Semantic Structure and Interpretability of Word Embeddings

By Lutfi Kerem Senel, Ihsan Utlu, Veysel Yücesoy, Aykut Koç, Tolga Cukur. 2017

118 citations recorded by the source corpus.

Dense word embeddings, which encode meanings of words to low-dimensional vector spaces, have become very popular in natural language processing (NLP) research due to their state-of-the-art performances in many NLP tasks. Word embeddings are substantially successful in capturing semantic relations among words, so a meaningful semantic structure must be present in the respective vector spaces. However, in many cases, this semantic structure is broadly and heterogeneously distributed across the embedding dimensions making interpretation of dimensions a big challenge. In this study, we propose a statistical method to uncover the underlying latent semantic structure in the dense word embeddings. To perform our analysis, we introduce a new dataset (SEMCAT) that contains more than 6500 words semantically grouped under 110 categories. We further propose a method to quantify the interpretability of the word embeddings. The proposed method is a practical alternative to the classical word intrusion test that requires human intervention.

Subfields: Artificial Intelligence, Natural Language Processing, Representation & Self-Supervised Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

In Representation & Self-Supervised Learning

Citations in the corpus

Analysis Methods in Neural Language Processing: A Survey (2018, methodology): “Since the evaluation is costly for high-dimensional representations, alternative automatic metrics were considered (Park et al., 2017; Senel et al., 2018).”
Evaluating word embedding models: methods and experimental results (2019, background): “[37], humans can group words by inference using concepts that word embeddings can gloss over.”
What's ur Type? Contextualized Classification of User Types in Marijuana-Related Communications Using Compositional Multiview Embedding (2018, background): “[11], [12] provides relevant intuitions for our work.”
Reconstructing Capsule Networks for Zero-shot Intent Classification (2019, background): “Since a word can have different meanings in different contexts, some interesting recent studies (Shen et al., 2018; S¸enel et al., 2018) suggest that different dimensions of a word embedding may tend to represent different semantic meanings.”
DirectProbe: Studying Representations without Classifiers (2021, background): “Controlled test sets (¸Senel et al., 2018; Jastrzebski et al., 2017) are designed and errors are analyzed to reverse-engineer what information can be encoded by the model (e.g., Marvin and Linzen, 2018; Ravichander et al., 2020; Wu et al., 2020).”
Quantifying Positional Biases in Text Embedding Models (2024, unknown): “Further research has shown that vector operations, such as adding embeddings, can produce new vectors that represent the semantic meaning of their components [22].”