Cultural Shift or Linguistic Drift? Comparing Two Computational Measures of Semantic Change (1606.02821v2)

Abstract: Words shift in meaning for many reasons, including cultural factors like new technologies and regular linguistic processes like subjectification. Understanding the evolution of language and culture requires disentangling these underlying causes. Here we show how two different distributional measures can be used to detect two different types of semantic change. The first measure, which has been used in many previous works, analyzes global shifts in a word's distributional semantics, it is sensitive to changes due to regular processes of linguistic drift, such as the semantic generalization of promise ("I promise." -> "It promised to be exciting."). The second measure, which we develop here, focuses on local changes to a word's nearest semantic neighbors; it is more sensitive to cultural shifts, such as the change in the meaning of cell ("prison cell" -> "cell phone"). Comparing measurements made by these two methods allows researchers to determine whether changes are more cultural or linguistic in nature, a distinction that is essential for work in the digital humanities and historical linguistics.

• The paper introduces a dual approach comparing global and local word embedding measures to differentiate gradual linguistic drift from abrupt cultural shifts.
• It shows that local measures better capture noun changes influenced by cultural innovation, while global measures more effectively track verb evolution.
• The study's findings support a domain-specific selection of semantic change detection tools for digital humanities and historical linguistics research.

Analyzing Semantic Change: A Dual Perspective on Cultural and Linguistic Shifts

The paper "Cultural Shift or Linguistic Drift? Comparing Two Computational Measures of Semantic Change" presents a refined approach to understanding semantic change through the application of two distinct distributional measures. These measures aim to disentangle semantic changes driven by cultural transformations from those spurred by intrinsic linguistic evolution, a crucial task for digital humanities and historical linguistics.

Methodological Framework

At the core of this paper is the comparison of two measures: a global measure and a local neighborhood measure, both applied to diachronic word embeddings. The global measure assesses changes by analyzing shifts in a word's overall vector semantics across time periods. It's well-suited to capturing gradual semantic drift, such as shifts occurring due to subjectification. The local neighborhood measure, on the other hand, evaluates changes by examining alterations in a word's nearest semantic neighbors, thereby aligning more closely with abrupt cultural changes.

Results and Analysis

The paper, grounded in comprehensive statistical analyses across six historical corpora in multiple languages, underscores a cardinal distinction in how these measures capture semantic evolution. The local neighborhood measure exhibits heightened sensitivity to changes in nouns, a part of speech frequently driven by cultural shifts owing to new technologies or societal changes. Conversely, the global measure more effectively captures changes in verbs, which are typically subject to regular processes of linguistic drift.

This dichotomy is further illustrated through a series of case studies. The authors elaborate on examples such as "gay," "virus," and "cell" underlining cultural shifts effectively detected by local neighborhood changes. Conversely, words like "actually" and "must," which have undergone well-documented regular linguistic shifts, highlight the efficacy of the global measure.

Implications for Research and Practice

The findings advocate for a domain-specific approach in selecting computational tools for semantic change detection. For applications focused on cultural phenomena, such as digital humanities research, employing the local neighborhood measure could provide significant insights. In contrast, for linguistic studies concentrating on phenomena like grammaticalization, the global measure aligns more closely with research needs.

Furthermore, by juxtaposing these measures, researchers can gain nuanced insights into the genesis of semantic change, distinguishing between cultural and linguistic influences. This has substantial theoretical implications for the understanding of language evolution and practical impacts on the design of natural language processing systems sensitive to historical language dynamics.

Future Directions

The methodology and findings open several avenues for future work. Refinement of computational models to enhance sensitivity and accuracy of these measures could broaden applicability. Moreover, expanding the analysis to encompass additional languages or less-resourced linguistic contexts may validate the universality of the proposed measures' sensitivity distinctions. As AI continues to advance, integrating these insights into machine learning frameworks could potentiate automated systems' ability to detect and understand historical linguistic changes.

In summary, this paper presents a robust framework that successfully differentiates between cultural and linguistic drivers of semantic change, providing a potent tool for researchers in both computational linguistics and digital humanities.