Observer's paradox: data from the interviews in the Ustja River Basin Corpus

In this work the data from the Ustja River Basin corpus is analyzed in terms of the so-called Observer’s Paradox. Observer’s paradox was introduced in the works of William Labov. He discusses the idea, that all of the linguistic data collected through the interviews is, in fact, different from the casual speech of the individuals observed. It is paradoxical in the way, that the only way to obtain the «casual» speech data is to systematically observe one speech. This idea caused a lot of debated around the validity of the data collected using the interviews. Some of the authors were arguing, that the individual's speech is shaped primarily by the listeners, which may explain why the observed speech is indeed different from the «casual» speech. However, here is no quantitative studies concerning the effects of the interviewer. In this study I will use 178 interviews from the corpus annotated for 13 different linguistic variables (one variable can have dialectal or standard realization) and the exact timestamps of those variables in the interviews. In this work I will hypothesise, that the usage of dialect realzations will increase towards the end of the interview, as the speaker will lose self-awareness and starts to speak more casually during the interview. When the individual interviews were analyzed, I have found that the strategies are different among the variables, but the prevailing strategy is still the increase of usage of dialect realizations towards the end of the interview. Then, when the Generalized Linear Models were applied to the data, I have found that the position has little or no effect on the realization of variables, when all the data is taken in consideration. When the interaction with gender and variables were analyzed, some variables have demonstrated the increase of dialect realizations towards the end of the interview (with few exceptions). The timestamp seems to be a good predictor for the realization of a variable when the data obtained from male speakers is studied.