ANTH 392 Annotated Bibliography_Emily Perkovic

docx

School

University of Victoria *

*We aren’t endorsed by this school

Course

392

Subject

Anthropology

Date

Oct 30, 2023

Type

docx

Pages

Uploaded by emilyperkovic

Emily Perkovic ANTH 392 A01 V00865638 March 16th, 2020 Artificial Intelligence and It’s Role in Deciphering Linear A No language has drawn as much attention as Linear A, the undeciphered Minoan language. Many attempts have been made to decipher the writing but due to low the number of tablets, poor quality of samples, short character sequences and other compounding factors it has never been done (Hooker, 1975). In recent years computers have been used to decipher other ancient languages such as Cuneiform (Mostofi & Khashman, 2014 and Bogacz et al., 2017). Using pattern recognition and intelligent recognition software it may be possible to use deciphered root languages as a key to decipher Linear A. Through recent genetic testing we can determine which civilizations closely interacted with the Minoans. Clues about the Minoans origins have become more readily available through while-genome sampling (Lazaridis et al., 2017). The ability to use genetics to narrow the search for Linear A’s roots, will increase chances of deciphering Liner A. Combing computer learning and the writings of likely genetic ancestors of the Minoans will give us the best possibility of finally deciphering Linear A. Bogacz, B., Klingmann, M. and Mara, H., 2017, November. Automating Transliteration of Cuneiform from Parallel Lines with Sparse Data. In 2017 14th IAPR International Conference on Document Analysis and Recognition (ICDAR) (Vol. 1, pp. 615-620). IEEE. Bogacz, Klingmann and Mara present the first approach for automized learning of transliterations on cuneiform tablets. The team utilized 3D- acquisition and word-spotting based on a corpus of parallel lines. Using tracings of cuneiform tablets provided by Yale University part-of-speech tagging was implemented and they were able to learn the correspondence between distinctive features and transliteration 1

Emily Perkovic ANTH 392 A01 V00865638 March 16th, 2020 tokens. Unfortunate this approach was not able to yield fully automatized transliterations. The authors attribute this failure to the sparsity of the data, inconsistent labeling and variety of tracing styles. This article proves that the application of artificial intelligence to ancient languages is an emerging topic and possibly of high value. Automation will be key moving forward as the number of qualified experts capable of transcribing cuneiform is low. This study also highlights past machine learning attempts that have not worked, and what the potential downfalls of that method were. One problem that was encountered by Bogacz, Klingmann and Mara was the lack of whitespace in cuneiform. The lack of whitespace makes if very hard for a computer to determine where one word ends and another begins, this is where the need for human understanding of the writing’s topic comes into play. Overcoming whitespace detection will make full computer learned deciphering possible. Hooker, J.T., 1975. Problems and Methods in the Decipherment of Linear A. Journal of the Royal Asiatic Society , 107 (2), pp.164-172. In this article J.T. Hooker describes the various hurdles with deciphering Linear A. The greatest obstacle according to Hooker is within the material it’s self. There are a relatively low number of tablets and of those tablets most are fragmented or very well worn, making deciphering even harder. Another problem faced with deciphering them comes with how short and personal most of the inscriptions are. Short strings of characters are harder to decipher as well as text with lots of proper names. He purposes roots for the language in either Greek or Semitic or a language of the Anatolian group, with the caveat however that their influence is only seen in very small fractions. This article outlines some of the problems that may be encountered while working 2

Emily Perkovic ANTH 392 A01 V00865638 March 16th, 2020 on this project. It also provides an idea of languages Linear A may draw from. Lazaridis, I., Mittnik, A., Patterson, N., Mallick, S., Rohland, N., Pfrengle, S., Furtwängler, A., Peltzer, A., Posth, C., Vasilakis, A. and McGeorge, P.J.P., 2017. Genetic origins of the Minoans and Mycenaeans. Nature , 548 (7666), pp.214-218. Lazaridis et al. collected genome-wide data from a total of nineteen individuals from Crete, Greece and southwest Anatolia. They set out to answer four main questions, the two most significant are as follows: Are Minoans and Mycenaeans genetically distinct populations, and how are the two groups related? After conducting principal component analysis (PCA) it was determined that the Minoans and Mycenaeans were homogeneous, sharing over three-quarters of there ancestry from Neolithic farmers. This discovery supports the idea of genetic similarity between the two groups. The two groups do differ is some aspects as Mycenaeans were also found to have DNA linking them to hunter- gatherers of eastern Europe and Siberia. Unlocking the genetic history of the Minoan people could give us a better idea as to the origins of Linear A. This study helps to provide a clue to the civilizations which predated the Minoans and may have influenced their language. Comparing Neolithic markings and proto-writing to Linear A may help shed light on it’s origins. Mostofi, F. and Khashman, A., 2014, August. Intelligent Recognition of Ancient Persian Cuneiform Characters. In IJCCI (NCTA) (pp. 119-123). In this article Mostofi and Khashman created an artificial neural network (ANN) and used it to identify noisy images of Cuneiform 3

Your preview ends here

Eager to read complete document? Join bartleby learn and gain access to the full version

Access to all documents
Unlimited textbook solutions
24/7 expert homework help

Emily Perkovic ANTH 392 A01 V00865638 March 16th, 2020 characters. The ANN was successfully trained after 243 iterations and 13 seconds of image processing. After training was complete it took the network 0.07 seconds to identify a pristine Cuneiform image. The network was then tested with three levels of noise image to determine its accuracy. At the highest level of noise tested images were identified correctly 89.1% of the time. This article demonstrated the ability for an ANN to identify potentially degraded or damaged samples, while still maintaining a high level of accuracy. A large portion of Linear A tablets are degraded and weathered so this form of identification would be beneficial. This study also shows general positive results, which support the idea that an Artificial Neural Network if trained properly can quickly identify and differentiate characters. 4