Rashi was right: Machine learning confirms unique status of some Talmudic tracts
New study shows that the ‘special tractates’ of the Babylonian Talmud have distinct linguistic features, as commented on by medieval sages
Gavriel Fiske is a reporter at The Times of Israel
Rabbinical commentators on the Talmud noted in the medieval era that a handful of sections of the great corpus stood out linguistically from the rest. Over generations of scholars, the existence of these so-called “special tractates” was considered to be a clue that could further elucidate how the Talmud was compiled and edited.
Now via modern data analysis, a team of contemporary researchers has shown that these “special tractates” do indeed display a distinct use of language. After feeding nearly the entire Talmudic corpus into machine learning algorithms to parse the Aramaic, they confirmed the theories of Rashi and other medieval scholars.
“We have provided the first comprehensive statistical proof of what humans intuitively have been aware of for centuries,” said Dr. Jakub Zbrzeżny, one of the authors of “A computational analysis of the special Talmudic tractates,” published in April in The Journal of Jewish Studies.
This kind of research is “an exploration into the great linguistic diversity of the Jewish world. It brings that complexity into the light. It’s not as simple as we may think,” he told The Times of Israel.
Aramaic, the ancient Semitic language that was the lingua franca of the Levant during the time of the Second Temple, survives as a spoken language among few neo-Aramaic speakers, as a written language in the liturgy of the Syriac
Church, and in the Jewish world largely due to its use in the Talmud. It is far more diverse than most people realize, Zbrzeżny said.
“Aramaic is a complex collection of various dialects. The Aramaic that Jesus spoke wasn’t like the Aramaic that the Jews in Iraq would speak. The Aramaic of biblical times is not like Aramaic of the Talmud, and that the Aramaic of the Babylonian Talmud is not like the Aramaic of the Jerusalem Talmud,” he said, referring to the two Talmudic corpora, one compiled in Iraq and the other in the Land of Israel.
Both Talmuds are based on the Mishnah, the Jewish oral law, which was written down in Hebrew and compiled around 200 CE by Rabbi Judah HaNasi. Sections of the Mishnah were later debated and expounded upon in the great rabbinical academies, and eventually the two Talmuds were compiled in Aramaic from these discussions, the Babylonian around 500 CE and the Jerusalem perhaps a hundred years before that.
Having “the machine” compare the language of the Babylonian Talmud and the Jerusalem Talmud was the key to confirming the rabbis’ labeling of the special tractates in the Babylonian, Zbrzeżny said.
First, the researchers entered a large amount of text from the two Talmuds into the database, enabling the machine-learning algorithm to learn both versions of Aramaic, he explained.
Then, they gave the computer further unlabeled Aramaic sections from both Talmuds and the algorithm was able to accurately determine the origin of the texts, confirming that the program understood the linguistic differences between them. Eventually, the entire Aramaic corpus of both Talmuds was entered.
“However, the machine assigned a number of lines to the Jerusalem Talmud that we knew were from the Babylonian,” said Zbrzeżny. In this way, “the machine is telling you, that this line from what we know was the Babylonian Talmud has, for example, a 90% probability of belonging to the Jerusalem.”
“The number of mislabeled lines was very high for [the sections] that scholars intuitively recognized as the ‘special tractates,’” he continued.
“Those who compiled the Talmud were aware of the linguistic diversity within the Talmud itself, and were aware when certain passages stood out. The medieval rabbis in Europe also noticed that certain tractates stood out linguistically,” he said.
These rabbis did this “simply by scholarly intuition, by reading those texts and spotting apparently archaic passages or conservative spelling,” Zbrzeżny said.
Paging Alexander the Great
One of the “special tractates,” Tractate Tamid, which is concerned with the daily sacrifices in the Temple, was found to have a large number of lines flagged by the algorithm, but team member Noam Eisenstein, an MA student at Tel Aviv
University, observed that these lines all dealt with stories about Alexander the Great. If those lines were removed from the equation the tractate would just have normal Jewish Babylonian Aramaic, a possible indicator that those sections were compiled and added from a separate source.
The scholars also found a new phenomenon: some tractates display a “conspicuous dialectical uniformity… an especially uniform language in some tractates may point either to its specific redaction or to the scribal work of the copyist of the manuscript,” the authors wrote.
However, Zbrzeżny said that the research results, while promising, aren’t groundbreaking yet.
“Machine learning doesn’t replace a human; it facilitates our human research. It doesn’t answer the question of when the Talmud was compiled or why we see such linguistic diversity,” he said.
Academic arguments about “why certain passages stand out, do they represent spoken Aramic or are they archaic… we don’t have the answer yet. Further research along these lines can help the academic debates, yes, and the machine can likely spot further linguistic features that scholars haven’t yet identified.”
An adventurous scholar
Zbrzeżny, a Polish-born lecturer in Second Temple Judaism in the Divinity Department at the University of Aberdeen, participated in the Talmud study during three years of postdoc research in Israel. The team was led by Prof. Lee-Ad Gottlieb from Ariel University and Dr. Eshbal Ratzon from Tel Aviv University, and funded by the Cogito Foundation in Switzerland.
Reached via Zoom while on a working visit to the United Arab Emirates, Zbrzeżny recalled that during his post-doctoral research with Israeli scholars, he also immersed himself in local spoken Arabic, a language whose “dialect diversity gives good insights into how Aramaic might have worked in the past, as neo-Aramaic dialects spoken today provide similar analogies.”
This immersion enabled him to pursue a new project focusing on “deepening our understanding of the Tanach [Hebrew Bible] and the New Testament” through the language, dialect and customs of isolated Arabic villages.
“Many religious Muslims in the West Bank love the Bible, they have copies of the Bible,” but can’t stand the modern Arabic versions, which they say is in bureaucratic language or sounds like a newspaper, Zbrzeżny explained.
By engaging in dialogue with the locals and listening to how they “translate” or “interpret” the Bible in “their own rural dialect,” it’s possible to get “fantastic insights… we have reasons to believe that these ‘non-experts’ may provide readings of these texts that Western scholars are not able to discover on their own.”
This process, called “contextualized reading,” is “an experimental approach, but it has the potential to be highly promising if a proper methodology is developed,” Zbrzeżny said. “There are passages in the Bible whose meaning remains uncertain, but [these locals] have their own understanding, and we see if they provide insights into the original meaning that is lost. The world of Semitic languages such as Aramaic and Arabic still await new discoveries.”
