Statistical Restoration
Posted: March 22nd, 2022, 9:19 am
It has been many years in the making, but the CNTR is now pleased to announce the beta version of the first computer-generated Greek New Testament, which for now I am going to call the “Statistical Restoration” (SR). The creation of the SR text offers a paradigm shift in the field of textual criticism, replacing the subjectivity of human editors with the use of objective statistical and computational methods. All the earliest manuscript evidence is fed directly to a computer program as raw data, which generates a Greek New Testament to reflect the most probable text based on statistical analysis and algorithms designed to mimic the processes of textual criticism, weighing both internal and external evidence. The SR offers a number of significant advantages:
• Weighs the raw manuscript data consistently and objectively without human intervention based solely on scientific principles.
• Can be updated automatically whenever new witnesses are added.
• Automatically comes with morphological parsing, Enhanced Strong’s Numbers, English interlinear, and the Universal Apparatus.
• Will later be released under an open license for use by the global Church, after it moves from the beta stage into production.
A number of breakthroughs had to occur along the way in order to accomplish this feat, including the automatic determination of variant unit boundaries and their relationships to each other, the classification of homophones based on the orthographical-priority method, and rating the statistical reliability of manuscripts against the corpus of data. The scientific underpinnings for this are explained in six different papers which are planned to be submitted for publication, three of which have already been presented at the Society of Biblical Literature conferences. I won’t go into all of the details here, but more information about the SR text can be found in this overview video: https://greekcntr.org/resources/4SRGNT.mp4
Note that this beta version of the SR text is subject to change as it is scrutinized by others for possible flaws. This version is about 1.5% different than the Nestle-Aland 28th edition, but some them are merely orthographical differences based on how I deal with phonology. Depending on the options used, it could be made to be less than 1% different, but the goal here was not to try to replicate any particular critical text, but rather to apply a scientific-based, data-driven approach derived from the raw data. It takes about two seconds for the program to generate one Greek New Testament, and the text was optimized by running batches of 1000 texts at a time with varying parameters.
The beta version is now visible on the CNTR website in the collation view: https://greekcntr.org/collation/index.htm. Try doing a hard refresh (Ctrl-F5) if you don’t see the updates. If you would like to help, I would welcome feedback about any places where you don’t think the computer got it right, based on the dataset that it is working from. Certainly, there are plans for the dataset to be expanded in the future.
• Weighs the raw manuscript data consistently and objectively without human intervention based solely on scientific principles.
• Can be updated automatically whenever new witnesses are added.
• Automatically comes with morphological parsing, Enhanced Strong’s Numbers, English interlinear, and the Universal Apparatus.
• Will later be released under an open license for use by the global Church, after it moves from the beta stage into production.
A number of breakthroughs had to occur along the way in order to accomplish this feat, including the automatic determination of variant unit boundaries and their relationships to each other, the classification of homophones based on the orthographical-priority method, and rating the statistical reliability of manuscripts against the corpus of data. The scientific underpinnings for this are explained in six different papers which are planned to be submitted for publication, three of which have already been presented at the Society of Biblical Literature conferences. I won’t go into all of the details here, but more information about the SR text can be found in this overview video: https://greekcntr.org/resources/4SRGNT.mp4
Note that this beta version of the SR text is subject to change as it is scrutinized by others for possible flaws. This version is about 1.5% different than the Nestle-Aland 28th edition, but some them are merely orthographical differences based on how I deal with phonology. Depending on the options used, it could be made to be less than 1% different, but the goal here was not to try to replicate any particular critical text, but rather to apply a scientific-based, data-driven approach derived from the raw data. It takes about two seconds for the program to generate one Greek New Testament, and the text was optimized by running batches of 1000 texts at a time with varying parameters.
The beta version is now visible on the CNTR website in the collation view: https://greekcntr.org/collation/index.htm. Try doing a hard refresh (Ctrl-F5) if you don’t see the updates. If you would like to help, I would welcome feedback about any places where you don’t think the computer got it right, based on the dataset that it is working from. Certainly, there are plans for the dataset to be expanded in the future.