This project was made for Google's 2018 Summer of Code on behalf of CLTK.
Changes found for the CLTK from this project may be found here: \ https://github.com/cltk/cltk/commit/52e28ecec67aad56b21d29c08b32c05683add826
Andrew Deloucas \ email: adeloucas@g.harvard.com
Willis Monroe \ Github: https://github.com/willismonroe \ Tyler Kirby \ Github: https://github.com/TylerKirby
GSoC Profile: https://summerofcode.withgoogle.com/organizations/4928035770335232/
We develop the Classical Language Toolkit (CLTK) because we believe it is revolutionizing the study of the ancient world. It is doing so by removing barriers to entry for those doing natural language processing (NLP) in Classical languages (namely, the surviving literature of the entirety of Eurasia and north Africa, from roughly 3000 B.C. to A.D. 1500).
Due to how academic disciplines have evolved over the past 200 years, our earliest civilizations are often studied in isolation from one another. This is tragic, for today we know that the ancient world – from Rome to Mesopotamia to India to China – consisted of deeply interconnected networks of ideas, technologies, art, and beliefs. As a framework for multidisciplinary research, the CLTK will help scholars discover the commonalities of what were once thought disparate cultures.
As software, the CLTK is a suite of NLP tools suited to the special needs of ancient languages. We have have three goals: The most basic is to offer low-level libraries for doing NLP in particular Classical languages (e.g., Ancient Greek, Sanskrit). Developed with an extensible architecture, our code is easily hacked to support new languages. Second, the CLTK offers tools for students and scholars to do reproducible scientific research. For instance, it has version-controlled linguistic corpora and a suite of functions for stylometrics. Third, it is a framework for multidisciplinary language research. With pre-trained models (such as Word2Vec for vector space models), we provide easy-to-use tools to capture the transmission and evolution of knowledge, from the earliest human societies to the dawn of the modern era.
GSoC Profile: https://summerofcode.withgoogle.com/organizations/5332375953735680/
The mission of the Cuneiform Digital Library Initiative (CDLI) is to collect, preserve and make available images, text and metadata of all artifacts inscribed with the cuneiform script. It is the sole project with this mission and we estimate that our 334,000 catalogue entries cover some two-thirds of all sources in collections around the world. Our data are available publicly at https://cdli.ucla.edu and our audiences comprise primarily scholars and students, but with growing numbers of informal learners.
At the heart of cdli is a group of developers, language scientists, machine learning engineers, and cuneiform specialists who develop software infrastructure to process and analyze curated data. To this effect, we are actively developing two projects: Framework Update https://cdli.ucla.edu/?q=news/cdli-core-update and Machine Translation and Automated Analysis of Cuneiform Languages https://cdli-gh.github.io/mtaac/. As part of these projects we are building a natural language processing platform to empower specialists of ancient languages for undertaking automated annotation and translation of Sumerian language texts thus enabling data driven study of languages, culture, history, economy and politics of the ancient Near Eastern civilizations. As part of this platform we are focusing on data standardization using Linked Open Data to foster best practices in data exchange and integration with other digital humanities and computational philology projects.
GSoC Profile: https://summerofcode.withgoogle.com/projects/#5184805973524480
This project focuses on integrating Cuneiform Digital Library Initiative (CDLI) corpora into the Classical Language Toolkit (CLTK). Currently, CLTK houses several functions developed by Dr. Willis Monroe; the difficulty in utilizing these functions is due to the variables having to be presented in a reconstructed normalized form of Akkadian, which is not how tablets are either written traditionally or stored by CDLI. The goal of this project is to enable CLTK to reconstruct a normalized spelling out of CDLI’s ATF text and thus create data available for two fundamental uses: 1) allowance of individuals to learn and practice Akkadian with real and novel reading exercises; and 2) with further class development, be analyzed on a mass scale.
At its core, this project highlights the discrepancy between current CTLK functions and their (in)ability to utilize the world’s largest archive of Akkadian data: CDLI. CDLI uses ATF (ASCII Transliteration Format), which is currently unreadable by CLTK.
confronted before CLTK’s Akkadian corpus can be fleshed out: 1) CLTK must be able to read a variety of cuneiform writing styles in ATF by means of an ASCIIUnicode converter, 2) for writing that does not provide full inflection, CLTK must know how to read writing conventions via declination and lemmatization, and 3) the exposition must become recognizable for additional classes to work. Thus, this project is established in three parts: \ 1) ASCII-Unicode conversion, 2) Normalization, and 3) Implementation, or case study.
The development of an ASCII-Unicode converter shall be first established for functions in which CLTK is familiar: the goal of this converter is to establish a way to encode unrecognizable data into Unicode. This process will take the established methods of typography for CDLI in ASCII and replace the characters with Unicode. This will allow current and proposed functions to be used. The end result of this converter will additionally provide a ‘sign transliteration’.
Because ATF-inflected data is uncommon to find, part 2 of this project involves creating a class that, in processing typical ATF data, references and abides by grammatical rules via lemmatization, declension, stress, and morphology: that is to say, normalization. By utilizing Huehnergard’s grammar, we will be able to create a class that follows set orthographic rules. Dr. Willis Monroe has already developed a decliner class, which establishes how Akkadian nouns, based upon their gender, are declined by case and number, as well as a tool to recognize radicals in a verb. Beginning with a set corpus of nouns, I will perform declinations of (ir)regular nouns and their forms in order to catch any inconsistencies; when corrections are made, its output will create a “lookup table” that will aid the process of creating a function where any noun form would return their lemma. This process will be repeated, albeit with verbs. From such functions, a final table will take form as a “lemmatizer.” Upon finalizing the lemmatizer, the goal is to finalize the class such that, upon exposition of the converter above, CLTK can normalize the provided ATF corpus and be able to operate any additional functions.
The case study of this project is composed of two exercises: 1) a long-form analysis of a singular text and 2) a sample of a larger corpus. This section of the project should be able to utilize parts 1 and 2 in order to become fully capable of being analyzed like any other NLP.
Due to the scope of this project, this proposal will focus on the Old Babylonian dialect of Akkadian, which has rules well established in Dr. John Huehnergard’s “A Grammar of Akkadian.” The texts being used will be the Code of Hammurapi (henceforth Codex Hammurabi) and Georges Dossin’s 'Archives Royale de Mari 1' (ARM 01).
I will be dedicating 20 hours per week to this project. Below is a table breaking down work and goals, hours dedicated, and their relation to the Google Summer of Code schedule. Evaluations established by GSoC will be completed outside of the allotted time mentioned.
(4 weeks) \ • Learn string manipulation \ • Learn Dictionary / List Comprehension \ • Familiarize myself with collections library functions \ • Total hours = 80 hours
(3 weeks) \ • Develop and Complete Part 1, Converter \ • Total hours = 60 hours
(12 weeks) \ • Develop and Complete Parts 2 and 3, Normalization and Implementation \ • Total hours = 240 hours
(2 weeks) \ • Complete Evaluations and Write Article \ • Set for Publication \ • Total hours = 40 hours