don't boost street if parse is ambiguous

sanitizer/_text_pelias_parser.js

@@ -52,6 +52,36 @@ function _sanitize (raw, clean) {
   return messages;
 }
 
+function isAmbiguousStreetOrVenueParse(solutions){
+  if (!solutions || solutions.length < 2) {
+    return false;
+  }
+
+  const topTwoSolutions = _.take(solutions, 2);
+  const streetOnlySolution = topTwoSolutions.find(solution => {
+    return solution.pair &&
+      solution.pair[0].classification.label === 'street';
+  });
+  const venueOnlySolution = topTwoSolutions.find(solution => {
+    return solution.pair &&
+      solution.pair[0].classification.label === 'venue';
+  });
+
+  const restOfSolutionIsTheSame = 
+    topTwoSolutions[0].pair.length === topTwoSolutions[1].pair.length &&
+    _.every(_.zip(topTwoSolutions[0].pair.slice(1), topTwoSolutions[1].pair.slice(1)), ([p1, p2]) => 
+       p1.span.body === p2.span.body && p1.classification.label === p2.classification.label
+    );
+
+  return (
+    streetOnlySolution &&
+    venueOnlySolution &&
+    streetOnlySolution.pair[0].span.body === venueOnlySolution.pair[0].span.body &&
+    restOfSolutionIsTheSame
+  );
+}
+
+
 function parse (clean) {
 
   // parse text
@@ -95,6 +125,23 @@ function parse (clean) {
   // '    VVVV NN SSSSSSS AAAAAA PPPPP      '
   let mask = solution.mask(t);
 
+  // (Theory) If the parser's two top solutions were labeling the entire query as a venue/place,
+  // and labeling the entire query as a street, then the parse is telling us it's pretty 
+  // ambiguous, so we shouldn't bother boosting street or place matches in our search.
+  //
+  // This helps with queries like "wrigley field" that have a word (field) that could be
+  // a street suffix but is also a likely venue word. 
+  //
+  // (Reality) There's no way to tell if the parser labeled the "whole query" because
+  // punctuation in the incoming query is unlabeled, so in reality this checks if
+  // the top two solutions were venue-only and place-only and labeled identical parts of the query
+  if (isAmbiguousStreetOrVenueParse(t.solution)) {
+    delete parsed_text.street;
+    delete parsed_text.venue;
+    mask = mask.replace(/S/g, ' ');
+    mask = mask.replace(/V/g, ' ');
+  }
+
   // the entire input text as seen by the parser with any postcode classification(s) removed
   let body = t.span.body.split('')
     .map((c, i) => (mask[i] !== 'P') ? c : ' ')

test/unit/sanitizer/_text_pelias_parser.js

@@ -311,6 +311,14 @@ module.exports.tests.text_parser = function (test, common) {
   cases.push(['Air & Space Museum D', { subject: 'Air & Space Museum' }, true]);
   cases.push(['Air & Space Museum DC', { subject: 'Air & Space Museum' }, true]);
 
+  // ambiguous venue
+  cases.push(['mccarren park', {
+    subject: 'mccarren park',
+  }]);
+  cases.push(['wrigley field', {
+    subject: 'wrigley field',
+  }]);
+
   // admin areas
   cases.push(['N', { subject: 'N' }, true]);
   cases.push(['Ne', { subject: 'Ne' }, true]);