From f80453e78dd157cdb5f66a401ea4493eccb95c3d Mon Sep 17 00:00:00 2001
From: David Blackman <david@whizziwig.com>
Date: Tue, 5 May 2020 09:42:08 -0400
Subject: [PATCH 1/2] start of fixing ambiguous streets

---
 sanitizer/_text_pelias_parser.js | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/sanitizer/_text_pelias_parser.js b/sanitizer/_text_pelias_parser.js
index 8ed3b1baf..8f146e0bf 100644
--- a/sanitizer/_text_pelias_parser.js
+++ b/sanitizer/_text_pelias_parser.js
@@ -52,6 +52,23 @@ function _sanitize (raw, clean) {
   return messages;
 }
 
+function isAmbiguousStreetOrPlaceParse(query, solutions){
+  if (!solutions) {
+    return false;
+  }
+  const wholeStreetSolution = solutions.find(solution => {
+    return solution.pair &&
+      solution.pair[0].label === 'street' &&
+      solution.pair[0].span.body === query;
+  });
+  const wholePlaceSolution = solutions.find(solution => {
+    return solution.pair &&
+      solution.pair[0].label === 'place' &&
+      solution.pair[0].span.body === query;
+  });
+  return wholeStreetSolution && wholePlaceSolution;
+}
+
 function parse (clean) {
   
   // parse text
@@ -108,6 +125,19 @@ function parse (clean) {
     }
   }
 
+  // If the parser generated two solutions, one where the whole query was classified as a
+  // street, and another where it was classified entirely as a place, discard both solutions
+  // since we don't really know which to use.
+  // 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
+  if (isAmbiguousStreetOrPlaceParse(t.span.body, t.solutions) &&
+    (parsed_text.street || parsed_text.place)) {
+    delete parsed_text.street;
+    delete parsed_text.place;
+    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 : ' ')

From a5594d968d128e5e9330c01394353acf190f4a90 Mon Sep 17 00:00:00 2001
From: David Blackman <david@whizziwig.com>
Date: Tue, 5 May 2020 15:29:34 -0400
Subject: [PATCH 2/2] works

---
 sanitizer/_text_pelias_parser.js           | 41 ++++++++++++++--------
 test/unit/sanitizer/_text_pelias_parser.js |  5 +++
 2 files changed, 31 insertions(+), 15 deletions(-)

diff --git a/sanitizer/_text_pelias_parser.js b/sanitizer/_text_pelias_parser.js
index 8f146e0bf..8ae2d4c8b 100644
--- a/sanitizer/_text_pelias_parser.js
+++ b/sanitizer/_text_pelias_parser.js
@@ -52,25 +52,31 @@ function _sanitize (raw, clean) {
   return messages;
 }
 
-function isAmbiguousStreetOrPlaceParse(query, solutions){
+function isAmbiguousStreetOrPlaceParse(solutions){
   if (!solutions) {
     return false;
   }
-  const wholeStreetSolution = solutions.find(solution => {
+
+  const topTwoSolutions = _.take(solutions, 2);
+  const streetOnlySolution = topTwoSolutions.find(solution => {
     return solution.pair &&
-      solution.pair[0].label === 'street' &&
-      solution.pair[0].span.body === query;
+      solution.pair.length === 1 &&
+      solution.pair[0].classification.label === 'street';
   });
-  const wholePlaceSolution = solutions.find(solution => {
+  const placeOnlySolution = topTwoSolutions.find(solution => {
     return solution.pair &&
-      solution.pair[0].label === 'place' &&
-      solution.pair[0].span.body === query;
+      solution.pair.length === 1 &&
+      solution.pair[0].classification.label === 'place';
   });
-  return wholeStreetSolution && wholePlaceSolution;
+
+  return (
+    streetOnlySolution &&
+    placeOnlySolution &&
+    streetOnlySolution.pair[0].span.body === placeOnlySolution.pair[0].span.body
+  );
 }
 
-function parse (clean) {
-  
+function parse (clean) {  
   // parse text
   let start = new Date();
   const t = new Tokenizer(clean.text);
@@ -125,12 +131,17 @@ function parse (clean) {
     }
   }
 
-  // If the parser generated two solutions, one where the whole query was classified as a
-  // street, and another where it was classified entirely as a place, discard both solutions
-  // since we don't really know which to use.
+  // (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
-  if (isAmbiguousStreetOrPlaceParse(t.span.body, t.solutions) &&
+  // 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 (isAmbiguousStreetOrPlaceParse(t.solution) &&
     (parsed_text.street || parsed_text.place)) {
     delete parsed_text.street;
     delete parsed_text.place;
diff --git a/test/unit/sanitizer/_text_pelias_parser.js b/test/unit/sanitizer/_text_pelias_parser.js
index fb998eec7..433f1d4b0 100644
--- a/test/unit/sanitizer/_text_pelias_parser.js
+++ b/test/unit/sanitizer/_text_pelias_parser.js
@@ -308,6 +308,11 @@ 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',
+  }]);
+
   // admin areas
   cases.push(['N', { subject: 'N' }, true]);
   cases.push(['Ne', { subject: 'Ne' }, true]);