Distributional social semantics: Inferring word meanings from communication patterns

Area of Science:

• Computational Linguistics
• Psycholinguistics
• Sociolinguistics

Background:

• Distributional models of lexical semantics learn word meanings from co-occurrence statistics in large text corpora.
• These standard models overlook the crucial social and communicative aspects of language processing central to usage-based theories.
• Previous research indicates social information is vital for language learning and processing.

Purpose of the Study:

• To investigate how social information embedded in communication patterns contributes to acquiring unique aspects of word meaning.
• To explore a new pathway for developing distributional models by integrating social and communicative data.
• To demonstrate the utility of analyzing large-scale online communication for linguistic insights.

Main Methods:

• Analysis of communication patterns from over 330,000 users on the online forum Reddit.
• Processing of approximately 55 billion words of text to extract social and word usage statistics.
• Development and testing of enhanced distributional models incorporating social information.

Main Results:

• Social information derived from communication patterns allows for the acquisition of unique semantic properties of words.
• Findings demonstrate that social context provides statistical information valuable for word meaning acquisition.
• The study validates the integration of social data into distributional semantic models.

Conclusions:

• Social and communicative aspects of language are essential for a comprehensive understanding of word meaning acquisition.
• Analyzing large-scale social communication data offers a novel and effective approach for advancing distributional semantics.
• This research opens new avenues for building more adaptive and socially-aware computational models of language.