10

Working sessions
50

Hours of working
2

Number of participants
Applied Linguistics | Sber
Study of gender speech patterns of operators
You need to create accurate bot personas for effective dialogue. One of the important characteristics of these personas is the tone of voice and the vocabulary of standard phrases. The family of virtual assistants Salut consists of representatives of masculine and feminine and we decided to check how much their speech patterns correspond to the features of male and female speech and whether such patterns exist in reality.

My role: VUI Researcher, methodology designer
Credits: Anna Aksenova, Applied Linguist, chapter Bots / Sber
Platform: SberBank mobile app and Smart Devices of Salute / Sber
Aproach and Key components of the project
The project was a desk study consisting of two parts - an introductory study to test the methodology and an analytical study, which included an analysis of scripts of call center operators' dialogues.
The main objectives of the study were:
  • Is the speech of men and women different?
  • Do these differences persist in our speech domain?
  • How aware are the clients of these differences?
  • Can these differences be used for scripting guidelines
Methods used:
  • Lemmatization
  • Quantitative analysis
  • Calculation of the relative frequency of words and its application
What have been done before
A competition to determine the gender of a tweet's author (2019)
Best determination quality >90%, certain patterns identified
Research on pronouns in the speech of men and women

Pronouns build our cognitive portrait, reflect our communication strategy
Exploring Strategies of Dialogue Building by Men and Women
Women - Supportive
Men - Competitive
Development of methodology and aproach
We analyzed the profile literature and research and identified a basic approach to analyzing gender patterns. We were particularly assisted by the literature on criminology and research on speech patterns in culture.

Based on our analysis, we identified 3 main areas of focus:
  • Vocabulary
  • Morphology and morphemics
  • Syntax
This fragment of research Miro research board with 3 main areas of research focus
Quantitative analisys
We used scripts of the call center and chat operators as the basis of our research. During research 3,000 dialogues of male and female operators were analyzed.
The first thing we did was a quantitative analysis of the texts, and already at this stage we discovered a number of interesting things...
Table with operator dialog scripts
There was no noticeable difference in the distribution of parts of speech between male and female operators. This is largely due to the fact that even in cases that allow freer communication, the operators have to adhere to the rules of communication adopted by the bank.
Distribution of parts of speech in male and female operators
When analysing pronouns, there were more visible differences in feminine and masculine speech. We found interesting correlations in the use of the pronoun 'I' and 'You', as well as in the frequency of using the pronoun 'This'.
Distribution of parts of speech in male and female operators
What have we learned
Men use the pronouns "I" and "You" more often
Analysis of dialogues revealed that male operators clearly try to divide areas of responsibility — "I am responsible for this...", "and you should do this". In general, men's speech is more directive.
Female operators use the pronoun We more often
Women tend to offer cooperation in solving issues — "Let's see with you", "Let's try together"...
Binding to a channel
Male operators use more words of foreign origin and jargon: 'id', 'bug', 'widget', while women use clerical: 'in addition', 'in general', 'exactly', 'quite'
Lexical diversity
In general, the speech of male operators was characterized by a large vocabulary and lexical diversity. There was a desire to demonstrate competence in the issue, confidence and intelligence.
Position in communication
Female operators demonstrated greater empathy toward clients, restraint, and modesty. This was expressed in shorter and simpler phrases, a balance between the length of the client's and the operator's speech, and the absence of special terms and neologisms.
Role in dialogue
Male operators tried to demonstrate the role of an expert and explain the reason for the problem and the logic to the client. Associated with this is the accentuated use of the adjectives THIS and THAT. At the same time, female operators acted in the role of performer and helper.
As a result of the project we formed a matrix of features of speech patterns of male and female operators. This material allowed us to make improvements to the Virtual Assistants Guide. The next stage will be the development of metrics to analyze the effectiveness of changes in dialog scenarios and A/B tests.