Rotated Word Vector Representations and their Interpretability

Park, Sungjoon; Bak, JinYeong; Oh, Alice

doi:10.18653/v1/d17-1041

Cited by 42 publications

(49 citation statements)

References 36 publications

(30 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…16 In this kind of work, a word embedding model may be deemed more interpretable if humans are better able to identify the intruding words. Since the evaluation is costly for high-dimensional representations, alternative automatic metrics were considered (Park et al, 2017;Senel et al, 2018).…”

Section: Other Methodsmentioning

confidence: 99%

Analysis Methods in Neural Language Processing: A Survey

Belinkov

Glass

2019

Transactions of the Association for Computational Linguistics

387

297

View full text Add to dashboard Cite

The field of natural language processing has seen impressive progress in recent years, with neural network models replacing many of the traditional systems. A plethora of new models have been proposed, many of which are thought to be opaque compared to their feature-rich counterparts. This has led researchers to analyze, interpret, and evaluate neural networks in novel and more finegrained ways. In this survey paper, we review analysis methods in neural language processing, categorize them according to prominent research trends, highlight existing limitations, and point to potential directions for future work.

show abstract

Section: Other Methodsmentioning

confidence: 99%

Analysis Methods in Neural Language Processing: A Survey

Belinkov

Glass

2019

Transactions of the Association for Computational Linguistics

387

297

View full text Add to dashboard Cite

show abstract

“…Note, however, that it was shown in [27] that total interpretability of an embedding is constant under any orthogonal transformation and it can only be redistributed across the dimensions. With a similar motivation to [27], [28] proposed rotation algorithms based on exploratory factor analysis (EFA) to preserve the expressive performance of the original word embeddings while improving their interpretability. In [28], interpretability was calculated using a distance ratio (DR) metric that is effectively proportional to the metric used in [27].…”

Section: Related Workmentioning

confidence: 99%

“…With a similar motivation to [27], [28] proposed rotation algorithms based on exploratory factor analysis (EFA) to preserve the expressive performance of the original word embeddings while improving their interpretability. In [28], interpretability was calculated using a distance ratio (DR) metric that is effectively proportional to the metric used in [27]. Although interpretability evaluations used in [27] and [28] are free of human effort, they do not necessarily reflect human interpretations since they are directly calculated from the embeddings.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Semantic Structure and Interpretability of Word Embeddings

Senel

Utlu

Yücesoy

et al. 2018

IEEE/ACM Trans. Audio Speech Lang. Process.

View full text Add to dashboard Cite

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 (SEM-CAT) 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.

show abstract

“…Though powerful and fairly easy to implement with specialized packages (e.g., the Gensim library; Rehurek & Sojka, 2010), these new methods still suffer in part from a crucial drawback shared with LSA, in that the embeddings used to assess semantic similarity are highdimensional mathematical spaces whose intrinsic meaning can be challenging to apprehend (Smalheiser & Bonifield, 2018). Though there has been research into techniques that attempt to address this issue (e.g., Luo, Liu, Luan, & Sun, 2015;Park, Bak, & Oh, 2017), generally these approaches make both the interpretation of the dimensions of the semantic space and understanding of the influence of specific keywords difficult. Further, though some of the simplicity of using word2vec comes from using pretrained embeddings, these spaces may not be optimal for particular applications, and training new embeddings can present several challenges (Smalheiser & Bonifield, 2018).…”

Section: Quantifying Semantic Contentmentioning

confidence: 99%

Beyond frequency counts: Novel conceptual recurrence analysis metrics to index semantic coordination in team communications

Tolston¹,

Riley

Mancuso

et al. 2018

Behav Res

View full text Add to dashboard Cite

Semantic alignment is a key process underlying interpersonal and team communication. However, semantic similarity is difficult to quantify, and statistical approaches designed to measure it often rely on methods that make the identification of the relative importance of key words difficult. This study outlines how conceptual recurrence analysis (CRA) can address these issues and can be used to detect conceptual structure in interpersonal communication. We developed several novel CRA metrics to analyze communication data reported previously by Mancuso, Finomore, Rahill, Blair, and Funke (Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 58, 405-409, 2014), gathered from teams who worked cooperatively on a logic puzzle under different cognitive biasing contexts. CRA, like other measures of semantic coordination, relies on parameters whose values affect estimates of semantic alignment. We evaluated how the dimensionality of semantic spaces affects metrics quantifying the conceptual similarity of communicative exchanges, and whether metrics calculated from top-down, a priori semantic spaces or bottom-up semantic spaces empirically derived from each data set were more sensitive to biasing context. We found that the novel CRA measures were sensitive to manipulations of cognitive bias, and that higher-dimensional, bottom-up semantic spaces generally yielded more sensitivity to the experimental manipulations, though when the communication was evaluated with respect to specific key concepts, lower-dimensional, top-down spaces performed nearly as well. We conclude that CRA is sensitive to experimental manipulations in ways consistent with prior findings and that it presents a customizable framework for testing predictions about interpersonal communication patterns and other linguistic exchanges.

show abstract

Rotated Word Vector Representations and their Interpretability

Cited by 42 publications

References 36 publications

Analysis Methods in Neural Language Processing: A Survey

Analysis Methods in Neural Language Processing: A Survey

Semantic Structure and Interpretability of Word Embeddings

Beyond frequency counts: Novel conceptual recurrence analysis metrics to index semantic coordination in team communications

Contact Info

Product

Resources

About