Word embeddings are a core component of modern natural language processing systems, making the ability to thoroughly evaluate them a vital task. We describe DiaLex, a benchmark for intrinsic evaluation of dialectal Arabic word embeddings. DiaLex covers five important Arabic dialects: Algerian, Egyptian, Lebanese, Syrian, and Tunisian. Across these dialects, DiaLex provides a testbank for six syntactic and semantic relations, namely male to female, singular to dual, singular to plural, antonym, comparative, and genitive to past tense. DiaLex thus consists of a collection of word pairs representing each of the six relations in each of the five dialects. To demonstrate the utility of DiaLex, we use it to evaluate a set of existing and new Arabic word embeddings that we developed.
Given our benchmark, we generate a test bank consisting of 260,827 tuples. Each tuple consists of two word pairs (a, b) and (c, d) from the same relation. For each of our six relations, we generate a tuple by combining two different word pairs from the same relation. Once tuples have been generated, they can be used as word analogy questions to evaluate different word embeddings as defined by Mikolov et al. (Mikolov et al., 2013a). A word analogy question for a tuple consisting of two word pairs (a, b) and (c, d) can be formulated as follows: ”a to b is like c to ?”. Each such question will then be answered by calculating a target vector t = b − a + c. We then calculate the cosine similarity between the target vector t and the vector representation of each word w in a given word embeddings V. Finally, we retrieve the most similar word w to t If w = d (i.e., the same word) then we assume that the word embeddings V has answered the question correctly.
Moreover, we also extend the traditional word analogy task by taking into consideration if the correct answer is among the top K, with K in {5, 10}, closest words in the embedding space to the target vector t, which allows us to more leniently evaluate the embeddings. This is particularly important in the case of Arabic since many forms of the same word exist, usually with additional prefixes or suffixes such as the equivalent of the article "the" or possessive determiners such as "her", "his", or "their". For example, consider one question which asks راجل to ست is like أمير to "?", i.e., "man to woman is like prince to ?", with the answer being "أميره" or "princess". Now, if we rely only on the top-1 word and it happens to be "للأميرة" which means "for the princess" in English, the question would be considered to be answered wrongly. To relax this, and ensure that different forms of the same word will not result in a mismatch, we use the top-5 and top-10 words for evaluation rather than the top-1.
After running jupyer notebook and opening Evaluating_Word_Embeddings.ipynb. The user would need to define the following constants:
- MODEL_PATH - Path to the pre-trained model
- BENCHMARKS_PATH – Path to folder containing the benchmark files, where lines are 4-tuples of words, split into sections by “: SECTION NAME” lines.
- TOP_K - The final aggregated report includes a top-1, top-5, and top-10 accuracy result for each section, in which the b word is searched for in the model’s top-k most similar words
- DUMMY4UKNOWN (bool, optional) – If True - produce zero accuracies for 4-tuples with out-of-vocabulary words. Otherwise, these tuples are skipped entirely and not used in the evaluation.
Output
- An xls workbook containing the accuracy spreasheet for each benchmark file included.
- A logger file containing the gensim output of the evaluate_word_analogies() used.
Please use the following citation if you use Dialex:
@inproceedings{abdul-mageed-etal-2021-dialex,
title = "{D}ia{L}ex: A Benchmark for Evaluating Multidialectal {A}rabic Word Embeddings",
author = "Abdul-Mageed, Muhammad and
Elbassuoni, Shady and
Doughman, Jad and
Elmadany, AbdelRahim and
Nagoudi, El Moatez Billah and
Zoughby, Yorgo and
Shaher, Ahmad and
Gaba, Iskander and
Helal, Ahmed and
El-Razzaz, Mohammed",
booktitle = "Proceedings of the Sixth Arabic Natural Language Processing Workshop",
month = apr,
year = "2021",
address = "Kyiv, Ukraine (Virtual)",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/2021.wanlp-1.2",
pages = "11--20",
abstract = "Word embeddings are a core component of modern natural language processing systems, making the ability to thoroughly evaluate them a vital task. We describe DiaLex, a benchmark for intrinsic evaluation of dialectal Arabic word embeddings. DiaLex covers five important Arabic dialects: Algerian, Egyptian, Lebanese, Syrian, and Tunisian. Across these dialects, DiaLex provides a testbank for six syntactic and semantic relations, namely male to female, singular to dual, singular to plural, antonym, comparative, and genitive to past tense. DiaLex thus consists of a collection of word pairs representing each of the six relations in each of the five dialects. To demonstrate the utility of DiaLex, we use it to evaluate a set of existing and new Arabic word embeddings that we developed. Beyond evaluation of word embeddings, DiaLex supports efforts to integrate dialects into the Arabic language curriculum. It can be easily translated into Modern Standard Arabic and English, which can be useful for evaluating word translation. Our benchmark, evaluation code, and new word embedding models will be publicly available.",
}