<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Grace Woo on Medium]]></title>
        <description><![CDATA[Stories by Grace Woo on Medium]]></description>
        <link>https://medium.com/@grace_52247?source=rss-4b41ded9a6f9------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*LcUdNgWSA-lwJbwLa_FF3Q.jpeg</url>
            <title>Stories by Grace Woo on Medium</title>
            <link>https://medium.com/@grace_52247?source=rss-4b41ded9a6f9------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 01:39:14 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@grace_52247/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Visualizing the Ebb and Flow of Feelings at Your Company]]></title>
            <link>https://medium.com/@grace_52247/visualizing-the-ebb-and-flow-of-feelings-at-your-company-1152ef4c7386?source=rss-4b41ded9a6f9------2</link>
            <guid isPermaLink="false">https://medium.com/p/1152ef4c7386</guid>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Grace Woo]]></dc:creator>
            <pubDate>Fri, 01 Jun 2018 21:55:42 GMT</pubDate>
            <atom:updated>2018-06-01T21:55:42.906Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zN9p652n0RUY_ycy_CsNhA.jpeg" /><figcaption>What was the sentiment cultivated at this moment in time? Here’s how we can help make a guess.</figcaption></figure><p>In our previous post, we explored the rhythm of speech in a group setting. <a href="https://medium.com/cultivate-ai/cultivating-equal-airtime-7970cd183dd4">Inspired by a TechCrunch transcript from Leena Rao and the founders of Honest Company</a>, we analyzed the airtime distribution between founders and found the timing and rhythm of Leena’s moderation role to carry weight despite not necessarily “speaking” the most. We found that airtime is not necessarily associated with speech time but may just as well be associated with visual appearance, body language, as well as emphatic cues.</p><p>While it was straightforward to look at the relative airtime amongst team members and ultimately deduce a rhythm for how to create a productive conversation, we had yet to look at the semantic content itself. Analyzing sentiment against time is a complex task with many dimensions and angles to consider. There are many tools available on the market for this type of analysis. For example, you can play with the open source Python NLTK sentiment analysis <a href="http://text-processing.com/demo/sentiment/">here</a>.</p><p>It has long been known that the amount of airtime each participant contributes is a factor in predicting overall output. With this in mind, Cultivate’s dashboard is created from analysis beyond counting airtime. The data science team has developed several in-depth natural language processing techniques to produce metrics from business e-mail, chat records and recorded forms of team conversation.</p><p>For example, one of the approaches currently being implemented at Cultivate is sentence-by-sentence analysis capturing words that fall into pre-defined sentiment categories like: afraid, amused, angry, annoyed, apathetic, happy, inspired and sad. As such, an overall cumulative score can be visualized over the course of a conversation for each one of these sentiments.</p><p>Here is the overall normalized content for scores associated with words that were used during the last Presidential debate. This shows a very constant and even pacing in the professional rhetoric. It is only when we zoom in very closely, that we can see some of the subtle shifts in the positioning and ranking amongst the different sentiments:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/762/1*dmoB36mkaaiMm2hKg-XHhw.png" /></figure><p>Overall, as the conversation progressed throughout the debate, the statistics show AFRAID and ANGRY as the dominating sentiments when examining the rhetoric. Later on, we can see ANGER dominating the conversation and words associated with HAPPY moving away from the other sentiments:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/790/1*BvJi-1SJ7Q-uezwgVF7YVg.png" /></figure><p>The x-axis on these graphs show only sentence count and is, unfortunately, not correlated with precise timing data. Without timing information from the audio itself, it is difficult to get a handle for how the ebb and flow of sentiment changed throughout the course of the debate. In order to get some of these sentiments correlated with some video data, we need to first associate the sentences with the actual speaker. Then, using hand-transcribed text, we need to coordinate the actual timestamps with the sentence-based sentiment analysis. Finally, we put two and two together to visualize a .srt file that looks like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/760/1*NzQOnCYIRhEfoQOk4T1aCw.png" /><figcaption>A .srt file records a manually cleaned-up version of what was said and best guess precise timing data for when it was said</figcaption></figure><p>The .srt file has an ID number which increases in sequence. There is also a timing start and end for each phrase that is mentioned. Cultivate analysis led by Chief Data Scientist Andy Horng adds speaker data along with a best guess for next speaker.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/570/1*9I7h_8HcaFbX0RIrPpvJiw.png" /><figcaption>Sequence numbers are included at the beginning of the file with the speaker annotated. Then, the best guess for the next speaker is included at the end of the speech segment</figcaption></figure><p>Combined with Cultivate’s method of sentiment analysis using semantic data, we are able to get a temporal-based estimate of what the emotional energy being generated was at the time. Ultimately, we want to be able drive positive energy within a team conversation using speech and semantic content.</p><p>Using a little ASCII hack in a SRT file, I visualized some of the emotion scores generated by Cultivate metrics. Specifically, I looked based off of the zoomed-in scores at what point the discussion became more angry or more happy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/231/1*ujqcByaQJcIrMBSbcEcNpg.png" /><figcaption>Moderator Wallace has a change in emotion early on according to Cultivate metrics</figcaption></figure><p>Cultivate’s semantic analysis conveys quite a bit of information about sentiment. I took two screenshots of the video corresponding to when the analysis suggested more angry rhetoric vs. more happy rhetoric. See if you can tell the difference:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/585/1*8crqtY6r7NE27KLPOsr0hw.png" /></figure><p>Overlaying video data with the plethora of sentiment data available using Cultivate’s analysis gives potentially more data than facial recognition. -But, more on that later.</p><p>To see the full video, see this <a href="https://youtu.be/tpUesms6Tzg">YouTube link</a>.</p><h3>About Cultivate</h3><p>Cultivate’s AI-powered platform enables engagement and inclusion by helping you understand and improve your workplace communications.</p><p>For more information on what we are doing at Cultivate, check out <a href="http://cultivateai.com/">our website</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*oLYJhouZ7giM2kH-yr6dBQ.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1152ef4c7386" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cultivating Equal Airtime]]></title>
            <link>https://medium.com/cultivate-ai/cultivating-equal-airtime-7970cd183dd4?source=rss-4b41ded9a6f9------2</link>
            <guid isPermaLink="false">https://medium.com/p/7970cd183dd4</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[leadership-development]]></category>
            <category><![CDATA[teamwork]]></category>
            <dc:creator><![CDATA[Grace Woo]]></dc:creator>
            <pubDate>Mon, 12 Mar 2018 21:42:02 GMT</pubDate>
            <atom:updated>2018-03-12T21:50:25.127Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_kV9X7lnP5ByYw9zyJJOjw.jpeg" /></figure><p>In my previous article with Cultivate, we used a variety readily available transcription and common shell scripting tools to analyze speech patterns in TechCrunch pitch competitions.</p><p>Inspired by <a href="http://amj.aom.org/content/early/2017/04/29/amj.2016.1215.abstract">a study out of Columbia University</a>, which found differences in the types of questions posed to men and women in a pitch competition setting, we also decided to conduct our analysis through the lens of gender. Our most interesting finding: in the four TechCrunch pitches we analyzed, we observed that <a href="https://medium.com/cultivate-ai/unpacking-the-role-of-gender-ed-speech-in-tech-35b51627dcfc">women tend to use the word <strong>we </strong>more often in their pitches than men do</a>.</p><p>As next step we decided to look at what <strong>we</strong>-focused airtime could mean from a data analytics perspective — this time by looking at how airtime is shared between team members.</p><p>We referred to an article in the <a href="https://www.nytimes.com/2016/02/28/magazine/what-google-learned-from-its-quest-to-build-the-perfect-team.html">New York Times</a> documenting Google’s well-known “Project Aristotle”: a project analyzing why some work teams thrive and others struggle. As study of team dynamics and productivity, there are many takeaways from this project in reference to verbal airtime.</p><p>To paraphrase the high-density article: “good” teams can look very different from one another in terms of team makeup, but share certain behavioral commonalities. Namely, team members speak roughly the same amount in meetings, and the group has a high emotional sensitivity to its members.</p><p>We were interested to see whether it was possible to take a data-driven approach to the former trait, which researchers in the article refer to as “equality in distribution of conversational turn-taking.” Again we decided to use publicly available TechCrunch meetings and conversations as our data set.</p><h3>Calculating Airtime</h3><p>One limitation of the free transcription tools available through YouTube is the ability to distinguish different speakers. For example, in the previous post, we were not able to automatically distinguish between content spoken by the person giving the pitch and questions asked by the judges. To close this gap, we wrote a simple Python tool which annotates the transcripts output by open source tools (specifically, the .json transcript files written out using <a href="https://github.com/agermanidis/autosub">Ager Manidis</a> transcription tools, which contain the start time, end time, and content of each statement).</p><p>Our human-assisted script asks for additional hand-entered annotation for who is actually speaking. After I went in and hand annotated each one of these segments, it became possible to distinguish each individual participant in the group conversation.</p><p>Next, I wrote another Python script to both record the name of each person speaking and tally up the total amount of airtime for each participant.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/850/1*d6QvTPPoxqZBN92iGEPleQ.png" /></figure><p>For this blog post, I chose to try out our tools on a <a href="https://www.youtube.com/watch?v=wZchcGX1hYI">TechCrunch interview</a> moderated by technology journalist Leena Rao and featuring Brian Lee and Jessica Alba of <a href="https://www.honest.com/">The Honest Company</a>.</p><p>Here’s a visualization of the results — a histogram showing seconds of airtime for each of the three interview participants:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/540/1*YlhZpO_RQnnafEcGmq-nrw.png" /></figure><p>The actual airtime for Leena Rao, Jessica Alba, and Brian Lee was 216.094, 517.888, and 555.264 seconds respectively, in a clip that was 1670 seconds long (approx. 27 minutes). The clip also contains about 382 seconds (approx. 6 minutes) of non-speaking time.</p><p>Going off <a href="https://www.nytimes.com/2016/02/28/magazine/what-google-learned-from-its-quest-to-build-the-perfect-team.html">Project Aristotle</a>’s findings, we would expect a similar share of airtime between the co-founders to be indicative of a more productive team. In this interview, that appears to be the case, with co-founders Jessica Alba and Brian Lee speaking a similar 517.888 and 555.264 seconds each respectively.</p><p>What role might a discussion moderator (or perhaps, a manager) play in delegating equal airtime? In her interview with Alba and Lee, moderator Leena Rao delegates most of airtime to the two participants, speaking the least at 216.094 seconds of airtime.</p><p>How did Leena use that airtime to lead the discussion? After hand annotating each person’s role in the transcript file, we were able to see that Leena’s well spaced questions over time are likely what was able to drive a balanced conversation:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/555/1*4M74kKSNg86mRnVIm2MtDQ.png" /></figure><h3>Facilitating Balanced Participation</h3><p>As a manager or a discussion leader, balancing airtime can come in many different forms. In her Forbes article “5 Simple Things You Can Do To Get People To Speak Up In Meetings,” business author <a href="https://twitter.com/@erikaandersen">Erika Andersen</a> cites five qualitative suggestions for leaders looking to boost team participation in discussions:</p><ul><li><strong>Give people lead time</strong> — If you’re planning to bring up a new topic or conversation in a meeting, let the team know in advance.</li><li><strong>Frame it up</strong> — Be specific when you ask questions. Your team needs context to be able to voice opinions usefully.</li><li><strong>Stop talking</strong> — Let your team get a word in. Your team can’t voice their opinion if you’re taking all of the airtime.</li><li><strong>Take it in</strong> — If you shoot down a team member’s idea, when possible tell them what you liked about the idea before you raise your concerns. This will encourage them to speak again in the future.</li><li><strong>Use what you hear</strong> — Reward your team for speaking by putting their ideas into action and giving them credit.</li></ul><p>Looking at Leena’s participation timeline alone, we can infer that she is employing at least two of Andersen’s suggested strategies for facilitating discussion.</p><p>Her comments are evenly spaced and give Alba and Lee time to contribute. She also spends significantly more time listening and processing the content she hears in comparison to Jessica and Brian, taking up right around 20% of the total airtime. That’s some pretty smooth driving!</p><p>However, verbal airtime is just one dimension of a discussion, and likely only a coarse indicator of a healthy team and a strong discussion facilitator. Diving into the actual content of the conversation or exploring other communication modes such as e-mail could in the future provide a more fine-grained understanding of how teams with well-balanced airtime operate, and how to encourage that behavior.</p><h3>About Cultivate</h3><p>Cultivate’s AI-powered platform enables engagement and inclusion by helping you understand and improve your workplace communications.</p><p>For more information on what we are doing at Cultivate, check out <a href="http://cultivateai.com/">our website</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*5ZGjU4Xl47UCsp7WyoKzJg.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7970cd183dd4" width="1" height="1" alt=""><hr><p><a href="https://medium.com/cultivate-ai/cultivating-equal-airtime-7970cd183dd4">Cultivating Equal Airtime</a> was originally published in <a href="https://medium.com/cultivate-ai">Cultivate AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Unpacking the Role of Gender(-ed speech) in Tech]]></title>
            <link>https://medium.com/cultivate-ai/unpacking-the-role-of-gender-ed-speech-in-tech-35b51627dcfc?source=rss-4b41ded9a6f9------2</link>
            <guid isPermaLink="false">https://medium.com/p/35b51627dcfc</guid>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[venture-capital]]></category>
            <category><![CDATA[pitching]]></category>
            <category><![CDATA[women-in-tech]]></category>
            <category><![CDATA[gender]]></category>
            <dc:creator><![CDATA[Grace Woo]]></dc:creator>
            <pubDate>Mon, 26 Feb 2018 17:16:52 GMT</pubDate>
            <atom:updated>2018-02-26T17:16:52.771Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uNm3BxSmmd6Comaq1zLJ_w.jpeg" /></figure><p>The Super Bowl was great and with only the blink of an eye, the 2018 Winter Olympics are now under way. It has been four years since I was invited to write for the ACM (American Computing Machinery) to share thoughts for <a href="https://dl.acm.org/citation.cfm?id=2634549.2604999&amp;coll=portal&amp;dl=ACM">how female computer scientists can lead in the work place</a>. I am intrigued by the most recent conversations in the tech world and even more specifically in Silicon Valley surrounding the role of women in the tech sphere. When <a href="https://cultivateai.com/">Cultivate</a> suggested that I use my data scientist role to share some recent insights on the topic, I gathered the tools required and got in sync with some of the team’s ongoing activities.</p><p>Cultivate is able to take an organization’s digital data, tease out important social dynamics, and reflect back to participating stakeholders valuable information about organizational engagement, inclusion, and wellness. In particular, understanding women’s responsibilities in the work environment is an essential issue in fostering a healthy workplace that is both functional and trustworthy.</p><p>Inspired by Cultivate’s technology and prior analysis revealing differences between male and female communication patterns, we’ll take a closer look at recent TechCrunch pitches under the same lens.</p><h3>Talking Gender in the VC world</h3><p>In 2017, the <a href="http://amj.aom.org/content/early/2017/04/29/amj.2016.1215.abstract">Academy of Management Journal published the result of a carefully designed experiment from Columbia University</a> revisiting at an expert level the fact that women have different speech patterns than men, and that these speech patterns might actually impact who succeeds professionally. Specifically, the study forms a theory around “Promotion” (focused on hopes, achievements, advancement, and ideals) vs. “Prevention” (concerned with safety, responsibility, security, and vigilance) questions and answers.</p><p>Analyzing Q&amp;A data from TechCrunch Disrupt New York between 2010 and 2016, the study asks blind participants to label questions and answers between entrepreneurs and venture capitalists as either “Promotion” or “Prevention”. The authors then go on to show how women are ultimately at a disadvantage when they provide “Prevention”-natured answers rather than “Promotion”-nature ones. This difference in behavior yields a significant difference in funding pattern:</p><p>“TechCrunch Disrupt entrepreneurs who were asked mostly prevention questions but gave mostly promotion responses went on to raise an average of $7.9 million in total funding. Conversely, those who responded to mostly promotion questions with mostly prevention answers went on to raise an average of only $563,000. An entrepreneur who is asked to defend her startups market share would be better served by framing her response around the size and growth potential of the overall pie than by merely stating how these plans to protect her share of the pie.”</p><h3>“How Women Undermine Themselves With Words”</h3><p>Women speak differently than men. <a href="http://tracymarchini.com/2010/09/30/differences-in-male-and-female-speech/">Tracy Marchini</a> and <a href="https://www.theguardian.com/world/2007/oct/01/gender.books">Deborah Cameron</a> describe how women tend to use more qualifiers and tend to converse in a more exploratory manner. An article by Goop (a women’s lifestyle publication) suggests common words used by women that might cause them to come across as less confident or competent in the workplace: “<a href="https://goop.com/work/career/how-women-undermine-themselves-with-words/">How Women Undermine Themselves With Words</a>”:</p><ul><li><strong><em>just</em></strong> — makes one come across as apologetic</li><li><strong><em>actually </em></strong>— makes one appear surprised that they have a question or objection</li><li>qualifiers (such as “I’m no expert in this <strong><em>but</em></strong> …” and “I know you have been researching this for a long time <strong><em>but</em></strong> …”) — undermines one’s position before they’ve stated it</li></ul><p>Wanting to understand if there was any support for these kinds of gendered speech patterns in an entrepreneurial setting, Cultivate analyzed four pitch transcripts from TechCrunch Disrupt SF 2017 using a combination of <strong>autosub</strong>, a transcription tool by <a href="https://github.com/agermanidis/autosub">Ager Manidis</a>, and common shell scripting tools.</p><p>We analyzed and grouped the transcripts into two sets based on the gender of those leading the pitch (two male-led pitches and two female-led pitches), which we’ve anonymized as <strong>SetA</strong> and <strong>SetB</strong>. Then, using shell scripting tools <a href="https://en.wikipedia.org/wiki/Wc_(Unix)"><strong>wc</strong></a>, <a href="https://en.wikipedia.org/wiki/Grep"><strong>grep</strong></a> and <a href="https://en.wikipedia.org/wiki/Gnuplot"><strong>gnuplot</strong></a>, Cultivate conducted a simple word analysis to see whether or not it was true that the women who pitched for <a href="https://www.youtube.com/watch?v=ZC3xl_UHhWY&amp;t=12s">Oneva</a> and <a href="https://www.youtube.com/watch?v=FVHxe7vKlR8&amp;t=131s">Future Family</a> actually had a different distribution of commonly used words than the men who pitched for <a href="https://www.youtube.com/watch?v=TnJYxAI3HrQ">ColorMass</a> and <a href="https://www.youtube.com/watch?v=UFvbygXm1Hc&amp;t=9s">Pi</a>.</p><p>Our initial analysis (using shell commands <strong><em>cat SetA.count | grep just </em></strong>and <strong><em>cat SetB.count | grep just</em></strong>, respectively)<em> </em>yielded <strong><em>14 </em></strong>occurrences of the word <strong><em>just</em></strong><em> </em>for SetA and <strong><em>16 </em></strong>occurrences of the word <strong><em>just </em></strong>for<strong><em> </em></strong>SetB<em>.</em></p><p>Continuing to reference the Goop article, we went on to check for occurrences of the words “actually” and “but”: finding <strong><em>9</em></strong> occurrences of the word <strong><em>actually </em></strong>for SetA and <strong><em>13 </em></strong>occurrences of the word <strong><em>actually </em></strong>for SetB<em>. </em>Likewise, our analysis yielded <strong><em>27 </em></strong>occurrences of the word <strong><em>but </em></strong>for SetA and <strong><em>25 </em></strong>occurrences of the word <strong><em>but </em></strong>for SetB.</p><p>To better contextualize these initial numbers, we used <a href="https://en.wikipedia.org/wiki/Gnuplot">gnuplot</a> to figure out the actual distribution of commonly used words between the sets:</p><p><strong>Most frequently used words in SetA, in order of frequency: (</strong>214 occurrences of the word <strong>the,</strong> 176 <strong>and,</strong> 165 <strong>to, </strong>138 <strong>a,</strong> 121 <strong>we,</strong> 117 <strong>of,</strong> 116 <strong>that,</strong> 111 <strong>you,</strong> 88 <strong>in,</strong> 69 <strong>so,</strong> 67 <strong>is,</strong> 59 <strong>are,</strong> 58 <strong>i,</strong> 58 <strong>for,</strong> 54 <strong>your,</strong> 50 <strong>on,</strong> 42 <strong>it,</strong> 42 <strong>have,</strong> 40 <strong>with,</strong> 40 <strong>our,</strong> 38 <strong>can,</strong> 36 <strong>what,</strong> 34 <strong>this,</strong> 30 <strong>as,</strong> 29 <strong>really</strong>)</p><p>Visualized as a vertical bar graph below:</p><p>**<br> **<br> **<br> **<br> ** **<br> ** ****<br> ** ****<br> ** ****<br> ** **** **<br> ** **** **<br> ** **** ** ** ** **<br> ** **** ** ** ** ****<br> ** **** ** ** ** ****<br> ** **** ** ** ** **** **<br> ** **** ** ** ** **** **<br> ** **** ** ** ** **** ** ** **<br> ** **** ** ** ** **** ** ** ** ** * ** **<br> ** **** ** ** ** **** ** ** ** ** * ** ** **<br> ** **** ** ** ** **** ** ** ** ** * ** ** ** ** * ** ** ** ** **<br> ** **** ** ** ** **** ** ** ** ** * ** ** ** ** * ** ** ** ** ****</p><p><strong>Most frequently used words in SetB, in order of frequency: ( </strong>212 occurrences of the word <strong>the,</strong> 136 <strong>to,</strong> 127 <strong>you,</strong> 107 <strong>and,</strong> 107 <strong>a,</strong> 104 <strong>of,</strong> 96 <strong>that,</strong> 82 <strong>we,</strong> 63 <strong>so,</strong> 62 <strong>in,</strong> 54 <strong>have,</strong> 53 <strong>it,</strong> 51 <strong>can,</strong> 50 <strong>is,</strong> 43 <strong>are,</strong> 40 <strong>for,</strong> 37 <strong>i,</strong> 34 <strong>what,</strong> 34 <strong>they,</strong> 33 <strong>this,</strong> 31 <strong>your,</strong> 31 <strong>now,</strong> 30 <strong>on,</strong> 29 <strong>how,</strong> 29 <strong>be</strong>)</p><p>Visualized as a vertical bar graph below:</p><p>**<br> **<br> **<br> **<br> **<br> **<br> **<br> **<br> ** **<br> ** ****<br> ** ****<br> ** **** ** ** **<br> ** **** ** ** ** **<br> ** **** ** ** ** ****<br> ** **** ** ** ** ****<br> ** **** ** ** ** **** **<br> ** **** ** ** ** **** ** ** **<br> ** **** ** ** ** **** ** ** ** ** * **<br> ** **** ** ** ** **** ** ** ** ** * ** ** ** ** * **<br> ** **** ** ** ** **** ** ** ** ** * ** ** ** ** * ** ** ** ** **</p><p>From this analysis, we found that there was no significant difference amongst the occurrences of the words <strong><em>just</em></strong>, <strong><em>actually</em></strong> and <strong><em>but </em></strong>between the two pitch sets. Rather, there was an apparent disparity in occurrences of the words <strong><em>we</em></strong> and <strong><em>you. </em></strong>Our distribution shows that SetA tends to use <strong>we </strong>(121 occurrences) more than <strong>you</strong> (88 occurrences) whereas SetB tends to use <strong>you </strong>(127 occurrences) more than <strong>we </strong>(82 occurrences).</p><p>Using this data, can you infer which set contains the female-led pitches and which contains the male-led pitches? Skip past these YouTube tips to find out.</p><h3>Transcribing and Analyzing Text with YouTube</h3><p>Interested in conducting your own text analysis with YouTube? Here are some of the tools and tricks we gathered while writing this post:</p><p><strong>YouTube Tip #1: Using YouTube’s built-in transcription tool</strong></p><p>To watch and select specific time segments within these videos, use the transcription feature built into YouTube.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*95d1lVhgBFDAOD08HnV7CQ.png" /></figure><p>YouTube has a built-in transcription tool that operates while the video is playing. In order to access it, click on the three dots (…) at the bottom right of the video. Then click on “open transcript” to see the scrolling content on the right. As a side note, I did a quick search through the transcripts to see whether words like “like” and “um” showed up more often in pitches featuring women at the forefront. I actually could not discern that these words showed up more often in these polished pitches.</p><p><strong>YouTube Tip #2: Analyzing YouTube videos external to YouTube</strong></p><p>I used Linux open source tools to download the video and run it through additional analysis. Alternatively, there are tools on the web, such as <a href="http://y2mate.com/">this</a> one where you can download the .mp4 file. Once the video is downloaded, I prefer the use of <a href="https://en.wikipedia.org/wiki/VLC_media_player">vlc</a> to play the videos back with closed captioning information. On YouTube, the files are saved as <a href="https://developer.mozilla.org/en-US/docs/Web/API/WebVTT_API">.vtt </a>files but standard <a href="https://fileinfo.com/extension/srt">.srt</a> files can also be used to playback txt.</p><p><strong>YouTube Tip #3: Watching your video slowed down or sped up</strong></p><p>Click on the gear icon aka the settings icon in the toolbar. In that menu, you should be able to see the third row with the option “speed”. By default, this is set to normal, but you can gear this down by a factor of 4 or up by a factor of 2 to watch content. This is useful when transcribing or attempting to analyze a phrase that isn’t immediately clear for content.</p><p>Shameless plug: you can watch <a href="http://cultivateai.com/">Cultivate</a>’s interview at TechCrunch Disrupt SF 2017 on YouTube <a href="https://www.youtube.com/watch?v=nMaw1bAW14w">here</a> :).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*L1Gmvb26FnkALH701N0zqw.png" /></figure><h3>The Proverbial “You”</h3><p><strong>… and now for the gendered breakdown of the pitch sets according to our simple word analysis at Cultivate:</strong></p><p>The words in SetA were comprised of the transcripts from <a href="https://www.youtube.com/watch?v=ZC3xl_UHhWY">Oneva</a> and <a href="https://www.youtube.com/watch?v=FVHxe7vKlR8">Future Family</a> (female-led pitches). The words in SetB were comprised of the transcripts from <a href="https://www.youtube.com/watch?v=TnJYxAI3HrQ">ColorMass</a> and <a href="https://www.youtube.com/watch?v=UFvbygXm1Hc">Pi</a> (male-led pitches).</p><p>Thus, our results show that the female-led pitches tended to use the word <strong>we</strong> more frequently than the word <strong>you</strong> (about 37% more frequently, in fact) and the male-led pitches tended to use the word <strong>you</strong> more frequently than the word <strong>we </strong>(about 55% more frequently).</p><p>Although these results are not conclusive, they do suggest that speech patterns amongst male and female entrepreneurs might differ not only on in how they talk about themselves and their company (as suggested by Goop and the Colombia study), but also <strong>who</strong> they talk about when they do so.</p><h3>About Cultivate</h3><p>Cultivate’s AI-powered platform enables engagement and inclusion by helping you understand and improve your workplace communications.</p><p>For more information on what we are doing at Cultivate, check out <a href="http://cultivateai.com/">our website</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dnllJfc-5N9iq4fBXSl5sw.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=35b51627dcfc" width="1" height="1" alt=""><hr><p><a href="https://medium.com/cultivate-ai/unpacking-the-role-of-gender-ed-speech-in-tech-35b51627dcfc">Unpacking the Role of Gender(-ed speech) in Tech</a> was originally published in <a href="https://medium.com/cultivate-ai">Cultivate AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>