It is not always clear how exactly to translate or interpret ancient Hebrew Bible texts. There are therefore countless translations and commentaries of every Bible passage. According to theologian Christiaan Erwich, the computer can help to arrive at an objective analysis.
A pastor who wants to interpret a Bible passage in his sermon has no simple task. You can easily find nine or ten comments about one piece of text and they all say something different. Take the following sentence from Psalm 75:“I judge you.” One says:Here speaks a king, the other:a prophet. And that goes on endlessly. “Especially the 'who's who' question gives many translators headaches,” says Christiaan Erwich.
Erwich is a theologian specializing in the Old Testament, the oldest part of the Bible that has been handed down in Hebrew. Because the texts are so old – this part dates from before the beginning of the Christian era – they are not always easy to fathom. Moreover, at least a third of the Old Testament consists of poetry, which makes it even more difficult. It explains why there are so many different interpretations of it.
They are all subjective interpretations based on individual assumptions. But according to Erwich, you can also arrive at an objective interpretation by just looking at the text. That is why, as a PhD student, he developed a computer model that makes statements based purely on the text. He renamed the computer model MiMi, Hebrew for 'who who'. With the help of MiMi he was able to find out that the 'I' from the aforementioned sentence from Psalm 75 refers to:God.
Featured by the editors
MedicineWhat are the microplastics doing in my sunscreen?!
AstronomySun, sea and science
BiologyExpedition to melting land
Ambiguous language
To develop MiMi, the theologian specialized in computational linguistics methods. An important question within that field is how to make computers intelligent enough to understand and produce human language. The computer is not yet very good at dealing with ambiguous reference words.
Take the following sentence:"The councilors refused to permit the protesters because they were afraid of violence." This is an example sentence that computer scientists use to test the intelligence of the computer. Although a person can reason well in this sentence to whom the personal pronoun 'she' refers, this is still a bridge too far for the computer. Because there are so many ambiguous reference words in old Bible texts, you can use these texts excellently to test a computer program, a so-called algorithm.
Before you have an algorithm that interprets reference words in a text, you need to train the computer. Erwich did this by teaching the system linguistic rules, such as:two pronouns that are close and grammatically similar refer to each other. Based on that rule, the computer can see that in the sentence "He addresses his people," "he" and "his" are related. They are grammatically similar because they both have the grammatical features 'third person', 'singular' and 'masculine'.
Digging through 150 psalms
Based on those rules, Erwich first hand-labeled 150 psalms. For each word that referred to a person, he indicated which other word it was associated with. Scientists call the extra information you add to a text 'annotations'. To investigate whether the rules he had drawn up were unambiguous, he also had a colleague label a number of those psalms. The computer then compared their annotations and found that they matched 82 percent.
Then it was the turn of MiMi, the computer algorithm. The agreement between the annotations was now 40 percent, a lot lower. Still, that's not just a bad score, the researcher explains. Because such an analysis of persons in a text actually consists of two parts. Firstly, detecting words that refer to persons, such as pronouns. The computer turned out to be super good at that, even better than the researcher. But relating those words to each other was more complicated for MiMi.
“Where MiMi only gets information from the immediate surrounding text itself, I can get extra information from the entire Psalm corpus, from the entire Hebrew Bible and I have experience as a human being in this world. I use all that knowledge to interpret reference words. A computer does not have all that extra knowledge, so it is not surprising that it scores lower. In that respect, my research has also been an experiment in exploring what kind of tasks the computer can take over from me when it comes to interpreting 'fuzzy data'.”
App for pastors
The algorithm still has to be developed further, but after that Erwich expects that the computer will bring about a major change in the field. Now the comments are often not reasoned from within the text, but from knowledge outside it. “Then people know, for example, what function a psalm has had within an Israeli temple, and on that basis it is assumed that the I-figure must be a king or priest. This algorithm only looks at the text.”
That does not mean, however, that people themselves are no longer involved. “You can use those comments as hypotheses that you can then disprove or confirm with such an algorithm. You can then discard some of those theories.”
Erwich is already thinking of an application in an app for pastors. “A well-trained pastor may put two commentaries side by side to explain a Bible passage in his sermon. But in assessing those comments, a computer algorithm like MiMi can help him on his way. It is a tool for him to interpret the text even better.”