LAGTvecs

This dataset represents word vectors for eight subcorpora based on the LAGT (Zenodo) corpus of lemmatized ancient Greek texts. Each word is represented by a 100-dimensional vector, which has been obtained by means of the CBOW word2vec. The exact parameters of the training procedure and other details are described in our paper "A Distributional Semantic Approach to the Religious and Moral Dynamics in the Ancient Greek Texts", currently under review. The scripts used for generation of these vectors are available for reuse as a supplementery material to this paper, either from Zenodo  are directly from Github.

We make the data available as a single pickle file, which allows to load the vector data directly into Python by the following piece of code:

import requests
import pickle
import io

url = "https://zenodo.org/records/10685268/files/vectors_sample1m_dict.pickle?download=1"
vectors_dict = pickle.load(io.BytesIO(requests.get(url).content))

It returns a dictionary object, with names of individual subcorpora as keys (see the table below) and Gensim's keyed vectors for each subcorpus as values. For instance, to get ten most similar words to the term θεός in Greek texts from the Classical Period, you run:

vectors_dict["pagan_classical"].most_similar("θεός")

The eight subcorpora are as follows:

subcorpus

works_n

tokens_n

lemmata_n

pagan_archaic

59

338991

199023

pagan_classical

454

4129342

2042961

pagan_hellenistic

148

2533938

1303655

pagan_roman_peak

602

10903481

4820083

pagan_roman_late

229

5446413

1970470

christian_roman_peak

113

2082493

869103

christian_roman_late

96

3833534

1544889

jewish

93

2121316

814648