Diachronic Analysis of the Italian Language exploiting Google Ngram
1 like542 views
The document presents a diachronic analysis of the Italian language using Google Ngram data from over five million books spanning from 1500 to 2012. It describes a method called temporal random indexing (TRI) to detect shifts in word meanings over time, demonstrating its effectiveness in evaluating meaning changes and establishing a benchmark for the Italian language. The authors propose future work to expand the dataset and apply the methodology to the English language.
Diachronic Analysis of the Italian Language exploiting Google Ngram
- 1. Diachronic Analysis of the Italian Language exploiting Google Ngram
- 2. Hello! Pierpaolo Basile Annalina Caputo Roberta Luisi Giovanni Semeraro Department of Computer Science University of Bari Aldo Moro - Italy
- 3. Background TRI P. Basile, A. Caputo, G. Semeraro. Temporal random indexing: A system for analysing word meaning over time. IJCoL vol. 1: Emerging Topics at the First Italian Conference on Computational Linguistics. Corpus with Temporal Information Dictionary/Random Vectors Temporal Random Indexing Word Space Word Space Word Space Word Space Word Space ▪ Several WordSpaces for several time periods ▪ Word vectors are comparable across WordSpaces
- 4. Motivation 1 Detect meaning shift Marty, in 2015 people will surf on the web!!!
- 5. Motivation 1 Detect meaning shift Surf!?!?! On the web!?!?!?
- 6. Motivation 1 Detect meaning shift Surf!?!?! On the web!?!?!? surf the Net/Internet to use the Internet When was this meaning introduced?
- 7. Motivation 2 Large corpus ▪ Build a method for computing TRI relying on a very large corpus ▪ Google Ngram for the Italian language ▫n-grams (up to five) extracted from Google Books ▫over five million books spanning the years from 1500 to 2012 ▪ covers several languages including Italian analysis is often described as 1991 104 5 N-gram occurrences books
- 8. Methodology 1. Run TRI on the Italian Google Ngram ▫build a WordSpace for each time period (10 years) 2. Provide for each word a time series 3. Search significant changes in the time series
- 9. cossim ( , ) Time Series Several time series Γ at the time interval k log frequency point-wise cumulative cossim ( , ) Word frequency in each time period k Cosine similarity between word vectors across two time periods Considers a cumulative vector of the previous k-1 time periods
- 10. Change point detection ▪ Mean shift of Γ pivoted at time period j ▪ Search statistical significant mean shifts ▫bootstrapping approach under the null hypothesis that there is no change in the meaning
- 11. Evaluation Dataset Build a benchmark for meaning shift detection for the Italian language ▪ extract a set of words by pooling data by running several system settings ▪ find correct change points in a dictionary (Sabatino Coletti/Etimologico Zanichelli)
- 12. Evaluation Results Method Accuracy TRIpoint 0.3086 TRIcum 0.2963 TRR1point 0.2716 log freq 0.2346 TRR2point 0.1728 TRR1cum 0.1605 TRR2cum 0.1235 Accuracy: the year predicted by the system should be equal or greater than the year reported in the gold standard TRR1 and TRR2 are variants of TRI based on Reflective Random Indexing
- 13. Conclusions and Future Work ▪ TRI method with point wise detection provides good results ▫it overcomes the baseline based on log-frequency ▪ We provide a benchmark for the evaluation of meaning shifts for the Italian language ▪ Future work: extend the dataset and provide an evaluation for the English language