Complexity of Word Collocation Networks: A Preliminary Structural Analysis

By Shibamouli Lahiri. 2013

91 citations recorded by the source corpus.

In this paper, we explore complex network properties of word collocation networks (Ferret, 2002) from four different genres. Each document of a particular genre was converted into a network of words with word collocations as edges. We analyzed graphically and statistically how the global properties of these networks varied across different genres, and among different network types within the same genre. Our results indicate that the distributions of network properties are visually similar but statistically apart across different genres, and interesting variations emerge when we consider different network types within a single genre. We further investigate how the global properties change as we add more and more collocation edges to the graph of one particular genre, and observe that except for the number of vertices and the size of the largest connected component, network properties change in phases, via jumps and drops.

Subfields: Artificial Intelligence, Natural Language Processing, Evaluation & Benchmarks

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 Evaluation & Benchmarks

Citations in the corpus

Unbounded cache model for online language modeling with open vocabulary (2017, background): “• The book Corpus This is a dataset of 3,036 English books, collected from the Project Gutenberg 4 [40].”
Machine Learning Models that Remember Too Much (2017, methodology): “We now relax this assumption and assume that the attacker uses an auxiliary vocabulary collected from publicly available corpuses: Brown Corpus,2 Gutenberg Corpus [43],3 Rotten Tomatoes [62],4 and a word list from Tesseract OCR.5 Obviously, this public auxiliary vocabulary requires no prior knowledg”
Statistical Model Aggregation via Parameter Matching (2019, unknown): “We quantitatively compare the two approaches by calculating the UCI coherence [28] scores over the Gutenberg dataset [26] consisting of 3000 books separate from the ones used to train the two models.”
A Neural Grammatical Error Correction System Built On Better Pre-training and Sequential Transfer Learning (2019, methodology): “For pre-training, we use the Gutenberg dataset (Lahiri, 2014), the Tatoeba4 dataset, and the WikiText-103 dataset (Merity et al.”
Gender Stereotypes Differ between Male and Female Writings (2019, methodology): “In the first experiment, we use a dataset by Lahiri (2013), which consists of 3,036 English books written by 142 authors.”
Temporal Reasoning on Implicit Events from Distant Supervision (2020, methodology): “We use the Gutenberg Dataset (Lahiri, 2014) as the source and collect 1 million paragraphs for this purpose.”
A Comparative Study of Synthetic Data Generation Methods for Grammatical Error Correction (2020, methodology): “…GEC Earlier GEC approaches focused on English as a Second Language Learners and made use of linear machine learning algorithms and developed classiﬁers for speciﬁc error types, such as articles, prepositions, or noun number (Gamon, 2010; Rozovskaya and Roth, 2010, 2014; Dahlmeier and Ng, 2012).”
SLM: Learning a Discourse Language Representation with Sentence Unshuffling (2020, methodology): “Sentences with typical discourse labels from samples in Jernite et al. (2017) are used as queries to see whether our model cap- tures a variety of different discourses and 1M sentence pairs from Gutenberg BookCorpus (Lahiri, 2014) are used as targets to search.”