cs.CLApr 8, 2026

Between Century and Poet: Graph-Based Lexical Semantic Change in Persian Poetry

Authors: Kourosh Shahnazari, Seyed Moein Ayyoubzadeh, Mohammadali Keshtparvar

Organizations: Sharif University of Technology · Amirkabir University of Technology

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Meaning in Persian poetry is both historical and relational. Words persist through literary tradition while shifting their force through changing constellations of neighbors, rhetorical frames, and poetic voices. This study examines that process using aligned Word2Vec spaces combined with graph-based neighborhood analysis across centuries and major poets. Rather than modeling semantic change as vector displacement alone, it treats lexical history as the rewiring of local semantic graphs: the gain and loss of neighbors, shifts in bridge roles, and movement across communities. The analysis centers on twenty target words, anchored by five recurrent reference terms: Earth, Night, two wine terms, and Heart. Surrounding them are affective, courtly, elemental, and Sufi concepts such as Love, Sorrow, Dervish, King, Annihilation, and Truth. These words exhibit distinct patterns of change. Night is more time-sensitive, Earth more poet-sensitive, and Heart shows continuity despite graph-role mobility. The two wine terms highlight probe sensitivity: one is broad and semantically diffuse, while the other is narrower and more stable. A lexical audit confirms that the corpus contains historically driven terms, poet-specific usages, and sparsely attested mystical vocabulary requiring caution. Overall, semantic change in Persian poetry is better captured as neighborhood rewiring than as abstract drift. For Digital Humanities, this approach restores local structure to computational analysis and supports interpretations closer to literary practice: persistence, migration, mediation, and selective transformation.

Figures & tables

Figure 1: Mean embedding drift against mean neighbor turnover for all twenty target words. Colors distinguish the recurrent reference words, the broader symbolic field, and the explicitly mystical layer; open diamonds mark words whose poet-side coverage remains thin.
Figure 2: Normalized component profile for the full word set, ordered by century-side signal. The columns report drift, turnover, graph-role volatility, century signal, and poet signal; a dagger marks thin poet-side coverage.
Most stable relational profilesStrongest historical reshapingStrongest poet imprint
WordScoreWordScoreWordRatio
eshgh / Love0.31mey / Wine0.79fanaa / Annihilation0.59
gham / Sorrow0.31soofi / Sufi0.67eshgh / Love0.60
rah / Path0.39darvish / Dervish0.62haqiqat / Truth0.62
baadeh / Wine0.40tariqat / Path0.60tariqat / Path0.64
del / Heart0.43rooz / Day0.59darvish / Dervish0.67
Table 1: Ranking summary for the full target set. The left block lists the most compact relational profiles, the middle block the strongest century-side reshaping, and the right block the words whose poet imprint most clearly outweighs the century signal.
Figure 3: Century-wise graph-role trajectories for all twenty target words. The upper heatmap retains absolute century values, while the lower heatmap centers each century on its mean so positive and negative values mark relative prominence within that period. Century 3 is omitted because its exceptionally thin graph does not sustain comparison with the denser periods that follow.
Figure 4: Century-wise graph-role trajectories for the full word set, continued.
Figure 5: Transition-level dynamics for the full target set. The upper panel reports neighbor turnover between adjacent centuries; the lower panel reports the extent of community reallocation, measured as one minus the overlap between adjacent communities.
Figure 6: Local century-to-century drift, deviation from the full-corpus reference model, and their agreement profile for all twenty target words. Adjacent drift is assigned to its resulting century. The lower panel classifies each word-period pairing by the joint behavior of the two measures, distinguishing stable usage, local fluctuation, robust change, and settled departure.
Figure 7: Overall similarity structure across the poet set, shown as a centered version of the raw mean-cosine matrix. Positive values indicate poet pairs whose affinity exceeds each poet’s general similarity to the wider set; negative values indicate comparatively weaker affinity.
WordDriftTurnoverMean hubHub dev.PressureLiterary mechanism
khaak / Earth0.830.800.002-0.000MixedThreshold, abasement, and materiality are redistributed by poet-specific ethical stance.
shab / Night0.830.690.002+0.000More time-sensitiveThe nocturnal field stays shared, but its historical weighting shifts with separation, darkness, and vigil.
mey / Wine0.820.940.001-0.001More time-sensitiveA broad symbolic probe absorbs convivial, mystical, satirical, and lyrical registers at once.
baadeh / Wine0.900.780.002+0.001More time-sensitiveA narrower wine lexicon keeps a more coherent convivial cluster across the corpus.
del / Heart0.740.800.002+0.000More time-sensitiveAn enduring affective center persists while its internal attachments are repeatedly rewired.
Table 2: Comparative profile of the five recurrent reference words. Mean hub reports raw degree centrality across viable century graphs; hub deviation reports the average departure from the century mean, so positive values indicate words that tend to sit above the prevailing hub level of their period.
Figure 8: Direct comparison between mey / Wine and baadeh / Wine . Both remain historically active, but baadeh / Wine keeps a narrower convivial neighborhood while mey / Wine disperses across a broader symbolic field.
Figure 9: Direct comparison of century-side and poet-side signal for all twenty target words. The bar panel compares the composite signal scores, and the plane situates the lexical panel across time-sensitive, mixed, and poet-sensitive regions.
WordFieldDriftTurnoverGraph roleDominant pressure
Reference words
khaak / EarthReference words0.830.80Community-migrantMixed
shab / NightReference words0.830.69Community-migrantMore time-sensitive
mey / WineReference words0.820.94Community-migrantMore time-sensitive
baadeh / WineReference words0.900.78Community-migrantMore time-sensitive
del / HeartReference words0.740.80Community-migrantMore time-sensitive
Table 3: Summary of the full lexical panel. Drift and turnover are mean adjacent century-side values; the graph-role column summarizes the dominant trajectory shape, and the pressure column reports whether century-side or poet-side differentiation is stronger.
More in this topic

Semantic Change Detection

Semantic change detection identifies how meanings, concepts, or scene contents differ across time in text, images, and other data.

See all 62 papers

Explore similar work

CardsList

  1. A Dynamic Atlas of Persian Poetic Symbolism: Families, Fields, and the Historical Rewiring of Meaning

    Apr 1, 2026Kourosh Shahnazari, Seyed Moein Ayyoubzadeh, Mohammadali KeshtparvarSemantic Change DetectionComputational Literary Studies

    Sharif University of Technology, Tehran, Iran. · Amirkabir University of Technology, Tehran, Iran.

  2. Echoes Across Centuries: Phonetic Signatures of Persian Poets

    Mar 15, 2026Kourosh Shahnazari, Seyed Moein Ayyoubzadeh, Mohammadali Keshtparvar

    Sharif University of Technology, Tehran, Iran. · Amirkabir University of Technology, Tehran, Iran.

  3. Eigenmood Space: Uncertainty-Aware Spectral Graph Analysis of Psychological Patterns in Classical Persian Poetry

    Feb 18, 2026Kourosh Shahnazari, Seyed Moein Ayyoubzadeh, Mohammadali KeshtparvarSpectral Graph EmbeddingComputational Literary Studies

    Sharif University of Technology, Tehran, Iran. · Amirkabir University of Technology, Tehran, Iran.