Making “concreteness” more concrete

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Science & Research — Social Sciences & Behavioral Studies, Mathematics & Computational Sciences · Depth: Expert, short

Summary

A 2026 paper by Yanting Li, Gregory Scontras, and Richard Futrell, titled "Making "concreteness" more concrete," challenges the common assumption that concrete words (e.g., "apple") share more cross-linguistic semantic features than abstract words (e.g., "appetite"). The researchers tested this hypothesis by employing multilingual aligned word embeddings to measure the distance between words and their nearest neighbors in other languages. They then examined if shorter distances correlated with higher concreteness ratings across six languages: Dutch, English, French, Cypriot Greek, Mandarin, and Portuguese. Their findings indicate that the relationship between concreteness and cross-linguistic distance was not uniform, varying significantly across the languages studied. This suggests that concreteness does not consistently correspond to cross-linguistic semantic relatedness. The study also demonstrates the potential of using aligned word embeddings to operationalize psycholinguistic constructs.

Key takeaway

For NLP Engineers or Research Scientists working on multilingual semantic models, you should reconsider assumptions about universal semantic feature sharing for concrete words. Your models might need to account for language-specific variations in how concreteness manifests cross-linguistically, rather than assuming uniform semantic relatedness. This suggests a need for more nuanced, language-aware approaches when designing or evaluating multilingual word representations.

Key insights

Cross-linguistic semantic relatedness of words does not uniformly predict their concreteness ratings.

Principles

• Concreteness varies cross-linguistically.
• Semantic features are not uniformly shared.
• Aligned embeddings operationalize constructs.

Method

Multilingual aligned word embeddings were used to measure cross-linguistic word distances. These distances were then correlated with psycholinguistic concreteness ratings across six diverse languages.

In practice

• Apply aligned embeddings to psycholinguistics.
• Re-evaluate cross-linguistic semantic hypotheses.

Topics

• Multilingual Word Embeddings
• Cross-linguistic Semantics
• Psycholinguistics
• Concreteness Ratings
• Semantic Features
• Natural Language Processing

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• Universal Conceptual Structure in Neural Translation: Probing NLLB-200's Multilingual Geometry — NLLB-200's internal geometry reflects language-universal conceptual structures, akin to human multilingual cognition.
• When Similar Means Different: Evaluating LLMs on Arabic--Hebrew Cognates — Current LLMs struggle with cross-lingual semantic disambiguation, especially false friends and loanwords, due to over-reliance on surface forms.
• Global vs. Local Sentence Embeddings for Brazilian Portuguese: Revisiting Monolingual Models in the Age of Foundation Models — Optimal model choice for PT-BR sentence embeddings balances vocabulary coverage and adaptation flexibility.
• Scene Abstraction for Lexical Semantics: Structured Representations of Situated Meaning — "Scene Abstraction" provides a structured framework to represent the situated, interpretive dimensions of lexical meaning using LLMs.
• Multilinguality of Large Language Models From a Structural Perspective — LLMs' multilinguality involves structural differences, with low-resource languages diverging more from English.
• The Cosine Similarity Trap: Why Embeddings Can’t Distinguish “War” from “Union” — Language models often cannot distinguish complementarity from conflict due to how embeddings encode relatedness, not relationship.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.