Diachronic Linguistic Patterns

• Diachronic linguistic patterns describe systematic historical shifts in word usage and form, quantified through measured frequency changes and co-occurrence analysis.
• Quantitative methods use frequency metrics, PPMI-based context analysis, and advection correction to separate intrinsic language change from topical drift.
• Empirical studies show topical advection accounts for 25%–30% of frequency variance, highlighting its role in understanding lexical innovation.

Diachronic linguistic patterns constitute the structural, distributional, and semantic regularities by which linguistic elements—words, forms, constructions—systematically shift in form and usage over historical time. Quantitative diachronic linguistics leverages large-scale corpora and computational models to isolate and explain these shifts, distinguishing genuine linguistic change from confounding factors such as topical drift. Core methodologies span frequency-based metrics, co-occurrence and topic modeling, embedding dynamics, and hybrid neural-symbolic architectures.

1. Theoretical Foundations and Diachronic Confounds

Corpus-based diachronic analysis traditionally interprets changes in the frequency of linguistic elements as evidence for selection (via social prestige, processing ease) or stochastic drift. However, frequency profiles are influenced not only by intrinsic linguistic competition or innovation but by fluctuations in the prominence of underlying discourse topics. For example, if a topic such as "computers" surges in the public sphere, all semantically associated words (e.g., “microchip”, “modem”) will rise in frequency, even absent intrinsic linguistic drivers. Uncontrolled, this topical covariance risks misattributing topic-driven surges to linguistic mechanisms of selection or drift (Karjus et al., 2018).

2. The Topical-Cultural Advection Model

To quantify and subtract the effect of topical drift, the topical-cultural advection model provides a robust, mathematically rigorous baseline for frequency change attributable solely to shifting extra-linguistic topics. Given a target word $\omega$ and time period $t$, one computes

$$\mathrm{advection}(\omega;t) = \frac{ \sum_{i=1}^m \mathrm{PPMI}(\omega,N_i) \, [\ln(f(N_i;t) + s) - \ln(f(N_i;t-1) + s)] }{ \sum_{i=1}^m \mathrm{PPMI}(\omega,N_i) }$$

where $N_i$ are the $m$ most strongly associated context words (high positive pointwise mutual information) and $s$ is a smoothing constant. This measure captures the mean, PPMI-weighted log-change of a word’s discourse neighborhood.

The model workflow is as follows:

• Corpus division into contiguous time slices (e.g., decades).
• POS-tagging, filtering to content words, normalization to pmw.
• Calculation of each word’s log-change across periods.
• Construction of co-occurrence matrices (window $\pm 10$), PPMI weighting.
• Selection of each target’s top $m$ PPMI neighbors per period.
• Computation of advection for each target/period.
• Advection correction by subtracting topic drift from raw log-change:

$$x(\omega;t) = \mathrm{logChange}(\omega;t) - \mathrm{advection}(\omega;t)$$

where $x(\omega;t)$ captures the residual, plausibly selection-driven change.

Empirical validation on COHA (Corpus of Historical American English, 1810–2009) demonstrates that, after smoothing, advection explains on average 25%–30% of frequency variance for common nouns. In an artificial genre-shift setting (COCA, academic → spoken subcorpora), advection accounts for $R^2=0.73$ of variance, confirming its utility in recovering known topic or stylistic transitions (Karjus et al., 2018).

3. Diachronic Patterns: Lexical Innovation, Diffusion, and Topical Coupling

Diachronic advection analysis shows that lexical innovations are strongly synchronized with rising topic advection. In a curated set of 73 neologistic nouns (1970s–2000s), 58% appear when their parent topic’s advection exceeds the upper 95% bound of its own historical “topic drift,” 37% synchronize with the mean, and only 5% appear during topical stagnation (one-sample $t$-test $p<0.001$). This suggests that the communicative need in a rising subspace of the lexicon orchestrates the timing and uptake of lexical innovation.

Qualitative tracing illustrates varying innovation–topic alignments:

Innovation	Period	Topic Advection	Semantic Interpretation
microchip	1970s	Well above topic mean	Coincides with surge in microelectronics
pantsuit	1970s	Near topic mean	Moderate communicative need
narratology	1980s	Well below topic mean	Niche, low-need academic uptake

Notably, the majority of "successful" new nouns correspond to moments of pronounced communicative need as quantified by topic advection (Karjus et al., 2018).

4. Methodological Implications and Corrections

By establishing the advection-corrected baseline, one can meaningfully apply genetic-inspired statistical tests to differentiate between drift and selection in lexical timeseries. After subtracting the topical baseline, residual outperformance ($x(\omega;t)>0$) or underperformance ($x(\omega;t)<0$) of topic drift becomes interpretable as (possibly) selection-like pressures or competitive displacement.

Further, decomposed and advection-corrected time series $ŷ(\omega;t)$ (obtained via cumulative sum of $x(\omega;τ)$) serve as de-trended frequency profiles, isolating frequency dynamics plausibly driven by factors intrinsic to the lexicon or grammar, rather than topic cycles.

This framework has been validated across raw PPMI-based context models and LDA topic models with highly similar $R^2$ performance (≈0.24–0.25 after smoothing), underscoring model robustness to topic inference formalism (Karjus et al., 2018).

5. Limitations and Potential Extensions

The advection model presupposes large, chronologically balanced corpora to estimate stable topic neighborhoods; low-frequency targets or short periods may benefit from context smoothing. Extremely high-polysemy or general-purpose terms may have diffuse or noisy topic vectors, suggesting scope for multi-sense or dynamic topic modeling approaches. The model’s cross-lexical (not self-temporal) focus means it captures the momentum of topic spaces, not intrinsic autocorrelation in an individual word's usage trajectory.

Possible extensions include:

• Integrating dynamic topic models (e.g., DTM, DLDA) for better longitudinal topic tracking.
• Adapting the framework to other cultural objects (e.g., memes, names, artifacts) wherever co-occurrence contexts are definable.
• Combining with detailed sociolinguistic metadata to in future separate cultural from social regime shifts (Karjus et al., 2018).

6. Broader Impact and Theoretical Relevance

The advection-based decomposition sharpens the precision of diachronic linguistic inquiry:

• Demarcates innovation from mere topical drift, reducing risk of attributing culturally induced frequency surges to internal linguistic forces.
• Supplies a corpus-internal null model for drift and selection tests, facilitating quantitative studies of linguistic evolution.
• Quantifies the proposition that vocabulary growth parallels rising communicative need in expanding topical subspaces.
• Empirically captures the punctuated, need-driven mechanism of lexical expansion, supplementing classical uniformitarian drift paradigms.

By operationalizing the portion of observed linguistic change due wholly to topical fluctuation, the advection model provides a mathematically transparent foundation for interpreting diachronic patterns in frequency, innovation, and selection under explicit control for cultural dynamics (Karjus et al., 2018).