The USCIS is looking to add a new dimension to their adjudication process. Their Chief Technology Officer (CTO), Rob Brown, is looking to implement natural language processing and machine learning into the pre-processing efforts in order to ensure efficiency when dealing with new applications. The USCIS has long backlogs of applicants in many employment based and family categories for certain foreign nationals. These backlogs have been stretched even further due to the Covid-19 pandemic and office closures.
But just how will natural language processing be of assistance? This blog will take a closer look into how NLP might help the USCIS, reduce backlogs, but also possibly result in some biases if the decision making process of the algorithm is inaccurate or unfair.
How Does NLP Work for Immigration?
First off, natural language processing for those who aren’t familiar is basically computer programming that seeks to process and analyze large amounts of natural language data. NLP could be useful in an immigration context by gathering an applicants name and other important biographic data such as their DOB, their marital status, and how many dependents they have or whether surnames are consistent throughout an entire form.
NLP could also be very important for the purpose of detecting fraud on an applicants USCIS form, and thus bringing these matters to the direct attention of an officer before they would even realize it. This would serve as an extra layer of security before an applicant was brought in for an interview, for example.
NLP can also be used more importantly to validate supporting documents that are attached to ones application. Many USCIS documents, such as the I-130 petition, the I-140 petition, and the I-485 application all require supporting documents in order for applicants to be eligible for immigration benefits. Language processing would assist adjudicators by sorting these documents, making sure they are valid, and also decreasing the amount of time it would take for a “swivel chair operation”. In a similar vein, NLP would function to cross check supporting documents with the original application to see if there are any other inconsistencies. This would at least serve as a pre-processing exercise that would make the adjudication process slightly more accurate.
However NLP wouldn’t have the capability of doing more detailed work for adjudicators such as reading an applicants written explanations. This is where the work of sentiment analysis would come in, where there is still a fair deal of uncertainty of how helpful such programming would be to USCIS officers in their adjudication process. For example, how would a machine learning program be able to identify that a couples marriage is legitimate based on their written statements they provide? Such technology will still take years, if anything, to develop.
Dealing with Biases
Overall, the NLP technology is a good start for the USCIS over the next few years. It will help adjudicators with more upfront processing and help “catch” applicants who have inconsistencies on their applications and also verify that certain documents have a seal of approval (from an employer, for instance). But ultimately any other type of machine learning program for immigration is still in the works, according to the USCIS. Hopefully such an investment in this technology will benefit the applicant and reduce their wait times.
For more information on this topic, please see the following link: https://www.fedscoop.com/uscis-automating-immigration-pre-processing/