The Canadian Shift: Still shifting?

Previous literature on the Canadian Shift describes this phenomenon as a change in progress in many dialects of Canadian English. However, elements of the shift are not found to be consistent, particularly in the lowering of [ɪ] and [ɛ] and the retraction of [æ]. This paper investigates apparent time data from eight native speakers of Canadian English from the Ottawa area to investigate the nature of the Shift in the region, as well as to better understand how the Shift is manifested here compared to previous literature. Results presented in this paper, which were collected as part of an ongoing study, show that younger speakers produce only [ɪ] and [ɛ] vowels more retracted than older speakers. These data will later be compared to results of a perception study in order to investigate the relationship between perception and production of a sound change in progress.


The nature of the Canadian Shift
The Canadian Shift is an established phenomenon present in many dialects of Canadian English (Sadlier-Brown and Tamminga 2008). The Canadian Shift is described as a pull-chain involving the front-lax vowel system. The shift involves the lowering and retraction of [ɪ] and [ɛ] (higher F1/lower F2), the retraction of [ae] (lower F2), and the centralization of [ʌ] (Clarke et al. 1995). This pull-chain is illustrated in Figure 1. The cause of this vowel shift is possibly related to the low-back merger, in which the vowel [ɔ] merges with [ɑ] in many dialects of Canadian English. With these two back vowels merging, a space is created within the vowel system. According to the vowel dispersion principle, the space vowels occupy is evenly spread out (Ashby and Maidment 2005). When a change in the vowel system like the low-back merger occurs, the remaining vowels may shift to accommodate the opening (Sadlier-Brown and Tamminga 2008).

The current experiment
The current experiment investigates the nature of the Canadian Shift as spoken by native speakers of Canadian English from Ottawa, Ontario. Ottawa provides a unique geographical and linguistic setting for which investigating conflicting elements of the Canadian Shift may prove useful. Particularly, due to its layout, the city includes urban, suburban, and rural areas. Previous literature has shown confounding results when investigating urban and rural communities, where age-related differences in vowel production are not statistically significant (e.g., Roeder 2012). Because of this, it may be useful to investigate one city with such a variety of communities in effort to tease apart the effect of age and the Canadian Shift. To do so, this study seeks to answer the following research questions: (1) a. In Ottawa, do younger Canadians produce more shifted vowels than older Canadians? b. Are there age-related differences in how vowels in the Canadian Shift are produced in Ottawa?
In order to address these questions, a production experiment was conducted on eight participants, four of which were between 18-30 years old, and four which were over the age of 60.

Participants
The current experiment investigated eight native, monolingual speakers of Canadian English from Ottawa. Four were between the ages 18-30, and four were over 60 years old. 2 All participants were recruited through social media (reddit, Facebook, and Instagram), or through word of mouth. Information collected from each participant includes their age and where they grew up.

Stimuli
The stimuli included the entirety of the Canadian English monophthong vowel system ([i, ɪ, ɛ, ae, u, ʊ, ʌ, ɑ]) situated in monosyllabic words between combinations of voiced and voiceless bilabial, alveolar, and velar stops. The carrier word occurred sentence-final, and there were a total of 103 sentences. While all monophthongs were included, this was just to provide a more complete picture of the vowel space, which is relevant when normalizing formant values from multiple speakers, and the results show only the vowels of interest ([ɪ], [ɛ], [ae], and [ɑ]). A full list of the stimuli is found in the Appendix.
Recordings were conducted in the Sound Patterns Laboratory's sound-attenuated booth at the University of Ottawa using a Philips Omnidirectional Handheld microphone (LBB 9080/55 NR552) and recorded in Audacity® (Dannenberg and Mazzoni 2015). The stimuli were presented through PowerPoint, each sentence was shown on a slide and the slides were advanced at the participant's own pace, the sentences were read through three times, and were randomized each time.

Acoustic analysis
Acoustic analyses for the data were conducted in Praat [V. 6.0.43] (Boersma and Weenink 2018). After importing the sound files, text grids were created manually for each of the sentence-final target words. The formant values (F1, F2, and F3) at the beginning, midpoint, and at the end of the vowels were automatically extracted using a script (Buss 2014). The results and analyses presented below focus only on the midpoint formant values.

Establishing the vowel space
The raw formant values from each token and speaker were normalized using the Nearey 1 method (Thomas and Kendall 2007) to look for any initial trends amongst the speakers. 3 Figure 2, younger speakers are shown in grey while older speakers are in black. It is important to discuss the logistics of using the Nearey 1 normalization formula. An advantage for Nearey 1 is that it provided normalized formant values that filter out physiological differences and maintain sociolinguistic variation. However, there are also disadvantages. These include the possibility of a skewed plot if an incomplete vowel inventory is provided, and individual dialectal differences can also result in skewedness. As this plot shows the values from just eight speakers, who did not produce diphthongs, the plot can be skewed. However, the plot is a useful tool to illustrate the comparison of two group's vowel space, for initial observations. Based on this plot, the younger speaker group's bit vowel does not appear much lower, but it is slightly more back. The bet vowel is slightly lower and more retracted, the bat vowel is lower, and the bot vowel is nearly overlapping between the two groups. These observations are further investigated through statistical analysis.

Mixed effects linear regression results
The normalized values for the vowels of interest ([ɪ], [ɛ], [ae], and [ɑ]) were used to run several mixed effects linear regression in R [V 3.5.1] (R Core Team 2018). In each model, the dependent variable was either the normalized F1 or F2 value, and the independent variables were age group (young) and speaker mean F0 (not reported here).
The statistical hypotheses in (3) and (4) were used to determine how to appropriately answer the research questions from (1), which are restated in (2).
(2) Broad Research Questions a. In Ottawa, do younger Canadians produce more shifted vowels than older Canadians? b. Are there age-related differences in how vowels in the Canadian Shift are produced in Ottawa?

The statistical models
Each vowel was run in its own model, with separate models for F1 and F2. 4 The following eight tables show each vowel's mixed effects linear regression models for F1 and F2. For these models, the older age group serves as a baseline. These results are discussed further in Sections 3.  4 An example of the model for bit vowel is > model.bit.F1.b <-lmer(nenormF1 ~ agegroup + meanF0 + (1 | speaker) + (1 | actual.word), data=data.bit, control=lmerControl(optimizer="bobyqa", optCtrl=list(maxfun=2e5))). 5 The p-values were calculated using a t-value to p-value online calculator.

Age and F1
In terms of lowering (higher F1), no vowel is significantly lower, as illustrated above in Tables 1, 3

Age and F2
In terms of retraction (lower F2), the results of the linear regression suggest significant retraction of [ɪ] and [ɛ], shown above in Tables 2 and 4 -backing chain shift, which is also found to be significant in Kettig and Winter (2017). However, this is not evident based on the data provided by these eight speakers.

Discussion
For the vowels involved in the Canadian Shift that were included in this analysis ( In this experiment, in terms of lowering, the younger speaker group did not produce any statistically lower vowels than the older speakers. Retraction was only significant for [ɪ] and [ɛ], and the retraction of the other vowels was not statistically significant. In terms of the statistical hypotheses in (3) and (4) the null hypothesis can be rejected as there was a significant difference in one of the formant values (F2) as produced by younger speakers.
As for the research questions outlined in (1), these can be tentatively answered given the results from the current sample population. In general, younger Canadians do produce more shifted vowels than older Canadians as expected based on previous literature. While significance was not found except for the retraction of [ɪ] and [ɛ], it is possible that this result can change with further analysis of the other 12 speakers in this study. For the second research question, there is evidence that age is a significant factor in predicting shifted vowels in accordance to what is expected with the Canadian Shift. At this point, these data suggest the Canadian Shift is active to some extent in Ottawa.

Next steps and future research
As this is an ongoing experiment, further data analysis is required before determining the behaviour of the Canadian Shift in Ottawa. Incorporating the remaining data from 12 other speakers will provide a better understanding of the way the Canadian Shift is produced in Ottawa. Specifically, more data may illustrate how younger speakers are producing the Shift, to what extent its behaviour is expected based on previous literature, as well as how to determine if the Shift is still an ongoing change in progress.
The data presented here show only a piece of what the Canadian Shift looks like in Ottawa. Not only due to a small sample size, but also in regards to the other side of a change in progressperception. Each of the participants in this experiment also completed a perception task, and final versions of this study will include the relationship between perception and production on a group and individual level.

Conclusion
What appears most salient from these data was how younger speakers produced a significant retraction of [ɪ] and [ɛ]. The other vowels involved in the Canadian Shift were not significantly shifted when produced by younger speakers, which at this point, only conflates the issue of inconsistencies from previous research in production. It is possible that Ottawa's mixture of urban and rural identities contributes to the manifestation of the Shift, as it did in Roeder's (2012) study. However, further analysis is required in order to determine the status of the Canadian Shift in Ottawa. If any conclusions can be made at this point, it is that there is still much to investigate on the Canadian Shift, and the way it is understood now may not be fully representative of what exactly this phenomenon entails.