Address Multilingual Compliance Issues Today
Compliance regulations around the world are tightening, requiring financial institutions to screen customers and transactions against terrorism and sanctions watchlists. At the same time, the penalties for non-compliance are increasing. With record fines and executive accountability,the risks of operating a financial institution have never been greater.
Operating in this increasingly global environment presents a whole new set of problems. When customer names are in one language and the watchlist is in a different language, you must still be able to evaluate your risk.
- Are you relying on translations of unknown quality ?
- What is your transliteration policy for personal names ?
- Are there watchlist hits that you are missing ?
Financial services firms must understand who their customers are and how they screen them. If your organization has customers whose names are of foreign origin and are screened against the OFAC SDN list, the UN sanctions list, or third party watchlists, you are increasing your risk exposure. Basis Technology has unique technology designed to help compliance officers understand these risks.
- Simple API
- Fast and Scalable
- Industrial-strength Support
- Easy Installation
- Flexible and Customizable
- Integration: Java or Web Services
- Platform: Unix, Linux, Mac or Windows
REX instantly scans through huge volumes of multilingual, unstructured text and tags key data. REX uses multiple approaches to achieve the most accurate results:
Statistical Entity Extraction
Statistical modeling with advanced linguistics solves the three biggest challenges in entity extraction: finding entities which cannot be exhaustively listed, finding entities which are yet unknown and using context to distinguish between similar entities, e.g., the place “Newton, MA” and the person “Isaac Newton”.
Field Training for Increased Accuracy
For users with text that is particularly challenging in format, style, or vocabulary, REX’s unique field training capability has multiple mechanisms to adapt its statistical model to their data. Users just add a quantity of their data (unannotated or annotated), and rebuild the model for maximum accuracy.
Rules expressed as regular expressions find entities which follow a pattern, such as dates, times, and email addresses. Many standard string patterns are included with REX; customers can customize by editing or adding their own rules, based on their specific needs.
Custom Entity Lists
Custom lists are helpful when users know that specific words or phrases in their data are almost never misspelled and always refer to the same thing (i.e., are unambiguous). REX comes with such lists for entity types like religions and nationalities.
RNI returns a confidence score based on a name’s similarity with existing names in the index. This “fuzzy” search automatically matches the names within large collections of documents and unstructured text, or rescues them from languishing databases.
Unlike expensive and less accurate legacy solutions driven by lists of spelling variants, RNI analyzes the intrinsic structure of each name component and performs an intelligent comparison using advanced linguistic algorithms. This approach is not limited to a particular list of variants and reduces the likelihood of both “false positives” (wrong matches) and “false negatives” (zero hits or missed matches). When only some components of a name match, RNI aligns input names with entries to recognize partial matches.
Customize to Your Need
- Set the minimum threshold of the confidence score to manage the precision and recall of the returned search results.
- Ignore a given list of words (“stopwords”) with respect to matching. (e.g., titles, honorifics)
- Force two name components to always match with a given score. (e.g., “Elizabeth” and “Lisbeth” always match at 90%)
- Force two names to always match with a given score. (e.g., “John Doe” and “Joe Bloggs” always match at 95%)
- Link multiple names to a single individual. (e.g., queries for “Marilyn Monroe” and “Norma Jeane Mortensen” include the same person)
RNT combines dictionary look-ups and transliteration to find the most accurate English spelling of a name. First, the foreign name is examined in user-supplied name dictionaries, known as gazetteers. If the name is not found, RNT transliterates the name into English by using linguistic algorithms and statistical modeling, then matches it using preferred name standards. For example, names written in Chinese are converted from ideographic characters into a phonetic representation. Names written in “unvocalized” Arabic (i.e., without short vowels) are automatically vocalized to enable a phonetic translation according to any of several user-selected standard systems.
- Generate “conventional spellings” of frequently appearing foreign names
- Process “unrecognized” names, i.e., those not appearing in any known catalog of foreign names
- Incorporate complex transliteration standards (such as the IC or U.S. Board on Geographic Names) for translating a name from a foreign alphabet into English
- Automatically resolve name spelling ambiguities in the source language, such as partial vocalization of Arabic, or word segmentation in Chinese
Financial institutions use the combination of REX, RNI, and RNT to create a process for extracting names and other essential entities, matching them to relevant watchlists, and then translating potential matches into standardized English for further investigation.
By integrating the Rosette components into your workflow, you can have greater confidence in your efforts to block terrorists’ and other criminals from gaining access to funds. The result is a reduction in risk of both compliance violations and harm to professional reputations.