Linguistics

Quantative and Qualitative Analysis of the Russian Accent

Fall 2021-Spring 2022

To keep myself engaged in linguistics through a difficult CS-filled senior year, I elected to take the Linguistics Thesis course, advised by Prof. McWhorter and Prof. Landman. A focus of mine that you can see as a constant throughout the past couple years of my studying has been on Russian accents in English - something that is of course near and dear to my heart being a daughter of Russian immigrants. I have previously focused on the perception of the Russian accent in a very small study which I found had interesting results that followed predictions I had. This time I wanted to expand the study and focus more on a quantitative analysis rather than hand-counting occurrences outright.

My initial attempt took me the entire Fall 2021 semester, where I tried to apply my work in my speech recognition class to this project. Needless to say, this proved to be far more difficult than I had anticipated because that model really was not fit enough to be used on imperfect English speech. The model at its core was trained on ARPAbet, a specifically-English phonetic alphabet that does not represent well non-English sounds, and there were many, many alignment issues that proved to be more difficult to fix than it was worth.

So, I scrapped everything and started anew, this time primarily using Praat and annotating everything by hand from the get-go. I would annotate the recordings with the phonemes that I would expect to be saying in a Standard American English accent and then I used the deviations in frequency in order to come to multiple interesting conclusions regarding potential frequency deviations. This setback of doing things by hand of course left me with much less data than I would have liked (if I had known I'd be doing this, I would have spent all of the first semester gathering my data) but I still think the conclusions I came to are sound! Ha. Sound. Hehe.

The things I was testing for involved phonetic deviations that I was expecting to see in a Russian accent. I focused my efforts on vowel alternations and certain consonantal alternations (mostly focusing on liquids for anything involving frequency, and then I really just had to hand-count stuff like r-trilling.) The results were most promising for instances of vowel-raising, particularly with the alternations between [æ]/[ɛ] and [ʊ]/[u]. The diagrams for both are above, sorted by the perceived accent strength that I have used before. And look! With perceived accent strength, the frequencies seem to be exponentially declining! Amazing!

With the consonants unfortunately there were not many occurrences to say whether it is a linear or otherwise increase in occurrences of deviations, but my conclusion was that liquid velarization and r-trilling were the best indicators of the presence and strength of a Russian accent. I had also attempted to look at high-front vowels as I have previously and see if I can find some kind of correlations there, but did not have any luck. There are many things to look at in the future, especially getting more data, but I hope that these findings can maybe be used to improve speech recognizers in the future!

This project was such a pleasure to work on, and I am so lucky to have been part of the Linguistics community at Columbia. A huge thank-you to Meredith Landman, Prof. McWhorter, and the entire thesis class for being amazing peers and wonderful friends.

You can find the final paper here

Hyphenated Minds: Russian Accent Perception Study

Spring 2021

One of my favorite parts of the Spring semester was working on my Linguistics project. Of course, this isn't to say that I did not enjoy the three other projects I was working on, but my project for the Hyphenated Minds class (taught by visiting professor Danko Sipka of ASU) was a one-person job; other than initial consultation with Danko to discuss the direction that I would like this project to take, I came up with this idea entirely by myself. It also ended up with quite interesting results, for a project whose experiment and analysis stage was completed in under two weeks!

This study focuses on the perception of the strength and intelligibility of Russian accents, specifically of people from St. Petersburg who moved to the United States. There were 7 recordings which 27 people judged on a Google Form.

Each recording was analyzed beforehand in Praat to determine the number of differences in pronunciation from Standard American English. After those scores are collected they are combined into one final score that was then used when doing the data analysis.

The results were really quite interesting! There was a clear positive, almost-linear correlation between the number of correct (or almost-correct, I also took Eastern-European/Slavic because people guessed that as the same rate as Russian) guesses and the number of phonetic alternations and a strong positive correlation between the perceived strength of the accent and the number of phonetic alternations. It was also nice to see that the more strong the accent was, the more certain people were of their answers; respondents barely voted 1 or 2 at perceived strengths of 4 or 5. The intelligibility score was the main loss here; according to respondents it seems like everyone was very much mutually intelligible so every subject got a score of 1 or 2. It again did result in some sort of upward trend, but it was not clear enough to determine if it was due to correlation.

The main problem that I ended up having with this data was a gross lack of it. Because of Columbia's shortened semester, I could not gather nearly the amount of recordings that I would have liked, and I would have then also made multiple Google Forms to randomize the order in which people listened to the recordings or the recordings which they would hear. Perhaps this can be done in a future iteration of the project, but in the case of this paper, it is important to note that all results have to be taken with a grain of salt.

Hyphenated Minds in general was an absolute gem of a class and hit really close to home during the course of the semester. I myself am a heritage speaker of Russian, and I have taken classes at Columbia specifically tailored to teaching Russian (mostly grammar, writing, and reading) to heritage speakers - something that I did not know was a field that people actually studied at the time, much less made whole courses for. Discussing with other classmates, who were often heritage speakers themselves, about the disconnect we often feel, the world's stance on bilingualism, and the fickle subject of revitalization were some of the most personal discussions that I have had at Columbia.

It was so wonderful to be able to do Linguistics research again! Even though I did not have time to complete this to the perfection I wanted it to be and there is so much left to look at with just this data, I thoroughly enjoyed annotating recordings, going through the form responses, and being able to draw my own conclusions from them. Hopefully I have more opportunities to do so in the future.

You can look at the paper in full here.

Languages of Africa: Maasai(Maa)

Fall 2020

For the Languages of Africa class taught by Prof. John McWhorter, I focused my final report on the Maa language, spoken by the Maasai people in Kenya and Tanzania. The paper focuses on two interesting aspects of the language, specifically the presence and use of advanced tongue-root vowel harmony and the way that words are borrowed into the language from Swahili and English. This is not my own research - it is more in a paper summary format - but personally I found it useful to condense the knowledge I found in other papers into a more comprehensive read. The Maa language is very intricate in many ways, and these two aspects are something I believe even non-linguists can appreciate.

The paper can be found here.

Field Methods: Balti

Spring 2020

As part of the Language Documentation: Field Methods course taught by Dr. Meredith Landman, I got the chance to work on an endangered language from the ground up. The language that the class worked on for the semester was Balti, a Tibetic language spoken in the Kashmiri region of Pakistan. Below are the three papers that I wrote for the course. If you want to look at the final class product, several students worked through the summer as well to create a comprehensive Balti dictionary.

Balti Phonology
Balti Morphology
Balti Syntax

Russian [i] vs. [ɨ] distinction

Spring 2020

During the Phonetics and Phonology class taught by Dr. Meredith Landman, I wrote a paper on Russian's [i] vs. [ɨ] distinction in English. My informants were two Russian-speaking relatives, whom I call Speaker 1 and Speaker 2. Using Praat, I analyzed recordings of both informants reading from a list of Russian and English words containing [i] and [ɪ] to see how they would pronounce them, to determine what a native Russian speaker would consider a phoneme in foreign words. The result is that the [i] vs. [ɨ] distinction exists in English, but [i] vs. [ɪ] distinction does not exist for Russian speakers.

An interesting difference is that because Speaker 1 learned to speak English and not just read it, when she is made aware of this English distinction she intentionally tries to pronounce it with the proper pronunciation, while Speaker 2 cannot distinguish the two sounds even after being made aware of there being a difference in English.

You can read the full paper here.

Thabulou

Summer 2015

As part of the Linguistics course I took in Summer 2015 from Johns Hopkins Center for Talented Youth, one of our ongoing assignments during the three weeks were to create a language from scratch, or a "ConLang". It could have any features that we wanted and its creation was used to test our learning throughout the course. It is unfinished in many ways, as it was done during three weeks and then I only worked on it for a little bit of time afterwards, but it was one of my favorite assignments to work on.

The result is a language that works similar to English in the sense that it is SVO, and it has its own working alphabet and grammar. The most underworked part is the vocabulary, as it is actually a bit difficult to make up words that don't sound like any words you know (which was my goal at the time). It is not able to be used for practical purposes either, as the words that I have come up with over the years are not from the Swadish list and instead are just words that I thought would be potentially useful at the time.

You can look at the documentation for Thabulou here.