family induction and edge construction. Stage Positive evidence Penalties and gates Output rule Family induction Eligible surface relation; static and occurrence-level embeddings; contextual and cluster overlap; seed and exemplar support; head cleanliness Fragment residue, function-word contamination, genericity noise, and proper-name inflation lower the merge score; staged thresholds keep weak cases as singletons Each token maps to one family; accepted heads default to the cleanest root form, with synthetic heads admitted only when several strong variants jointly clear the synthetic threshold Edge construction Multiview embedding support; weighted context overlap; cluster overlap; representative-context overlap; verse/document co-association; poet overlap; seed support; local co-activation Genericity is penalized in every layer; referentiality is negative in the core, near-neutral in the referential layer, and slightly positive in the bridge layer; layer thresholds and top-k sparsification prune generic neighborhoods more aggressively Each retained family keeps one primary atlas residency, while bridge participation records which retained families also sustain cross-layer attachment edges Table 8: Supplementary uncertainty summary for representative families and dynamic claims. Target Point estimate Uncertainty summary Summary Peymaaneh (Wine Vessel) Slope +2.51+2.51 95%95\% bootstrap CI [+1.37,+3.40][+1.37,+3.40]; peak-share 76%76\% Late strengthening remains stable under poet-bootstrap resampling. Sahbaa (Wine) Slope +0.60+0.60 95%95\% bootstrap CI [+0.29,+0.92][+0.29,+0.92]; peak-share 78%78\% The rise is smaller than Peymaaneh (Wine Vessel) but remains consistent. Golshan (Garden) Slope +5.06+5.06 95%95\% bootstrap CI [+3.53,+6.82][+3.53,+6.82]; peak-share 81%81\% Garden-space intensification is one of the clearest late family gains. Khaaqaan (Sovereign) Slope −1.74-1.74 95%95\% bootstrap CI [−2.92,−0.14][-2.92,-0.14]; peak-share 63%63\% Early concentration is consistent, even if the exact peak century is not knife-edged. Farrokh (Auspicious) Slope −2.06-2.06 95%95\% bootstrap CI [−3.06,−0.39][-3.06,-0.39]; peak-share 93%93\% Prestige-coded decline is unusually stable among the declining families. Baad (Wind) Slope −19.31-19.31 95%95\% bootstrap CI [−27.97,−11.27][-27.97,-11.27]; peak-share 67%67\% Mobility imagery is earlier-weighted with a strongly negative slope. Shab (Night) Slope +2.09+2.09 95%95\% bootstrap CI [−2.51,+7.24][-2.51,+7.24]; peak-share 51%51\% The family is temporally diffuse rather than sharply periodized, so it is treated as stable rather than monotonic. Khaak (Earth) Slope +7.03+7.03 95%95\% bootstrap CI [+2.08,+11.20][+2.08,+11.20]; peak-share 90%90\% Later weighting is real, but the family remains distributed across the whole chronology and still functions as an anchor. Core modularity Δ+0.020 +0.020 Comparable-variant span +0.012+0.012 to +0.030+0.030; permutation upper-tail 0.140.14 A modest but consistently positive structural separation signal. Cross-link ratio Δ−0.032 -0.032 Comparable-variant span −0.032-0.032 to −0.019-0.019; permutation lower-tail 0.050.05 The decline repeats across the main reconstructions. Courtly bridge strength Δ−0.033 -0.033 Variant span −0.033-0.033 to −0.007-0.007; flat in 2/42/4 variants Courtly weakening is visible but should be read cautiously. Sacred bridge strength Δ+0.049 +0.049 Variant span +0.042+0.042 to +0.071+0.071 Sacred-prophetic strengthening is the most secure bridge-level shift. A.2 Supplementary family inventory Table LABEL:tab:appendix_inventory gathers a larger set of representative families across the principal symbolic fields. The selection is not exhaustive. It is meant to show recognizable Persian forms, transliterations, and English glosses while keeping the family-level scale intact. Table 9: Supplementary inventory of representative symbolic families. Family Atlas role Temporal profile Note ساغر / Saaghar (Wine Cup) Core / Wine-Tavern Ritual Rising A vessel image that remains central even as late lyric clustering intensifies. پیمانه / Peymaaneh (Wine Vessel) Core / Wine-Tavern Ritual Rising A measure-and-intoxication family that strengthens with later lyric concentration. گلشن / Golshan (Garden) Core / Floral-Vegetal Rising Garden-space imagery expands while remaining lexically transparent at the family level. خرقه / Kherqe (Sufi Robe) Core / Ascetic-Mystical Rising A late-strengthening ritual family that becomes structurally prominent within the mystical field. خرابات / Kharaabaat (Tavern) Core / Ascetic-Mystical Peak Middle A tavern family marking the field’s recurrent traffic with wine imagery. بنفشه / Banafsheh (Violet) Core / Body-Beloved Declining A beloved-image family weighted more strongly toward earlier and middle centuries. خاقان / Khaaqaan (Sovereign) Referential / Courtly Bridge Declining A prestige head retained as a distinct courtly family rather than diffused into imagistic clusters. فرخ / Farrokh (Auspicious) Core / Mixed Field Declining Auspicious prestige language that loses weight across the later chronology. شب / Shab (Night) Core / Mixed Field Stable A durable lyric anchor with wide chronological spread. روز / Ruz (Day) Core / Mixed Field Stable A companion temporal anchor that remains broadly distributed. خاک / Khaak (Earth) Core / Mixed Field Stable Ethical and metaphysical weight gathers around a highly persistent elemental head. دریا / Daryaa (Sea) Core / Water-Sea Stable A broad elemental family that remains recognizably tied to the same symbolic actor. آتش / Aatash (Fire) Core / Light-Fire Stable An enduring intensity family linking flame, affect, and illumination. A.3 Field structure and robustness Table 10 relocates the family illustrations removed from the main community graph into tabular form. The wine-tavern field is the most persistent large field, the floral-vegetal field shows the strongest internal splitting pressure, and the ascetic-mystical field is most often pulled toward wine imagery. Table 10: Supplementary field inventory and structural tendencies. Field Illustrative families Mean active-member ratio Summary Wine-Tavern Ritual Saaghar (Wine Cup); Baadeh (Wine); Qadah (Goblet) 0.990 The most persistent large field in the atlas, with low split pressure and strong late continuity. Floral-Vegetal Gol (Rose); Golshan (Garden); Sarv (Cypress) 0.965 The field remains active throughout the chronology but shows the strongest internal differentiation. Body-Beloved Banafsheh (Violet); Sonbol (Hyacinth); Meshkin (Musky) 0.960 Bodily beauty imagery persists strongly while repeatedly leaning toward the floral field. Ascetic-Mystical Kherqe (Sufi Robe); Meykadeh (Wine House); Kharaabaat (Tavern) 0.857 The most visibly intermediary field, with recurrent outward pull toward wine-tavern ritual. Water-Sea Daryaa (Sea); Bahr (Sea); Qolzum (Sea) 0.980 A highly continuous elemental field with little internal fragmentation. Light-Fire Aatash (Fire); Forugh (Glow); Partow (Radiance) 0.985 An unusually stable field whose members retain a common intensity profile across bins. Mixed Field Shab (Night); Ruz (Day); Khaak (Earth) 0.885 A durable contact zone that repeatedly absorbs bridge attachment and historical retargeting. Across loose, base, strict, and poet-balanced reconstructions, the direction of the central dynamic claims remains stable: core modularity rises, cross-scope linkage declines, and sacred-prophetic attachment strengthens. The most cautious result continues to be courtly weakening, which remains negative in the base and poet-balanced reconstructions but flattens under looser and stricter activity requirements. The smallest high-confidence-only subgraph preserves the broad tendency toward separation, but it becomes too sparse to sustain the full contrast structure on its own. A.4 Supplementary graph view Figure A1 adds one more graph view without crowding the main text. It contrasts an earlier dense core snapshot with a later one. The later structure is more differentiated, but it remains continuous with the earlier network. Figure A1. Supplementary early and late core snapshots. Community colors follow Figure 3; the later snapshot shows the sharper differentiation of field structure discussed in the main text. A.5 Supplementary poet-level evidence The threshold-based core poet pool used in Figure 7 contains 3636 poets with at least 10,00010,000 verses and 1,0001,000 symbolic-family occurrences. This rule is coverage-driven rather than prestige-driven. It includes Saadi, Khaaqani, and Vahshi Bafqi, but excludes Hafez, whose mapped corpus in the present dataset reaches 5,1835,183 verses and falls below both the core and sensitivity floors. A broader 5858-poet sensitivity pool at 8,0008,000 verses and 500500 symbolic occurrences leaves the century-effect contrast unchanged. Tables LABEL:tab:appendix_poet_profiles–LABEL:tab:appendix_poet_neighbors provide the full verse-normalized field profiles, cluster assignments, and nearest symbolic neighbors for the core pool. Figure A2 adds a low-dimensional view of the same standardized field space. The first two principal components capture 74.0%74.0\% of the variance and make the cluster geometry easier to inspect: Khaaqani and Vahshi Bafqi remain close to the symbolic center, whereas Saeb, Jahan Malek Khatun, and Ashofteh Shirazi separate most strongly from the pooled mean profile. Table 11: Verse-normalized poet-field profiles for the threshold-based core poet pool. The pool retains poets with at least 10,000 verses and 1,000 symbolic-family occurrences. WT = Wine-Tavern, AM = Ascetic-Mystical, B = Body-Beloved, FV = Floral-Vegetal, LF = Light-Fire, WS = Water-Sea, MF = Mixed Field. Cluster assignments are reported separately in Table LABEL:tab:appendix_poet_neighbors. Poet C. Verses Sym. WT AM B FV LF WS MF Amir Khosrow Dehlavi 7 21,599 19,843 1714.9 170.8 1387.6 885.7 333.3 153.7 4541.0 Jahan Malek Khatun 8 12,276 12,720 165.4 31.8 1376.7 1770.9 432.6 127.1 6457.3 Khwaju Kermani 8 29,301 19,768 1128.3 158.4 1079.5 771.0 414.3 297.9 2897.2 Salman Savaji 8 13,129 10,573 844.7 207.9 1352.7 1132.6 401.4 320.7 3793.1 Seyf Farghani 8 10,279 8,205 803.6 74.9 1192.7 917.4 583.7 180.0 4230.0 Jami 9 39,753 26,396 1129.7 218.3 999.2 739.3 441.0 203.8 2908.7 Ahli Shirazi 10 16,064 13,948 1012.8 262.7 1194.6 1289.8 692.2 180.5 4050.0 Saeb 11 77,570 89,285 3250.2 213.0 1624.6 1338.4 666.5 771.0 3646.5 Hazin Lahiji 12 14,480 13,904 2138.8 367.4 1195.4 953.0 627.1 341.9 3978.6 Ashofteh Shirazi 13 12,767 11,349 1178.8 604.7 1294.7 1077.8 623.5 246.7 3863.1 Amir Mu’izzi 5 18,406 9,220 305.3 31.0 690.5 420.0 439.5 286.3 2836.6 Mas’ud Sa’d Salman 5 15,048 8,109 536.9 20.6 534.3 579.5 585.5 249.2 2882.8 Sana’i 5 26,079 11,615 265.3 126.2 524.9 300.6 380.4 148.8 2707.5 Anvari 6 13,308 5,946 248.7 32.3 401.3 310.3 259.2 214.9 3001.2 Attar 6 94,934 38,521 257.3 81.3 316.6 218.8 507.4 335.7 2340.5 Khaaqani 6 17,434 9,568 448.5 122.2 792.1 311.5 513.9 273.6 3026.3 Hakim Nezari 7 18,402 10,549 1431.4 172.8 501.6 380.9 301.6 144.0 2800.2 Mowlana 7 66,196 34,220 638.3 117.8 618.9 400.0 534.2 346.1 2514.2 Saadi 7 16,159 6,939 140.5 104.0 648.6 547.1 199.9 125.0 2529.2 Ibn Yamin 8 13,792 7,811 525.7 87.7 855.6 710.6 470.6 249.4 2763.9 Owhadi 8 14,683 8,760 345.3 149.2 869.0 506.0 397.7 96.0 3602.8 Mohtasham Kashani 10 10,716 5,985 286.5 42.9 911.7 691.5 465.7 328.5 2858.3 Vahshi Bafqi 10 10,596 6,058 380.3 100.0 699.3 668.2 448.3 260.5 3160.6 Bidel Dehlavi 11 32,333 17,434 462.1 63.7 834.1 634.6 458.0 586.1 2353.3 Fayyaz Lahiji 11 10,204 7,630 612.5 128.4 1004.5 1126.0 691.9 472.4 3441.8 Feyz Kashani 11 11,240 8,111 815.8 191.3 781.1 409.3 556.0 232.2 4230.4 Molla Ahmad Naraqi 13 11,356 5,202 548.6 115.4 443.8 383.9 685.1 250.1 2153.9 Qa’ani 13 20,343 9,561 532.9 72.3 884.3 516.1 390.3 336.2 1967.8 Reza-Qoli Khan Hedayat 13 10,028 5,179 444.8 324.1 386.9 240.3 523.5 431.8 2813.1 Ferdowsi 4 49,610 12,502 146.1 2.2 192.7 136.3 111.3 127.4 1804.1 Iranshahan 5 20,825 5,573 121.0 1.9 155.6 123.9 126.3 253.5 1893.9 Nasir Khusraw 5 10,944 3,039 145.3 50.3 325.3 287.8 246.7 120.6 1600.9 Nizami 6 27,768 9,392 266.9 32.1 415.9 361.6 315.8 189.4 1800.6 Elhami Kermanshahi 13 16,795 5,371 216.7 36.9 350.7 206.0 235.8 111.9 2039.9 Safi Alishah 13 44,939 10,124 677.4 72.1 182.7 53.6 239.4 163.3 864.3 Adib al-Mamalek 14 15,562 5,894 468.4 68.8 455.6 447.9 286.6 241.0 1819.2 Table 12: Cluster membership and nearest symbolic neighbors in standardized field space. Smaller mean-distance ranks indicate more central positions, while larger ranks indicate more outlying profiles. Poet Cluster Nearest neighbor Second neighbor Mean-dist. rank Amir Khosrow Dehlavi Lyric-amplifying Salman Savaji Jami 31 Jahan Malek Khatun Lyric-amplifying Seyf Farghani Salman Savaji 35 Khwaju Kermani Lyric-amplifying Jami Ibn Yamin 9 Salman Savaji Lyric-amplifying Khwaju Kermani Jami 24 Seyf Farghani Lyric-amplifying Feyz Kashani Ibn Yamin 22 Jami Lyric-amplifying Khwaju Kermani Ibn Yamin 12 Ahli Shirazi Lyric-amplifying Seyf Farghani Salman Savaji 32 Saeb Lyric-amplifying Hazin Lahiji Fayyaz Lahiji 36 Hazin Lahiji Lyric-amplifying Ahli Shirazi Ashofteh Shirazi 33 Ashofteh Shirazi Lyric-amplifying Hazin Lahiji Ahli Shirazi 34 Amir Mu’izzi Mediating-balanced Vahshi Bafqi Mohtasham Kashani 4 Mas’ud Sa’d Salman Mediating-balanced Amir Mu’izzi Vahshi Bafqi 10 Sana’i Mediating-balanced Anvari Nizami 7 Anvari Mediating-balanced Nizami Elhami Kermanshahi 15 Attar Mediating-balanced Mowlana Amir Mu’izzi 16 Khaaqani Mediating-balanced Mowlana Amir Mu’izzi 3 Hakim Nezari Mediating-balanced Sana’i Jami 19 Mowlana Mediating-balanced Khaaqani Attar 5 Saadi Mediating-balanced Sana’i Anvari 18 Ibn Yamin Mediating-balanced Vahshi Bafqi Mohtasham Kashani 2 Owhadi Mediating-balanced Sana’i Vahshi Bafqi 13 Mohtasham Kashani Mediating-balanced Ibn Yamin Vahshi Bafqi 6 Vahshi Bafqi Mediating-balanced Ibn Yamin Amir Mu’izzi 1 Bidel Dehlavi Mediating-balanced Qa’ani Mohtasham Kashani 25 Fayyaz Lahiji Mediating-balanced Mohtasham Kashani Salman Savaji 27 Feyz Kashani Mediating-balanced Khaaqani Vahshi Bafqi 17 Molla Ahmad Naraqi Mediating-balanced Mowlana Mas’ud Sa’d Salman 20 Qa’ani Mediating-balanced Mohtasham Kashani Amir Mu’izzi 8 Reza-Qoli Khan Hedayat Mediating-balanced Mowlana Attar 26 Ferdowsi Sparse-anchor Iranshahan Elhami Kermanshahi 30 Iranshahan Sparse-anchor Ferdowsi Elhami Kermanshahi 28 Nasir Khusraw Sparse-anchor Elhami Kermanshahi Nizami 23 Nizami Sparse-anchor Adib al-Mamalek Nasir Khusraw 14 Elhami Kermanshahi Sparse-anchor Nasir Khusraw Nizami 21 Safi Alishah Sparse-anchor Nasir Khusraw Elhami Kermanshahi 29 Adib al-Mamalek Sparse-anchor Nizami Nasir Khusraw 11 Figure A2. Supplementary projection of poet similarity space. Points show the threshold-based core poet pool in the first two principal components of the standardized field profiles; faint lines connect nearest neighbors, and colors follow the three clusters shown in Figure 7.