Perlo is an Oslo-based B2B startup building a smart risk prediction assistant that helps project managers in property development actively prevent cost overruns

Perlo

Despite my affinity for new technologies, using instant messaging at work has always struck me as counterproductive. Whether it’s Slack, Microsoft Teams, or similar solutions, the apparent benefits of these tools seem to come at the expense of overall . productivity These apps, combined with our dependence on smartphones, have redefined the rules of communication at work. It’s almost impossible to get away from them since they follow us everywhere, comfortably tucked in our pockets. The result: constant connectivity has become the default work culture. Employees are expected to read and respond to messages immediately or be penalized. Yet, the culture of connectivity is one of the main causes of distraction in the workplace. As far as I’m concerned, constant connectivity is counterproductive, and I will prove it to you. When connectivity becomes counterproductive People working in professional services (consultants, financial services, accountants, lawyers, etc.) are particularly affected by the culture of constant connectivity. It’s commonplace in these circles to think that being constantly available to the client is proof of professionalism and quality of service. Yet the relationship between constant connectivity and quality of work has never been proven. To verify this, two American researchers, Leslie A. Perlow and Jessica L. Porter, , one of the world’s most prestigious consulting firms. spent four years studying the behavior of professionals at the Boston Consulting Group (BCG) Before beginning the study, the researchers surveyed over 1,000 consulting professionals to better understand their work habits. 94% claimed to work more than 50 hours a week, and almost half more than 65 hours a week. This does not include the 20 to 25 hours spent checking email outside of work (at the time of the study, mainly on their BlackBerrys, but we can assume that this phenomenon persists, if not worsens with the use of smartphones). In their experiment at BCG, the researchers imposed a radical method by asking a small group of consultants to take an entire day off in the middle of their working week. During this day, they were forbidden to work and to read their emails or any other form of professional communication. A second group was similarly disconnected, but only after 6 p.m., once a week. This test aimed to measure the effects of total disconnection on the participants’ quality of work and well-being. The experiment was initially greeted with resistance, as most participants were afraid of disappointing their clients or reducing their chances of promotion. Measuring the quality of a consultant’s work can be very difficult, and although methods exist, their performance is often based on the number of hours worked. Presenteeism wins the day. With this in mind, the researchers were able to launch their test and force some of the staff to take a day off in the middle of the week. At the start of the experiment, participants were asked to rate the following statement: “ ” (rated from 1 for “ ” to 7 for “ ”). I feel respected for setting boundaries. strongly disagree strongly agree In the first month, the team responded with an average score of 3.7. After five months, this score rose to 5.2. The study also showed that all participants in this experiment wished to keep a portion of their time totally disconnected. Even more astonishing, 76% of those who didn’t participate in the study also wanted to be included in the experiment after hearing about it. Participants also reported greater job satisfaction, a greater chance of establishing a long-term career with the company, and a better work-life balance. Notably, participants also reported delivering better quality work to their customers. This experiment shows us that, despite the culture established in this group, its usefulness is unfounded. However, without questioning it, it seems to persist and even worsen with the new tools available. Why does constant connectivity persist? The answer may come as a surprise, but it’s pretty clear: it’s because it’s easy. Being constantly connected and reactive is easier than trying to create barriers to distraction. Being reactive is easier than planning long-term tasks. Letting yourself be disturbed is simpler than refusing to be interrupted. In short, the culture of connectivity benefits from our laziness. Cal Newport, author of , calls this phenomenon the “Principle of Least Resistance” and defines it as follows: Deep Work “ In a business setting, without clear feedback on the impact of various behaviors to the bottom line, we will tend toward behaviors that are easiest in the moment.” The Principle of Least Resistance: Two natural forces drive the Least Resistance Principle: If you can get an immediate response to a problem, work becomes “easier.” The need to plan disappears. We then indulge in a reactivity that takes the weight off our shoulders, to the detriment of the group’s concentration. Personal need: In an environment where constant connectivity is accepted, even encouraged, it’s easier to keep up with the general flow. Instead of developing personal organization skills, it’s easier to react to the next email. The fact that performance is indexed to reactivity encourages the preservation of this behavior. Performance measurement: It’s easier, for example, to forward an email to a whole team in the hope of receiving a constructive opinion than to reflect on the content of the email and make a personal analysis. The result: context-free emails asking only “What do you think?” abound in our inboxes (and distract us from more important tasks). The Least Resistance Principle is the primary source of our addiction to constant connectivity. Basically, being connected all the time is lazy. Instead of trying to structure our time, we prefer to let ourselves be carried along by the current of communication. We end up assessing our worth by the number of emails we read and process per day rather than by the tasks we actually complete. And our managers do the same because they are themselves carried by the Principle of Least Resistance. It’s possible to extricate ourselves from this negative connectivity, but it takes effort to set up rules that are beneficial in the long term. How do we escape constant connectivity? The method below aims to set limits on the amount of time you spend on communication so that you can regain control of your time and concentration: 1. Set strict, non-negotiable limits Start by choosing a time slot reserved for rest, during which all professional communication will be prohibited. For example, disconnect completely after 6 p.m. and during weekends. This first step aims to give you time to recharge your attention daily (using can also help). a closing ritual at the end of the day 2. Communicate transparently Your limits must be clearly communicated to your professional entourage. Talk about it with those around you, whether your superior or your customers. Explain that your work's result will be better if you can recharge your batteries daily. Everyone can understand this (and often, even envy you). 3. Define blocks of full concentration Connectivity is also a problem during the day, and these distractions prevent you from completing the most important tasks. To concentrate fully, define times when all forms of communication are forbidden. If you work in an office and are constantly disturbed, use headphones and explain that you are not to be disturbed when you wear them. The goal here is to reduce the constant distractions so that you can rediscover deep concentration. 4. Compartmentalize your schedule Once you’ve established the basics, organizing your time into specific compartments will be easier. For example, emails (twice a day is sufficient). Likewise, define . define specific times for reading and replying to time slots during which you will accept meetings The rest of the time should be devoted to deep, concentrated work. Inserting scheduled breaks into your day to will help you recharge your batteries. Tip: get some fresh air and take a walk As we saw with the BCG example, constant connectivity persists because it is almost impossible to demonstrate that its opposite is more effective. In a world where the Principle of Least Resistance reigns, alternatives to constant connectivity are often ignored. Challenging this paradigm takes courage. It requires reflection on the true meaning and value of work. If being constantly reactive doesn’t suit you, it’s time to . try another method Also published . here

Disconnect to Reconnect: The Counterproductive Culture of Constant Connectivity

THE EVENING SKY AT THE VERNAL EQUINOX

Why Python is The 1st Language

A blog post on the call and an audio replay can be found on Medium . here On June 29th, in partnership with Geektime we held a conference call with four crypto currency thought leaders (from Tezos, Bancor, DCG, and Polychain Capital) and over 400 participants. You can read a blog post and listen to a replay of the call . In the weeks since the call, Etheruem and other crypto currencies fell 60%, and then rose 50% and Tezos raised a record shattering $232 million in it’s ICO. Per the graph below, through the fist half of 2017, ICO’s are tracking to raise $2.5 billion this year, up 10X over 2016. Calling this space dynamic is an obvious understatement. here Two things became apparent to us from the call. First, ICOs are a poised to be a disruptive force in the global tech and finance ecosystems (here’s a great blog post on “ ”). Second, a follow up call was needed to go in to greater detail about ICOs, from the perspective of both companies (who should consider an ICO, what are the keys to a successful ICO, what are the downsides to an ICO…), and from investors (how do you put different ICOs in to context, what’s the best way to buy and trade ICOs …). To answer these questions, and others, we’ve put together another great panel of thought leaders who have all been part of successful ICOs. , who will go in to greater detail about their ICO experiences and give greater context to the entire ICO process and token ecosystem. Why Tokens Are Eating The World The hour long call will leave 20 minutes at the end of for questions. Click here to register for the call at GoToWebinar Speaker Bios: an identity platform with the goal of bringing mobile & digital democracy to the world by utilizing the to give everyone a digital identity. Civic recently completed a $33m token sale after raising $5.75m in venture capital. Prior to Civic Vinny started Gyft, which became the leading consumer platform accepting before selling it to First Data in 2014 and joining them as SVP of Product Development. Vinny has founded a number of other companies, including Click2Customers, Yola & Newtown Partners (VC) Vinny Lingham — Co Founder and CEO at Civic , Blockchain Bitcoin , a an Israel based decentralized financial platform backed by a breakthrough protocol for creating, issuing and trading real financial assets. Adam Perlow —Co-Founder and CEO at Zen Protocol a digital finance-focused investment bank and asset manager for the emerging tokenized and crypto-currency capital markets. Stan has over 15 years of experience in investment banking, corporate finance, M&A, and equities as CEO of The Argon Group, Managing Director at Rosemount Consulting Partners, EVP for TNK-BP, and as an Executive Director at Morgan Stanley. Stan Miroshnik—Co-Founder & CEO of The Element Group , crypto asset management and investment company. SuperBloom actively trades crypto assets,builds trading algorithms and tech tools, and accelerates/launches new companies that are pre-token sales. She previously founded a a Y-Combinator backed ed-tech company that served tens of thousands of users and bootstrapped 2 edtech companies to multi million revenues. Emmie Change — CEO of Superbloom, a Click here to register for the call at GoToWebinar If you liked this post, click the💚 below so other people will see it on Medium!

Why ICOs Are Poised 2 Disrupt Everything Conference Call — July 27th, 2pm-3pm est

One of my heroes is Gary Vaneychuck. His book blew my mind and I just finished reading . This guy is a legend and I am glad he is coming to my hometown of Atlanta for the in March. I’m stoked to finally meet him and see some raw passion. Jab, Jab, Jab, Right Hook Crush It Rainmaker conference hosted by Salesloft For those who are not familiar with Gary, and here are his top 5 quotes that will get your blood pumping and out of your seats. Say these quotes out loud and you will know what I mean. These are truly powerful words which he says with so much passion that you gotta feel it. watch this video Get off your ass and get noticed. Look yourself in the mirror and ask yourself, what do I want to do every day for the rest of my life? Do that. You didn’t grow up driving. You figured it out. You can market your ass off, but if your product sucks, you’re dead. It’s easy to dream about it, but much harder to execute it. Work! I love the enthusiasm that Gary projects and the #HustleMuscle he has to do big things. As I finished reading something big happened. My company, , an account-based marketing firm based in Atlanta, hit $1 million in ARR, and we did it in just nine months! We are now in the works to make a 3x run (as they say in the world) to triple our revenue in 2016. Crush It , Terminus startup Here’s how we did it and the story of our #HustleMuscle. In any startup, it’s a given that you will work crazy hours pouring your life and soul into your company. You’ll wear more hats than you even knew you could wear. You love it and hate it at the same time. But no matter what, you have the greatest influence on the success or failure of your company, and that’s what keeps you going. Because you want to succeed, you’re thinking about your startup while you are drinking, eating, and even sleeping. This lifestyle now seems perfectly normal, but after the past year, it begged a question… Are you crushing it or crushing yourself building a startup? As I reflect on 2015, it was a year some great teamwork and lots of hustle. Heck, we helped build an , , and probably the most important accomplishment is hiring 24 new rockstars on the which really feels like a real family. ABM category with G2Crowd did three massive #FlipMyFunnel conferences with over 1,000 attendees Terminus team So, I tried my best to look at each and every team member of Terminus and came up with 10 reasons why I feel that we are simply #CrushingIt and thankfully not crushing ourselves in the process. 10 Ways to Know You ’re #CrushingIt Building a Startup , employee #3 at Terminus, ran events for the state government and wanted to switch careers. worked in financial accounting before coming to Terminus. They are now rocking our sales development team setting a record number of demos every week with big smiles on their faces as if they have been doing this for years. It’s about doing something you’ve never done before. Amanda Faillo Brad Wilkerson and on our Customer Success team won’t go home until they have done everything to make sure each and every customer is happy. These ladies rock. We call them “Team Kydney” because we can’t live without them. It’s about giving 2,000%. Sydney Smith Karen Topham , who left Salesforce Pardot to head up our sales development team, comes in before anyone else gets in the office to set an example, especially for our new hires and . It’s about coming in early. Lucas Ulloque Kirby Oscar Geoffrey Hanan on our marketing team is all about building a to help innovative B2B sales and marketing professionals learn everything they can about account-based marketing. I joke that her title should be Director of Funnel Flipping. It’s about building a community. Nikki Nixon #FlipMyFunnel community , our sales leader, is so focused on building a kickass team to grow revenue. She took a step back and realized she needed help managing both the sales account executives and SDRs, which is why Lucas came on board. Now, her awesome team of AEs including , , and are laser-focused on driving the company forward to hit our 3X revenue goal. It’s about taking a step back to take 10 steps forward. Tonni Bennett Stuart Maron Chris Reene Zach Patterson is helping me write the . We’re so excited will be out on April 18, and she’s already getting a plan together for building our content library for sales enablement in Q2. It’s about getting to the finish line then running another lap. Lauren Patrick first-ever book about ABM Account-Based Marketing for Dummies , our recruiter and culture evangelist, is finding new ways to make the onboarding process for new “Terminators” smoother and giving every employee an amazing first week and beyond. It’s about dedicating time to do the right things. Kipp Ramsey and on the engineering team started a club for our company! We have several people in our company who are reading the book, along with their spouses, and getting together each month. Talk about a #WorkFamily. It’s about building a great culture. Stevie Vines Brandon Cummings book Our interns — , , and — are so far beyond the stereotypical interns who just print pages and pickup lunch. They are doing real work helping our company grow and solving real problems at a fast-paced startup while shaping their futures forever. It’s about helping build careers. Josh Fowler Adam Perlow Jeff Ronpirin Terminus CEO and co-founder, , always knew he wanted to be an entrepreneur. Our CTO and co-founder, , knew he wanted to build an amazing product. And that’s just what we’re doing while challenge the status quo of the B2B marketing and sales industry. It’s about doing something you always wanted to do. Eric Spett Eric Vass I’m curious to know what other startup founders think about building a business from the ground up. In my next post, I’d love to share more about how to not crush yourself and focus having a great time working and living your dream at a startup. How’s your startup experience? What does #Crushingit mean to you? Originally posted on LinkedIn Pulse

Are You #CrushingIt or Crushing Yourself Building a Startup? A Story of Raw Passion

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Muhammed Yusuf Kocyigit, Boston University;
(2) Anietie Andy, University of Pennsylvania;
(3) Derry Wijaya, Boston University. Table of Links Abstract and Intro
Related Works
Data
Method
Analysis and Results
Conclusion
Limitations
Ethics Statement and References
Appendix: Toxicity Measurement
Appendix: Correlation Over Time
Appendix: Wikidata
Appendix: Hyperparameter Sensitivity Appendix: Wikidata We use the the query in Figure 8 to select the set of individuals that were born in, residents of and citizens of the United States of America. The query takes the ethnic label manually. The ethnic label returns classes that are much more fine-grained then we aim for in this study so we manually create a dictionary to map each sub-group into our main categories presented in Figure 4 as African American, White American and Others. For African American the main rule was that the origin country for the ethnicity would be in the African Continent. We have also classified each European American(for example Italian American, Irish American etc.) ethnicity into White Americans. Finally we manually label individuals that are are significant figures that don’t contain the ethnicity label in their Wikipedia page. This was more prevalent in White Americans. This paper is available on arxiv under CC 4.0 license. Authors: (1) Muhammed Yusuf Kocyigit, Boston University; (2) Anietie Andy, University of Pennsylvania; (3) Derry Wijaya, Boston University. This paper is available on arxiv under CC 4.0 license. Authors: Authors: (1) Muhammed Yusuf Kocyigit, Boston University; (2) Anietie Andy, University of Pennsylvania; (3) Derry Wijaya, Boston University. Table of Links Abstract and Intro Related Works Data Method Analysis and Results Conclusion Limitations Ethics Statement and References Appendix: Toxicity Measurement Appendix: Correlation Over Time Appendix: Wikidata Appendix: Hyperparameter Sensitivity Abstract and Intro Abstract and Intro Related Works Related Works Data Data Method Method Analysis and Results Analysis and Results Conclusion Conclusion Limitations Limitations Ethics Statement and References Ethics Statement and References Appendix: Toxicity Measurement Appendix: Toxicity Measurement Appendix: Correlation Over Time Appendix: Correlation Over Time Appendix: Wikidata Appendix: Wikidata Appendix: Hyperparameter Sensitivity Appendix: Hyperparameter Sensitivity Appendix: Wikidata We use the the query in Figure 8 to select the set of individuals that were born in, residents of and citizens of the United States of America. The query takes the ethnic label manually. The ethnic label returns classes that are much more fine-grained then we aim for in this study so we manually create a dictionary to map each sub-group into our main categories presented in Figure 4 as African American, White American and Others. For African American the main rule was that the origin country for the ethnicity would be in the African Continent. We have also classified each European American(for example Italian American, Irish American etc.) ethnicity into White Americans. Finally we manually label individuals that are are significant figures that don’t contain the ethnicity label in their Wikipedia page. This was more prevalent in White Americans.

A Novel Method for Analysing Racial Bias: Collection of Person Level References: Appendix: Wikidata

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Muhammed Yusuf Kocyigit, Boston University;
(2) Anietie Andy, University of Pennsylvania;
(3) Derry Wijaya, Boston University. Table of Links Abstract and Intro
Related Works
Data
Method
Analysis and Results
Conclusion
Limitations
Ethics Statement and References
Appendix: Toxicity Measurement
Appendix: Correlation Over Time
Appendix: Wikidata
Appendix: Hyperparameter Sensitivity Analysis and Results In this section, we present results that are obtained using the group embeddings trained with the details provided in the Method Section. We will go over each method of analysis and results here. Correlation over Time We first examine the temporal relationships within the context of each group. We calculate the Pearson correlation between each pair of decades for every group as illustrated in Figure 5. For example, the top-right cell in Figure 5 would be the Pearson correlation between the vectors for the African American group embeddings for 1870 and 1990. The objective of this is two-fold. First, we want to see how the context around each racial group has changed, which decades have exhibited stronger changes compared to others. Second, this we use this as a sanity check mechanism similar to Garg et al. (2018). Our findings suggest that the context encompassing White Americans exhibits greater stability over time with most decades correlating over 0.6 with any other decade. In contrast, the context pertaining to African Americans demonstrates significant discontinuities. This highlights the potential disparities in the evolution of social contexts surrounding these groups throughout history. Additionally we show that the way African Americans are represented in books has drastically changed at the turn of the 20th century. This change splits the time frame between 1870-1900 and 1900-2000 corresponding to the Civil War and Emancipation and the Segregation and Civil Rights periods (which we have marked over the heatmap) with a certain delay, which we approximate as 10 years. We also test the significance of the observed difference with the Kolmogorov–Smirnov two-sample test (more details in Appendix: Correlation Over Time) and show that there is a notable change in the period between 1900s-1910s in the African Americans embeddings, indicating that the important social events of the time as well as the civil rights movement had a strong impact on the representation of African Americans in books. Literature is naturally influenced by socio-political events but in this work we computationally measure the rate and extent of this influence and show the details of this change as well. Additionally this analysis serves as a verification of the embeddings. The alignment between the results our method has captured and well known historical events is emblematic of the method’s capacity to effectively capture and illuminate large-scale trends within literary works. Consequently, our method’s ability to reveal outcomes that resonate with our understanding of historical events is a strong signal that the collection of individual representations can be used to measure the representation of the collection of individuals. Semantic Axes Our second approach to analyzing the context involves the use of semantic axes, as introduced by Lucy, Tadimeti, and Bamman (2022). Semantic Axes, broadly are collections of adjectives that define a specific relationship. For example an axis could be the ”Noble” axis where one (left-hand) pole is defined by words like base, common, and ignoble and the other (right-hand) pole is defined by words like aristocratic, august, and blue. The semantic axis is the vector connecting the mean of the left-hand pole toward the mean of the righthand pole. Later any word vector can be compared to this semantic axis via cosine distance and analyzed in this direction i.e., nobility in this case. We employ the entire collection of 732 semantic axes presented by Lucy, Tadimeti, and Bamman (2022), and represent the poles of these axes with vectors derived from the pre-trained embeddings released by (Hamilton, Leskovec, and Jurafsky 2016). Following Lucy, Tadimeti, and Bamman (2022), we exclude any semantic axis that doesn’t contain at least three words in both poles. Utilizing the learned group representations, we compute the difference in the cosine similarity between the vector for African American and White American groups to the semantic axes in the same way with An, Kwak, and Ahn (2018). We use the cosine similarity between each group embedding and the semantic vector, then taking the absolute difference between these similarities. In Table 1, we display these results, with each row corresponding to a decade and showcasing the top-2 semantic axes that exhibit the highest absolute difference between the groups in that decade. The pole words exhibiting greater similarity to each group is placed under its respective column. In Table 1 we observe that the most significant differences between African Americans and White Americans are in semantic axes related to social status. Between 1850- 1990, African Americans were associated with terms such as ”lowly”, ”outlawed” and ”inferior” while White Americans were associated with terms such as ”grand”, ”legal” and ”eminent”. Additionally, if one of the poles of a semantic axis can be identified to have a negative meaning African Americans are consistently related to the negative pole of of that semantic axis with only a few exceptions. Toxicity Finally, we investigate the toxicity of the context and how it has evolved over time. To achieve this, we employ a dictionary-based approach using toxic words provided by HurtLex (Bassignana, Basile, and Patti 2018). However, as words can lose or acquire toxic meanings over time; to account for this, we introduce a novel filtering method. First, we identify semantic axes with the highest average cosine similarity between their poles and the list of toxic words for the latest decade for which we have embeddings, 1990-2000. This process yields a set of axes in which toxic words are predominantly clustered around one of the poles. We select the top 10 semantic axes, and for each decade, we recompute the vectors for these axes and the similarity of toxic words to these axes. We assume that the meaning of a word with respect to toxicity has significantly changed if it switches its side (from being closer to one pole to being closer to the opposite pole) in more than half of these semantic axes. This method is effective in eliminating words that might no longer be toxic; however, we do not search the entire vocabulary for additional toxic words. While extending our approach to include this aspect could be beneficial, it constitutes a separate research direction and is beyond the scope of this paper. Our initial hypothesis posited that toxic words would cluster around semantic poles such as ”bad” or ”wrong.” Nonetheless, our experiments revealed that the most extreme poles where toxic words cluster are regular ↔ irregular, educated ↔ ignorant, superior ↔ inferior, and normal ↔ abnormal, with toxic words clustering around the latter in each case. In the Appendix , we provide further details on the filtering of toxic words. Figure 7 shows that the content in books around African Americans is always more toxic compared to White Americans. While the toxicity has dropped after the Civil War, in 1860s and 1870s, this changes and there is an upsurge in toxicity after the 1900s which marks the beginning of the segregation era and continues increasing to the Civil Rights movement (1950s). Finally, in the most recent decades while toxicity has somewhat decreased, African Americans are still exposed to more toxic words and context. This paper is available on arxiv under CC 4.0 license. Authors: (1) Muhammed Yusuf Kocyigit, Boston University; (2) Anietie Andy, University of Pennsylvania; (3) Derry Wijaya, Boston University. This paper is available on arxiv under CC 4.0 license. Authors: Authors: (1) Muhammed Yusuf Kocyigit, Boston University; (2) Anietie Andy, University of Pennsylvania; (3) Derry Wijaya, Boston University. Table of Links Abstract and Intro Related Works Data Method Analysis and Results Conclusion Limitations Ethics Statement and References Appendix: Toxicity Measurement Appendix: Correlation Over Time Appendix: Wikidata Appendix: Hyperparameter Sensitivity Abstract and Intro Abstract and Intro Related Works Related Works Data Data Method Method Analysis and Results Analysis and Results Conclusion Conclusion Limitations Limitations Ethics Statement and References Ethics Statement and References Appendix: Toxicity Measurement Appendix: Toxicity Measurement Appendix: Correlation Over Time Appendix: Correlation Over Time Appendix: Wikidata Appendix: Wikidata Appendix: Hyperparameter Sensitivity Appendix: Hyperparameter Sensitivity Analysis and Results In this section, we present results that are obtained using the group embeddings trained with the details provided in the Method Section. We will go over each method of analysis and results here. Correlation over Time We first examine the temporal relationships within the context of each group. We calculate the Pearson correlation between each pair of decades for every group as illustrated in Figure 5. For example, the top-right cell in Figure 5 would be the Pearson correlation between the vectors for the African American group embeddings for 1870 and 1990. The objective of this is two-fold. First, we want to see how the context around each racial group has changed, which decades have exhibited stronger changes compared to others. Second, this we use this as a sanity check mechanism similar to Garg et al. (2018). Our findings suggest that the context encompassing White Americans exhibits greater stability over time with most decades correlating over 0.6 with any other decade. In contrast, the context pertaining to African Americans demonstrates significant discontinuities. This highlights the potential disparities in the evolution of social contexts surrounding these groups throughout history. Additionally we show that the way African Americans are represented in books has drastically changed at the turn of the 20th century. This change splits the time frame between 1870-1900 and 1900-2000 corresponding to the Civil War and Emancipation and the Segregation and Civil Rights periods (which we have marked over the heatmap) with a certain delay, which we approximate as 10 years. We also test the significance of the observed difference with the Kolmogorov–Smirnov two-sample test (more details in Appendix: Correlation Over Time) and show that there is a notable change in the period between 1900s-1910s in the African Americans embeddings, indicating that the important social events of the time as well as the civil rights movement had a strong impact on the representation of African Americans in books. Literature is naturally influenced by socio-political events but in this work we computationally measure the rate and extent of this influence and show the details of this change as well. Additionally this analysis serves as a verification of the embeddings. The alignment between the results our method has captured and well known historical events is emblematic of the method’s capacity to effectively capture and illuminate large-scale trends within literary works. Consequently, our method’s ability to reveal outcomes that resonate with our understanding of historical events is a strong signal that the collection of individual representations can be used to measure the representation of the collection of individuals. Semantic Axes Our second approach to analyzing the context involves the use of semantic axes, as introduced by Lucy, Tadimeti, and Bamman (2022). Semantic Axes, broadly are collections of adjectives that define a specific relationship. For example an axis could be the ”Noble” axis where one (left-hand) pole is defined by words like base, common, and ignoble and the other (right-hand) pole is defined by words like aristocratic, august, and blue. The semantic axis is the vector connecting the mean of the left-hand pole toward the mean of the righthand pole. Later any word vector can be compared to this semantic axis via cosine distance and analyzed in this direction i.e., nobility in this case. We employ the entire collection of 732 semantic axes presented by Lucy, Tadimeti, and Bamman (2022), and represent the poles of these axes with vectors derived from the pre-trained embeddings released by (Hamilton, Leskovec, and Jurafsky 2016). Following Lucy, Tadimeti, and Bamman (2022), we exclude any semantic axis that doesn’t contain at least three words in both poles. Utilizing the learned group representations, we compute the difference in the cosine similarity between the vector for African American and White American groups to the semantic axes in the same way with An, Kwak, and Ahn (2018). We use the cosine similarity between each group embedding and the semantic vector, then taking the absolute difference between these similarities. In Table 1, we display these results, with each row corresponding to a decade and showcasing the top-2 semantic axes that exhibit the highest absolute difference between the groups in that decade. The pole words exhibiting greater similarity to each group is placed under its respective column. In Table 1 we observe that the most significant differences between African Americans and White Americans are in semantic axes related to social status. Between 1850- 1990, African Americans were associated with terms such as ”lowly”, ”outlawed” and ”inferior” while White Americans were associated with terms such as ”grand”, ”legal” and ”eminent”. Additionally, if one of the poles of a semantic axis can be identified to have a negative meaning African Americans are consistently related to the negative pole of of that semantic axis with only a few exceptions. Toxicity Finally, we investigate the toxicity of the context and how it has evolved over time. To achieve this, we employ a dictionary-based approach using toxic words provided by HurtLex (Bassignana, Basile, and Patti 2018). However, as words can lose or acquire toxic meanings over time; to account for this, we introduce a novel filtering method. First, we identify semantic axes with the highest average cosine similarity between their poles and the list of toxic words for the latest decade for which we have embeddings, 1990-2000. This process yields a set of axes in which toxic words are predominantly clustered around one of the poles. We select the top 10 semantic axes, and for each decade, we recompute the vectors for these axes and the similarity of toxic words to these axes. We assume that the meaning of a word with respect to toxicity has significantly changed if it switches its side (from being closer to one pole to being closer to the opposite pole) in more than half of these semantic axes. This method is effective in eliminating words that might no longer be toxic; however, we do not search the entire vocabulary for additional toxic words. While extending our approach to include this aspect could be beneficial, it constitutes a separate research direction and is beyond the scope of this paper. Our initial hypothesis posited that toxic words would cluster around semantic poles such as ”bad” or ”wrong.” Nonetheless, our experiments revealed that the most extreme poles where toxic words cluster are regular ↔ irregular, educated ↔ ignorant, superior ↔ inferior, and normal ↔ abnormal, with toxic words clustering around the latter in each case. In the Appendix , we provide further details on the filtering of toxic words. Figure 7 shows that the content in books around African Americans is always more toxic compared to White Americans. While the toxicity has dropped after the Civil War, in 1860s and 1870s, this changes and there is an upsurge in toxicity after the 1900s which marks the beginning of the segregation era and continues increasing to the Civil Rights movement (1950s). Finally, in the most recent decades while toxicity has somewhat decreased, African Americans are still exposed to more toxic words and context.

A Novel Method for Analysing Racial Bias: Collection of Person Level References: Analysis and Result

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Muhammed Yusuf Kocyigit, Boston University;
(2) Anietie Andy, University of Pennsylvania;
(3) Derry Wijaya, Boston University. Table of Links Abstract and Intro
Related Works
Data
Method
Analysis and Results
Conclusion
Limitations
Ethics Statement and References
Appendix: Toxicity Measurement
Appendix: Correlation Over Time
Appendix: Wikidata
Appendix: Hyperparameter Sensitivity Limitations Reliability of Word Embeddings The representation power and how much word embeddings can model our beliefs is an active point of discussion. This line of research is critical because it allows differentiating under which conditions word representations generate good abstractions to analyze. The critique can be summarized in two main lines: structural and statistical problems. Structural problems (Arseniev-Koehler 2021) relate to the Distributional Hypothesis (Harris 1954) that current word embeddings are based on. This is criticized by structuralism in multiple aspects. Primarily issues arise when the phrase a word is known by the company it keeps is interpreted as a word is the company it keeps. The representations of words or groups that we present are not to say that the content represents them precisely as the representations, but they are more closely linked to certain concepts if they are closer in the semantic space. The second important issue we must be aware of is that concepts are not always dichotomous or binary. They can be graded or related in multiple ways. We, however, in line with An, Kwak, and Ahn (2018) assume a dichotomous nature to a set of adjectives. The statistical problems mainly stem from the frequency of different words/concepts. Valentini, Slezak, and Altszyler (2022) shows that word representations tend to yield higher similarity scores between high-frequency words than between high and low-frequency words. Joseph and Morgan (2020) show that word representations produce similarity measures that better represent human beliefs in with higher frequency words. Loon et al. (2022) argue that word embedding methods like word2vec cluster high-frequency words together and low-frequency words together. They argue that people have an intuitive positive bias where positive words are more frequent and if a racial/social group(black names in their case) is less frequently represented in the corpus compared to their control group(white names). They are just likely to be related to negative words and this relation would not necessarily be an indication of bias in the corpus. Structural problems are related to how we analyze the results and make inferences from them. However, in this paper, we target statistical problems on two levels. First, we look at the distribution of the context words in the primary data where the word vectors were trained and observe that the context words for these groups are not significantly different, and on the frequency of names, we sample equal amounts of (word, context) pairs for both African Americans and White Americans and train the embeddings for these groups from scratch rather than using a pre-trained embedding that would depend on the frequency of names in the initial corpus. One specific limitation that we had to work around in this project was the issue of false positive matches. Even though we search for exact name+surname combination in books the resulting n-grams are not necessarily always about the same person that we intent on. While a person with the same name+surname combination likely belongs to the same racial group in our analysis, this might not be the same for other use cases making this a relevant limitation to consider. Another relevant limitation is the data size limitation. We have replicated our results with a manually curated name subsets however a smaller subset of n-grams was not a viable option. We have also observed that decades before 1850s return too few samples and the results are not meaningful for those decades. The question of how many n-grams would be enough for a new application is still an open question that we were not able to answer. This question would also extend the scope of this paper beyond a reasonable limit. This paper is available on arxiv under CC 4.0 license. Authors: (1) Muhammed Yusuf Kocyigit, Boston University; (2) Anietie Andy, University of Pennsylvania; (3) Derry Wijaya, Boston University. This paper is available on arxiv under CC 4.0 license. Authors: Authors: (1) Muhammed Yusuf Kocyigit, Boston University; (2) Anietie Andy, University of Pennsylvania; (3) Derry Wijaya, Boston University. Table of Links Abstract and Intro Related Works Data Method Analysis and Results Conclusion Limitations Ethics Statement and References Appendix: Toxicity Measurement Appendix: Correlation Over Time Appendix: Wikidata Appendix: Hyperparameter Sensitivity Abstract and Intro Abstract and Intro Related Works Related Works Data Data Method Method Analysis and Results Analysis and Results Conclusion Conclusion Limitations Limitations Ethics Statement and References Ethics Statement and References Appendix: Toxicity Measurement Appendix: Toxicity Measurement Appendix: Correlation Over Time Appendix: Correlation Over Time Appendix: Wikidata Appendix: Wikidata Appendix: Hyperparameter Sensitivity Appendix: Hyperparameter Sensitivity Limitations Reliability of Word Embeddings The representation power and how much word embeddings can model our beliefs is an active point of discussion. This line of research is critical because it allows differentiating under which conditions word representations generate good abstractions to analyze. The critique can be summarized in two main lines: structural and statistical problems. Structural problems (Arseniev-Koehler 2021) relate to the Distributional Hypothesis (Harris 1954) that current word embeddings are based on. This is criticized by structuralism in multiple aspects. Primarily issues arise when the phrase a word is known by the company it keeps is interpreted as a word is the company it keeps. The representations of words or groups that we present are not to say that the content represents them precisely as the representations, but they are more closely linked to certain concepts if they are closer in the semantic space. The second important issue we must be aware of is that concepts are not always dichotomous or binary. They can be graded or related in multiple ways. We, however, in line with An, Kwak, and Ahn (2018) assume a dichotomous nature to a set of adjectives. The statistical problems mainly stem from the frequency of different words/concepts. Valentini, Slezak, and Altszyler (2022) shows that word representations tend to yield higher similarity scores between high-frequency words than between high and low-frequency words. Joseph and Morgan (2020) show that word representations produce similarity measures that better represent human beliefs in with higher frequency words. Loon et al. (2022) argue that word embedding methods like word2vec cluster high-frequency words together and low-frequency words together. They argue that people have an intuitive positive bias where positive words are more frequent and if a racial/social group(black names in their case) is less frequently represented in the corpus compared to their control group(white names). They are just likely to be related to negative words and this relation would not necessarily be an indication of bias in the corpus. Structural problems are related to how we analyze the results and make inferences from them. However, in this paper, we target statistical problems on two levels. First, we look at the distribution of the context words in the primary data where the word vectors were trained and observe that the context words for these groups are not significantly different, and on the frequency of names, we sample equal amounts of (word, context) pairs for both African Americans and White Americans and train the embeddings for these groups from scratch rather than using a pre-trained embedding that would depend on the frequency of names in the initial corpus. One specific limitation that we had to work around in this project was the issue of false positive matches. Even though we search for exact name+surname combination in books the resulting n-grams are not necessarily always about the same person that we intent on. While a person with the same name+surname combination likely belongs to the same racial group in our analysis, this might not be the same for other use cases making this a relevant limitation to consider. Another relevant limitation is the data size limitation. We have replicated our results with a manually curated name subsets however a smaller subset of n-grams was not a viable option. We have also observed that decades before 1850s return too few samples and the results are not meaningful for those decades. The question of how many n-grams would be enough for a new application is still an open question that we were not able to answer. This question would also extend the scope of this paper beyond a reasonable limit.

Perlo

perlo.io

-- employees

Since n.d.

COMPANY RANKING

PERLO

EVERGREEN INDEX #7894

Perlo WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!

Perlo

perlo.io

-- employees

Since n.d.

COMPANY RANKING

PERLO

EVERGREEN INDEX #7894

RELATED COMPANIESRANK

STORIES ABOUT PERLO

Perlo WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!