Yeah, Um… So Like, Are Filler Words Considered Feminine?

Jennifer Beck, Jaymie Bernardo, Theo Chen, Karl Danielsen, and Calista Eaton-Steinberg

At some point in your life, you have probably experienced the intense awkward silence that comes about when it’s your turn to speak and you have no idea how to respond. Whether you’re not sure how to answer a question or you simply got lost in your train of thought, perhaps you’ve found yourself choosing one of these coping mechanisms to deal with that moment of dreaded stillness in the conversation: (1) you accept the silence and ponder your next move; (2) you fill the silence with filler words to buy time. Filler words such as “like,” “well,” and “um” are a common occurrence for people in conversation who are thinking of what to say. If you pay attention, you might notice that you use these words unconsciously in daily conversation, not even noticing when they slip out.

By observing, collecting, and analyzing video interviews, our study focuses on the correlation between gender and filler words in Californian college students. Studying the use of filler words in different genders of the cis-binary will allow researchers to better understand the way that gender and filler word usage interact. The purpose of this study is to clarify the assumption that women use more filler words than men due to persisting social pressures and the social implications of filler words.

Introduction and Background

Professor Eckert discusses in her linguistic studies that women typically have a different linguistic role in society compared to men (Eckert, 2012, pp. 90). When men speak, they try to keep up a persona that exudes confidence. As filler words explicitly foreground someone’s lack of confidence in speaking – they indicate that the speaker does not feel entirely certain about the things they are saying – men are presumed to more commonly avoid using filler words. In comparison, women generally assume a more mediating role in conversation (Van Herk, 2017, pp.110), so they might be expected to use more filler words.

Finding a connection between gender and filler word usage could indicate that one gender is less affected by the negative traits associated with filler words. In other words, one gender group may feel less social pressure to avoid filler words despite their pre-existing negative implications. Alternatively, one gender might actually prefer using filler words as modes of marking discourse to connect and organize the things they say in specific ways (Divett, 2014, pp. 37-42). A paper in the Journal of Language and Social Psychology found that men and women both use filler words equally when filling pauses, but that women use them more as discourse markers (Laserna et al., 2014, pp. 332-334). In this way, women use filler words to assert their authority in a conversation by directing its path and indicating it is their turn to speak. Due to the unprofessional associations with filler words, we hypothesize that women will use filler words more than men, as women face lower levels of societal pressure to sound professional. They may also utilize these words more often to direct conversation. We conducted a small-scale study of casual interactions between college-age men and women to assess the patterns of filler word use.


We analyzed 15 interviews of Californian college students posted on college-related YouTube channels. These casual one-on-one interviews asked random students basic questions about their college experiences. We looked at results from women interviewing men and men interviewing women and calculated the number of filler words (including “um/uh,” “like,” “yeah,” “so,” “I mean,” and “you know”) relative to the number of total words spoken.                                                                                                                                        

Previous research into this topic suggests that women do, in fact, use more filler words than men (Laserna et al., 2014, pp. 332-334). However, as gender roles become less important to our modern society, the previously discovered results may have become outdated. We set out to see if we could reproduce other studies’ outcomes in a modern, progressive college setting, while simultaneously seeking out answers as to what factors could cause the gendered differences in filler word usage.

While our final results matched those of previous studies in confirming a gender difference, the difference we found was not what we expected. Below are a couple of statistics from our data collection:

Figure 1: The most significant data from our research; note the difference between mean and median results.

Looking at the overall ratio result, our results did not support the previous findings on this topic. Women surveyed actually used significantly fewer filler words than men. Looking at the overall total words to filler words ratio, males displayed a 9.651 ratio, while females displayed a 11.885 ratio, showcasing a 2.233 difference in filler word usage between the two genders. Oddly enough, the median of the data contrasted this. The median female used more fillers than the median male. This could potentially mean that men tend more towards extremes, while women speak more similarly across the board. Indeed, one interview with a male revealed the most filler word usage of all interviews, as the male spoke with almost one filler word per five words.

In spite of the inconclusive results of our mean/median analysis, two segments of the data did show a clear trend. Across all interviews, women and men showed preferences as groups for different filler words. Women favored the word “like,” which is increasingly androgynous but still closely associated with the “valley girl” archetype. Men, in place of using the effeminate “like,” preferred words such as “yeah.” It appears that both genders selected their filler words carefully to index different personas, even if they used filler words at similar rates. This means that social pressure is still strongly at play in word choice, even if neither gender has a stronger need for the confidence lent by decreased filler usage.

Both genders together indicated another interesting trend: the presence of two, not one, spikes on the graph of filler ratios. Figure 1 below shows that there is a peak of people using ~6 words/filler word and one of people using ~13 words/filler word. This two-peak system indicates that there are likely two separate modes of speech people use, one casual with a higher ratio of fillers, and one formal with a lower ratio. Filler word use overall is likely distributed across two standard deviations centered at these spikes.

Figure 2: A histogram showing the number of interviews with a certain filler ratio. Make note of the two separate peaks – one at 6, and one at 13.



Our research shows that the differences in filler word usage across genders are more complex than previous findings suggest. Figure 2 below shows the transcript between two different interviews we observed, both being asked similar questions. You can see the female interviewee produces five filler words out of 43 words total. On the other hand, the male interviewee produces six filler words out of a total 40 words. The margin of filler word usage is slim here. As we mentioned before, females have been found to favor the filler word, “like” while men favored “yeah”. You will note that in this case, the male favored the word, “Uhm.”  While not every male favors the same word, overall data suggests that there is still a generally consistent difference between male and female filler word choice, especially in the use of “like.”

This could be a result of gender stereotypes for speech – “like” and “so” are associated more with femininity, while “um” and “yeah” seem more masculine. There aren’t rules for who can say what, but speech can be very gendered. Part of it might be conscious – for example, males might avoid “like” for fear of sounding feminine – but it might also be a result of who these people are spending time around and what kind of speech they naturally pick up from friends and family.

Figure 3: Transcripts of two interview segments, both involving the opposing gender. Extracted from ProWrite Admissions YouTube channel.


It is important to keep in mind that the data used for our results was extracted from online videos of causal speech. Casual speech with a fellow young person allows for a more comfortable setting, therefore allowing for more filler words to be used. Because these videos were spontaneous and filmed, it is also possible that certain participants were more nervous than others, causing them to use more filler words as they collected their thoughts. Some people are more anxious speaking spontaneously in front of a camera, which would definitely affect their mannerisms, while other people might love being filmed and thrive in the same situation, speaking with confidence and ease.

Our current research sought to analyze the long-lived stereotype of women using more filler words than men, which may exist due to the even older stereotype of women having less intelligence. With these results, we come to the conclusion that college-aged males within California use filler words more frequently in casual speech than college-aged women in California. This could result from a number of factors. For one, more male college students are in STEM fields (Blackwood, 2020) where interpersonal skills are de-emphasized, and students might use more fillers. Men could also be more willing to index a casual persona in interviews because there are fewer expectations against their intelligence that they want to combat. With the persisting sociological stereotypes that deem women less intelligent, women have to work twice as hard in order to gain the respect that men have, especially within the work field (Eckert, 2012, pp. 90). Women are held to different expectations than men, which could impede on filler word usage.

Furthermore, a strong negative social stigma exists around young women who use filler words, especially “like.” Frequent use of the word “like” is a characteristic of the valley girl accent, a Californian accent associated with wealthy, unintelligent, and annoying young women. (This NPR article talks about some other ways that women’s language is stigmatized and disrespected). Since women have to overcome these pre-existing stereotypes, it is possible that they consciously work harder at not using filler words.

Should this research be conducted in another state with another age range, or in a more formal setting, the results may differ. However, our data challenges a conventional understanding of filler word use, suggesting that this topic is very complex and requires further investigation. Potential future research could look into formal interviews between an employer and potential employee, and whether this context decreases filler word use, regardless of gender. Research could also look into stereotypes surrounding different filler words, and whether these stereotypes consciously affect filler word use.



Crimson Education. (2013). Home [YouTube Channel], from

Divett, S., Duvall, E., Graham, T. Robbins, A. (2014) How and why people use filler words (pp. 35-46). /files/2014/12/F2014-Robbins.pdf

Eckert, P. (2012). Three waves of variation study: The emergence of meaning in the study of sociolinguistic variation. Annual Review of Anthropology, 41, 87-100.

Laserna, C., Pennebaker, J., Seih, Y. (2014). Um . . . Who Like Says You Know: Filler Word Use as a Function of Age, Gender, and Personality. Journal of Language and Social Psychology. 33(3), 328-335. DOI: 10.1177/0261927X1452699 OR https://www.researchgate .net/publication /27 5005568_Um_Who_Like_Says_You_Know_Filler_Word_Use_as_a_Function_of_Age_Gender_and_Personality

ProWrite Admissions. (2017). Home [YouTube Channel]. YouTube. Retrieved November 16, 2020, from

Van Herk, G. (2017) Gender. What is Sociolinguistics? Wiley Blackwell. (pp. 97-115)

W.K.C., Kate Blackwood. (2020, July 1.). Gender gaps in STEM college majors emerge in high school. Cornell Chronicle.

Read more

Does She Listen to ‘Girl in Red’? Linguistic Markers in WLW Flirting

Tiffany Dang, Brianna Lombardo, Carlos Salvador Vasquez, Denisa Tudorache, Yuyin Yang

The present article focused on linguistic markers that are adopted by the Women Loving Women (WLW) population when identifying potential members of the WLW community. More specifically, this study focused on the strategies used by members of the WLW community for identifying fellow WLW with the intentions of pursuing a romantic or sexual relationship. Through analyzing popular YouTube videos featuring strategies on flirting with WLW, our first study captured the common beliefs regarding the need to take an extra step, and the possible methods on identifying WLW before taking any romantic or sexual advances. Followed-up by semi-structured interviews in study two with UCLA students who self-identify as WLW, we were able to examine the accuracy of the tips offered by the YouTube videos. This allowed for further investigation on the existence of specific linguistic markers adopted by WLW when flirting. We found that both popular YouTube videos and participants both discussed the need for WLW to take an extra step before they can comfortably pursue another woman and tend to make a conscious effort to not be too direct.

Introduction and Background

While there have been past studies done on examining the speech of gay men, particularly the California vowel shift among gay men (Podesva, 2011), and one that revealed a concept of gay-dar, the belief that gay men possess an ability to pick out each other in a crowd (Shelp, 2003), little research has been done on uncovering linguistic patterns within the Women Loving Women population (WLW). A member of the WLW community is loosely defined as anyone who identifies as a woman and differs from the mainstream preferences in terms of their sexual practice and identity (Eliason & Morgan, 1998). Due to being seen as deviant from the mainstream practices, they may feel the need to take different approaches when making romantic pursuits in order to establish a mutual understanding of their interest in women when talking to another individual. As WLW may often struggle with compulsory heterosexuality, the fear of being perceived as predatory, as well as the potential dangers that come with revealing their sexuality, we aimed to investigate whether there were any linguistic markers adopted among the members of the community to aid in implicitly seeking each other out. This study explores the ways WLW work around the potential barriers they face when pursuing romantic interests and when revealing their identity in hopes of gaining insight on ways to improve the inclusivity of a general community. We hypothesized that WLW would adopt practices where they refer to certain WLW-group-specific terminologies or features before making romantic or sexual advances towards another woman.


Study 1 collected people’s lay knowledge on identifying WLW by looking at popular YouTube videos that featured strategies on how to initiate romantic/sexual advances with a WLW. We found three relatively popular videos created by members of the WLW community who also covered a large realm of dating advice and made a list of those that were related to indexing sexual identity. In addition, we watched two videos that featured heterosexual dating advice and made note of the advice given to men to romantically or sexually pursue other women. By comparing the two lists of notes, we were able to identify potential strategies that are WLW group-specific.

Study 2 consisted of two semi-structured interviews that took place and were recorded through Zoom. We interviewed a total three members of the WLW community, with two of them being in a committed relationship. They were primarily asked to describe and draw from their past experiences. The interviews were guided by six open-ended questions (see Appendix A) with the interviewer following up with questions when necessary. Our questions focused on the WLW’s description of their experiences in establishing mutual interest in women using non-direct measures. Participants were recruited using snow-ball sampling and all answers were kept anonymous. After the interviews, we listened to the audio recordings and made notes of the different ways WLW chose to index their sexual identity as well as the cues they used to determine the sexual identity of their romantic interest.

Study 1 Results

In Study 1, we were able to uncover several recurring themes. One point made consistently across multiple videos was that the WLW always felt the need to immediately make their sexuality known once they realized they had feelings for the other party.

Reasons for this were that they did not want to confuse the other party into thinking that they just wanted a female friend, and they also did not want the other party to assume that the speaker is straight and think differently of them. WLW worry about giving ambiguous signals if they were to not reveal their sexual identity soon enough, which leads to the subtle incorporations of various cues in conversations, such as mentioning the pride parade, to demonstrate their sexual identity.

They also made mentions of lurking through the other party’s social media for signs pertaining to possible membership of the WLW community to know whether it would be appropriate for them to make romantic advances. WLW also tend to be cautious in making advances as they adopt a “flirting by not flirting” technique. This allows them to slowly determine if the other party has reciprocated romantic feelings without being too overbearing and only continue to proceed if there is a positive response.

Figure 1: A selection of videos on WLW flirting used in Study 1

WLW flirting:  Video 1      Video 2     Video 3

In contrast, when we explored flirting advice geared towards men to pursue women, there was no  mention for men to index their sexual identity to women before flirting or at any stage of the courting process. The videos generally focused on advising men to be indirect to increase excitement in women and how to appear playful and masculine.

Figure 2: A selection of videos on heterosexual flirting used in Study 1

Heterosexual Flirting: Video 1      Video 2

Although there was some overlap in advice given to women to pursue other women and given to men to pursue women, such as being subtle and indirect, the reasoning behind it was different,  and a clear difference was the need for WLW to drop hints about their sexual identity. Because there tends to be less confusion in intentions when a male approaches a female, neither party is advised to hint at their own sexual identity nor advised on how to determine the other party’s sexual identity. In contrast, a common theme across videos geared towards WLW is to use references to hint at their own gayness or try to determine whether the other party is gay before advancing.

Study 2 Results

Interview 1

A summary of common themes that arose in Interview 1 are presented in Table 1 below along with some illustrative examples given by the interviewee.

Table 1: Recurring themes and examples from Interview 1

Interview 2

To illustrate the results derived from Interview 2, Table 2 consists of the most important statements made by both Subject 1 and Subject 2 in the conversation. It is important to note that Subject 1 and Subject 2 have been in a WLW relationship for over a year. When answering the interviewer’s questions, they both reflected on when they first met and how this has changed or remained consistent. The middle column consists of what they answered similarly.

Table 2: Noteworthy excerpts from each subject of Interview 2 and areas of overlap


Study 2 Analysis

From our interviews we gathered that the majority of strategies available for Women-Loving Women to identify and flirt with other WLW are mostly non-linguistic in nature. In both interviews, WLW referred to style of dress as a primary identifier for fellow WLW. These and other aspects of popular WLW culture were also drawn upon during the flirting itself, which leads us to one overtly linguistic flirting strategy we found was used by WLW– compliments. Compliments between WLW referenced nonverbal yet mutually understood markers of WLW identity, so they were used to confirm sexuality and communicate an attempt to flirt, in addition to their function as simple compliments. Importantly, compliments between WLW and platonic ones between heterosexual women were said to differ solely in their content and not their form. We conclude that this arises from a need or desire for WLW to flirt “under the radar” to avoid the very real danger of homophobia and bigoted comments.

We also noted the potential for confusion and ambiguous interpretations of these, arguably necessary, nonverbal flirting methods. Subject 1 even described a trend among WLW to pull back on “standard” physical or verbal affection (at least among other WLW) as a way to avoid creating confusion since more open displays of platonic affection are expected among groups of women. This may contribute to a societal perception of WLW as being “colder” or “more masculine.” Future studies might investigate whether or not this is true among a larger sample size.

Figure 3: A meme employing WLW popular music artist ‘Girl in Red’ to euphemistically index a WLW identity


Discussion and Conclusions

Our ultimate takeaway from these interviews was a strong indication that, motivated by a possible fear of negative attention, members of WLW groups feel the need to be covert in romantic contexts. As a result of this covertness, we noticed a trend of relying on nonverbal cues (like clothing choice) more than an awareness of individuals phonetically or lexically indexing their “gayness.”

Even in situations where an individual might directly state “I like girls,” the implication of “I’m romantically interested in you” often remains covert. This gives the other individual a choice as to whether or not an interaction is romantic in nature, but can end up causing some confusion. Thus arises the stereotype that WLW do not flirt. In many cases, their advances can easily be interpreted as platonic interaction among women in a society where affection among women is more normalized than among men, and where revealing your sexuality to the wrong person can have negative repercussions.

Further Reading Recommendations: Although we did not cover this information in our study, there have been numerous studies done on the language WLW may use that distinguish their patterns from heterosexual women. Robin Lakoff in Language and A Woman’s Place (1975), defines stereotypical “women’s language features (WL)” as those associated with “heterosexual women’s performance of femininity.” She contrasts this with the existence of typical “men’s language features (ML),” thus creating a binary of “women’s speech v men’s speech.” It would be interesting to use this and analyze whether women in the WLW community use either one or both of the language features, and whether this could be a distinguishing feature.



 Eliason, M.J., Morgan, K.S. Lesbians Define Themselves: Diversity in Lesbian Identification. International Journal of Sexuality and Gender Studies 3, 47–63 (1998).

Lakoff, Robin (1975). Language and A Woman’s Place. Language in Society, Vol. 2, No. 1, 45-80.

Podesva, R. J. (2011). The California vowel shift and gay identity. American speech, 86(1), 32-51.

Rich, A. (1980). Compulsory heterosexuality and lesbian existence. Signs: Journal of women in culture and society, 5(4), 631-660.

Rieger, G., Linsenmeier, J. A., Gygax, L., Garcia, S., & Bailey, J. M. (2010). Dissecting “gaydar”: Accuracy and the role of masculinity–femininity. Archives of Sexual Behavior, 39(1), 124-140.

Shelp, S. G. (2003). Gaydar: Gaydar. Journal of Homosexuality, 44(1), 1-14.


Read more

“Sorry, I Didn’t Quite Get That: The Misidentification of AAVE by Voice Recognition Software”

Shannon McCarty, Lam Pham, Alora Thresher, Alexandria Wasgatt, Emma Whamond

This study investigates the transcription accuracy by AI speech recognition systems using natural language processing when interpreting standard American English dialects (SAE) versus African American Vernacular English (AAVE). We inspect the percentage of misidentified words, and the degree to which the speech is misidentified, by AI speech recognition systems through analyzing authentic speech found in YouTube videos. The accuracy of voice recognition with respect to AAVE will be determined by selecting for distinct AAVE features, such as G-dropping, the [θ] sound, reduction of consonant clusters, and non-standard usages of be. The methodology includes feeding YouTube clips of both SAE and AAVE through an AI speech recognition software, as well as examining YouTube’s auto-generated transcripts, which are created by automatic speech recognition based on the audio of the YouTube video. The purpose of this study is to bring attention to the needs of diversity in technology with regard to language variation, so that AI speech systems such as Amazon’s Alexa or Apple’s Siri are more accessible to all members of society, as well as to help destigmatize a variety of American English that has carried social, cultural, and historical stigma for centuries.

Introduction and Background

2020—American society finds itself at the thrilling forefront of technological innovation, yet is still plagued by racial inequality and systematic racism. Our study aimed to tackle a portion of this American duality from a sociolinguistic standpoint. We investigated AI speech recognition systems and the identification differences while processing speech of the standard American English dialect (SAE) versus African American Vernacular English (AAVE). SAE is most broadly defined as the most uniform, accepted, and understood language in the US. On the other hand, AAVE is not solely slang or a lesser form of SAE; rather, it is a language variety that has systematic linguistic patterns and carries social, cultural, and historical stigmas. We believe AI speech recognition systems will reflect our country’s racial biases by misinterpreting AAVE far more often than SAE. As these technologies become more commonplace and embedded in society, our study’s goal is to shed light on whether AI speech technology is inclusive of the AAVE dialect and its speakers.


An overview of AAVE linguistic features can be found in a paper discussing the matter by Erik Thomas (2007), though we narrowed our focus on the most defining features of AAVE in this study. These included G-dropping, the th sound (as in bath) becoming the f sound (as in fast), reduction of consonant clusters at the ends of words (wes side versus west side), and the use of the verb be (Singler, 1998).

The analysis of the main facets of the verb be included the dropping of be, the habitual be (Collins, 2006), the use of BIN and be done. The heavy focus on be was due to the fact that it is one of the biggest syntactic differences between SAE and AAVE (Lanehart, 2015). As a result, the variants of be  were most likely to affect the transcription performance of voice recognition software when processing AAVE. 

This study used the abundant linguistic resources available from YouTube to exhibit the use of SAE and AAVE dialects. We compiled audio clips of authentic, normal speech in both of these varieties and fed the audio clips into an AI transcription software, specifically, as well as examined YouTube’s AI-generated closed captions. It was important to ensure that the captions we used on YouTube were the auto-generated ones, and not captions that had been manually entered by the uploader.

To determine what percentage of speech and what types of features of the dialects were misidentified by the AI speech recognition systems, we quantified the data by categorizing the accuracy of the AI transcriptions into four groups: “All,” meaning All of the words were picked up, “Most,” “Few,” or “None.”  If every single word in the audio recording was transcribed correctly by the AI speech recognition system, then that audio clip was placed in the “All words picked up” group. If most words or only a few words were transcribed correctly by the AI speech recognition system, then that audio clip was placed in “Most” and “Few” groups, respectively. And if nothing in the audio clip was transcribed correctly by the AI speech recognition system, then that audio clip was placed in the “None” group.

Our hypothesis was that the percentage of interpretation inaccuracy in AI speech recognition systems while interpreting AAVE would be significantly higher than the inaccuracy recorded by speech recognition software while interpreting SAE (i.e. more AAVE audio clips would land in the “Few” and “None” transcription accuracy groups). This would lead to an expectation that voice recognition software is globally less effective for speakers of AAVE.


The results of this study were overall in line with the hypothesis—the transcription software picked up more SAE words than AAVE words. The vast majority of AAVE video samples collected had “Few” words transcribed correctly. Figure 1 shows the amount of AAVE words picked up by the transcription software. Our study found that for all of the clips of AAVE:

    • 16.7% “All” words were picked up
    • 16.7% “Most” words were picked up
    • 56.7% “Few” words were picked up
    • 10.0% “None” of the words were picked up

It is noteworthy that the transcription software was unable to pick up the majority of AAVE speech when the video clips were similar to the SAE video clips with regard to background noise, speaker volume, etc. We believe this is largely due to AAVE phonetics rather than AAVE syntax and word choice, which is what we were focusing on in this study.

Example 1: This clip was categorized under “Few” words being picked up; in the beginning of the segment, the speaker actually says it’s a whole buncha people in New York that we know from. (“African-American English in North Carolina”, The Language & Life Project)

On the other hand, all video samples of SAE fell under “All” or “Most” transcription accuracy. The SAE video clips had the majority of their words picked up and transcribed correctly. According to Figure 2:

    • 31.6% of the clips had all their words picked up
    • 68.4% had the majority of their words picked up

The results of transcribing SAE versus the results of AAVE are drastically different; we believe this is largely due to the phonetic differences between the two English varieties. The only time SAE was not picked up was when the speaker was speaking faster than normal. All of the words could have blended together, and as a result, the transcription software was not able to distinguish what was being said.

Another unexpected observation for both AAVE and SAE was that women speakers were not picked up as often as male speakers; AAVE female speakers made up 66.7% of the “None” category, despite comprising less than 20% of our AAVE samples.


We had some limitations in our project as well. Due to COVID and time restrictions, our sample size was regrettably small at only 49 clips. We also had a few clips become unusable between the time of our finding them and analyzing them. As a result of the small sample size, we ended up noticing a huge, unexpected difference between the transcription accuracy of male versus female speakers of AAVE, but we weren’t able to draw any solid conclusions as to whether this difference is representative of a larger population. However, the difference was quite apparent within the random pool we gathered, so a study with equal amounts of male and female AAVE speakers could address this issue more equitably.

Speakers that were present in the room would also have enabled a more accurate representation of AAVE and voice recognition software’s real-world interaction. Our computers and phones were limited due to the differing microphone quality, which could have influenced the amount of words the voice recognition software picked up. In addition, we had planned to use our phones’ dictation softwares as the primary method of data acquisition, which ended up being wholly impossible, so we only had the time to find and use one transcription site.

Many of the video clips were filmed in neighborhoods, with background noise included, such as cars going by, wind, and people talking. This could also have limited our transcription site and created another layer of ambiguity. However, SAE clips with similar amounts of background noise that were played were transcribed (mostly) correctly. So it is unclear if the background noise is fully a limitation or an example of voice recognition software not picking up AAVE.

Lastly, we began to wonder if the phonetics of AAVE had more of an effect on our voice recognition software than the syntax did—that is, we wondered whether the sound of AAVE was more impactful than the sentence structure. A few of us used the software to speak sentences that were syntactically AAVE, but not phonetically, as none of the researchers of this study were AAVE speakers. Those sentences were transcribed correctly, including AAVE syntax. However, we did not have enough time to pursue that avenue of research, but it would be a promising starting point for any future projects.

Discussion and Conclusions

This investigation was intended solely to pursue the question of whether there is a racially-based difference in accuracy of voice recognition software. Despite the unexpectedly small scope of our study, we believe our results are sufficient to prompt further investigation into the reasons as to why there is such a stark difference in accuracy.

To that end, a question that arises naturally is whether closing the gap in accuracy is as simple as writing and including a few more lines of code. If it is, then what’s preventing this from happening now? And if the voice recognition software instead needs to be re-constructed from the ground up, then that, in turn, spawns a whole host of follow-up questions (e.g. Who’s paying for this development? Who’s working on it? How long will it take?).

It is, as with any discussion of racially-based imbalances, also worth considering systemic variables at play. Another question worth pursuing is whether there is a higher average level of inaccessibility to the technology sector for speakers of AAVE compared to speakers of SAE, which would contribute to a broad range of consequences—one of which might be the discrepancy in accuracy of voice transcription softwares demonstrated in this study.

In 2016, Rickford and King cited Schneider (1996) to describe AAVE as “the US English dialect most examined by linguists for quite some time.” In the academic world, it’s common knowledge that the prejudices held against AAVE  — and its speakers — have no basis in fact. The drastic difference in voice recognition software transcription accuracy between SAE and AAVE is not only one more imbalance to correct in pursuit of a fully equitable society, but also a symptom of the systemic racism that influences all aspects of daily life. Studies like this one which stop at identifying the problem are only the first step; the next is to examine why the problem exists in the first place, so that work to resolve the inequity can begin.


Collins, C. (2006). A fresh look at habitual be in AAVE. CREOLE LANGUAGE LIBRARY, 29, 203.

Koenecke, A., Nam, A., Lake, E., Nudell, J., Quartey, M., Mengesha, Z., Toups, C., Rickford, J., Jurafsky, D., & Goel, S. (2020). Racial disparities in automated speech recognition. Proceedings of the National Academy of Sciences, 117(14), 7684-7689.

Lanehart, S. (Ed.). (2015). The Oxford Handbook of African American Language. Oxford University Press.

Lippi, R., Donati, S., Lippi-Green, R., & Donati, R. (1997). English with an accent: Language, ideology, and discrimination in the United States. Psychology Press.

Rickford, J. R., & King, S. (2016). Language and linguistics on trial: Hearing Rachel Jeantel (and other vernacular speakers) in the courtroom and beyond. Language, 92(4), 948-988.

Singler, John Victor. “What’s not new in AAVE.” American Speech 73.3 (1998): 227-256.

Thomas, E. R. (2007). Phonological and phonetic characteristics of African American vernacular English. Language and Linguistics Compass, 1(5), 450-475.

Read more