Following is a brief summary of my major projects:
The Center for Speech and Language Technologies (CSaLT) was founded in May 2015 at the Information Technology University, Punjab with an aim to contribute towards Speech and Language Technologies research for languages of developing countries and to employ these technologies to assist in socio-economic development of low-literate and non-affluent people through better information access.
In 2012, Raza created Polly along with Roni Rosenfeld (his doctoral thesis advisor at Carnegie Mellon) and Umar Saif (his thesis committee member and research collaboration partner). Polly is a voice-based, telephone-based service that engages low-literate and non-tech savvy users in light entertainment and spreads useful development-related information to them as they become more comfortable with the voice interface. Polly allows users to record a short-message their voice, to modify it using a choice of funny voice modifications and to optionally send the original or modified version of the message to their friends. Consequently its use spreads virally among the target population and results in spreading development-related information to low-literate masses. The simple entertainment acts as a soft incentive for users to train themselves and overcomes the scalability hurdle of explicit user training and motivation. It also allows Polly to organically spread among the population through word of mouth as well as scheduled message deliveries from one user to another. As users become more comfortable with the interface, Polly introduces them to development related services like job search, health information etc.
Seeded with 5 users, within a year Polly amassed 165,000 users from all over Pakistan who took part in more than 636,000 calls, including 200,199 forwarded voice messages. These users were also exposed to a Job-Audio Browser where they could listen to newspaper job ads, fit for low-skilled workers, and forward these ads to friends. At its peak Polly was spreading to 1,000 new users every day. The 728 job ads in the system were listened 386,000 times by 34,000 users. Polly was used primarily by low-educated young men for entertainment and other creative uses like voicemail, group messaging and telemarketing. Its viral spread crossed gender and age boundaries and attracted a large number of blind users. Its use remained primarily in low-socio-economic strata.
Following is a brief description of the various deployments of Polly since 2011:
Pilot Launch in Pakistan (2011)
We tested a pilot version of Polly in 2011 by giving its phone number to 32 low-literate office workers (the phone number was handed out with minimal to no explanation). In 3 weeks it organically amassed over 2,000 users who took part in more than 10,000 calls. We had to shut down the pilot because the growing user population overwhelmed our single telephone line.
Large-scale deployment in Pakistan (2012)
In May 2012 we procured 30 phone lines and re-launched Polly in Pakistan. In one week we saturated these 30 lines and had to impose daily usage quotas on Polly's users. The most important addition in May 2012 was the introduction of a development-related application (what we call 'the payload') as part of the dialog menu: A Job Audio Browser. We scanned Pakistani newspapers daily for advertisement about jobs that are appropriate for low-skilled, low-literate workers, recorded them in the local language, and made them available for audio-browsing by phone. A caller can browse job opportunities, and can even forward a promising one to a friend.
This deployment of Polly remained online in Pakistan for a full year. It was introduced through automated phone calls to 5 people. Within a year Polly amassed 165,000 users and resulted in over 636,000 interactions, including 200,199 forwarded voice messages and 22,104 forwarded job ads. At its peak it was spreading to 1,000 new users every day. The 728 job ads were listened 386,000 times by 34,000 users. Polly was used primarily by low-educated young men for entertainment and other creative uses like voicemail, group messaging and telemarketing. Its viral spread crossed gender and age boundaries, attracted a lot of blind users but remained primarily in the same socio-economic strata.
First Launch in India: Collaboration with Babajob.com (2013)
In July 2013, we teamed up with Babajob.com to launch Polly in Bangalore, India. Babajob.com is an entry level and informal job portal in India that maintains a database of tens of thousands of low-skill job opportunities in Bangalore area. We launched Polly in India to find out if it can become viral in a different country and also to be able to measure the true impact of the back-end service through Babajob. Finding out the necessary conditions for making Polly viral in India took some time but finally we succeeded in achieving not only virality but also exponential spread.
Remotely deployed in India, Polly did not take off immediately as it did in Pakistan. Instead, it initially entered a six-month long phase of fluctuating, intermittent activity. We experimented with various forms of seeding and it eventually transitioned into a viral phase, with sustained transmission that continued for five months but without (exponential) growth. Finally, interface adjustments in response to user feedback enabling plain-voice asynchronous voice-messaging resulted in an abrupt exponential and viral growth amassing 10,349 phone calls by 1,613 users over a span of seven days. Of these, 299 users also transitioned to the job service. User feedback and surveys suggest possible reasons for each phase. We studied the challenges of remote deployment and the interplay of user interface; language of the system; seeding mechanisms and active response to user feedback towards the uptake of the service.
Second Launch in India: Collaboration with Jharkhand Mobile Vaani (2014)
In May 2014, we launched Polly in India, in the Jharkand area, in collaboration with a popular local citizen-radio-over-phone platform, Jharkhand Mobile Vaani (JMV). Polly was launched in Jharkhand based on a "cross-selling" dissemination model where Polly advertised JMV and vice versa. As JMV is a utility-based service which is not inherently viral, our goal was to popularize it through Polly's exponentially growing users once it seeds Polly by advertising to its current user-base. Polly remained active for 54 days in which it received 19,042 calls from 4,428 users. Analysis of traffic revealed viral spread but no exponential growth. Polly was shut down due to high international calling costs. Polly's launch in Jharkhand showed that seeding via promos and advertisements has the potential to induce viral spread. The volume sustained itself even after the promo on JMV was turned off. Users' tendency to introduce new people to Polly in response to different types of promos shows that the content, mood and tone of the promo plays a vital role in influencing user's understanding of the service and its capabilities, and in turn effects user behavior.
Ebola Healthline: Polly in West Africa (2014-15)
The 2014 outbreak of Ebola in several West African countries threatened to engulf an even larger population if effective measures were not taken to prevent its spread. Ebola spreads primarily through blood and bodily fluids of the infected person. Its rapid spread was largely being attributed to careless local practices that bring people in contact with Ebola patients and the deceased; lack of reliable information being conveyed to the public; corruption and mismanagement on part of the governments and a sheer absence of trust. In this situation an information dissemination service like Polly promised to prove useful.
In late 2014, we launched Polly in Guinea in collaboration with the US embassy in Conakry to spread reliable health information about Ebola. The variant of Polly in Guinea was called Polly-Sante' (Polly-Health) and it spread audio health messages about prevention, symptoms, cure and aftermath of Ebola in eleven local languages. These messages were generated by healthcare organizations that were intervening on the ground (like CDC). Polly is still live in West Africa and has reached 7,600 users.
Hello-Rozgar: A Job Portal (2015-16)
In 2015, Polly was relauched in Pakistan and with a new service: Hello-Rozgar. Hello-Rozgar is a telephone-based portal for access to employment and skill-training opportunities. Employers, employees, trainers and trainees can register with Hello-Rozgar through its web or IVR interface. Registered users can post audio job ads; advertise skill training opportunities; listen and apply for job openings and vocational training vacancies.
The project's aim was to benefit people from under-served portions of the society (women, minorities, low-literate and/or belonging to geographically remote areas), by disseminating information to them about skill-training and employment opportunities; and to further benefit a subset of them via financial incentives in the form of vouchers for appropriate skill training programs.
Users interact with Hello Rozgar via missed calls i.e. they flash the phone number of Hello Rozgar and within a few minutes Hello Rozgar calls them back (hence subsidizing their air time cost). During the interaction it allows users to register themselves as either someone who is looking for a job (a potential employee), or as a person who wants to advertise job openings (a potential employer), or as a person looking for skill training opportunities (trainee). Users register with basic details like name, national ID card number, father's name, gender, age, profession/skills of interest, district and tehsil. Once registered, Hello Rozgar matches potential employers and employees based on profession and location. It then allows potential employees to apply against job vacancies. The details and of these applicants are then provided to the potential employers who can now contact them directly. In case of skill training opportunities, Hello Rozgar maintains a list of vocational training institutes run by TEVTA and PVTC throughout Pakistan and provides contact details of the relevant and closest training institutes to the users interested in skill training.
Over the last few months Hello-Rozgar has been called 63,000 times by more than 49,000 users. Currently, there are 2,600 registered employees, 407 registered trainees and 180 registered employers. Polly and Hello Rozgar are both currently live in Pakistan.
Baang is a voice-based, telephone-based forum for low-literate users. Users can post voice messages (baangs) of their own as well as listen to baangs posted by others; vote these up or down; mark posts as being inappropriate and post voice comments on them. Baangs can be browsed by popularity, recency or a mixture of the two (i.e. trending posts). Within 71 days of its launch Baang received 42,500 calls (involving 1,550 users) and accumulated 8,469 posts contributed by 888 users that were played 393,448 times (by 1,524 users). Overall, we gathered 172 hours of spontaneous speech data (106 hours of posts + 66 hours of comments) with an average recording duration of 45 seconds. Users also engaged deeply with the advanced functionality of baang resulting in 29,000 up votes (by 1,100 users), 14,000 down votes (by 730 users), 7,700 report abuse votes (by 600 users) and 11,000 comments (by 600 users). The posts were forwarded 2,083 times (by 191 users to 517 users).
Almost all recording contain actual content and most (94%) are civilized and free from foul language. Among the most popular baangs are religious poetry (naat, hamd etc.), quranic recitation, songs, jokes, famous quotes and users hosting programs as DJs. People generally tend to vote down and report abusive content and enthusiastically compete for votes. Baang also found a lot of uptake among blind users (38% in a randomly chosen sample of 148 users). Content posted on baang is regularly monitored by moderators to prevent abuse and misuse.
We intend to use Baang for information dissemination to low-literate and visually impaired masses, mining public sentiment, gathering linguistic resources of local languages of Pakistan and behavior change studies.
Sawaal: Polly Quiz (2015-16)
Sawaal is a voice-based, telephone-based forum for sharing general knowledge questions and competing with friends. Sawaal allows a user to attempts questions posted by the administrators as well as general public. Users can also post questions of their own. Each question requires multiple answer choices (3-5) and the one of these choices needs to be marked as the correct answer. Users can also forward questions to friends and challenge them to competitions. Within a month Sawaal accumulated 2,700 questions posted by 160 users. The available questions were attempted 74,000 times by 400 users and forwarded 6,800 times to friends. Most of the questions fall into the category of sports, religion and general knowledge. Posted content is closely monitored for signs of misuse.
We plan to use Sawaal to disseminate information and understand the effects of increased user engagement on behavior change.