Natural Language Processing and your Apps

Transcription

1 Session App Frameworks #WWDC17 Natural Language Processing and your Apps 208 Vivek Kumar Rangarajan Sridhar, Software Engineering Manager Doug Davidson, Senior Engineer 2017 Apple Inc. All rights reserved. Redistribution or public display not permitted without written permission from Apple.

2 Goal

3 Goal

4 Goal Natural language input typed text recognized handwriting transcribed speech

5 Goal Natural language output typed text recognized handwriting transcribed speech

6 Goal Natural language text typed recognized handwriting Confer intelligence transcribed speech NLP

7 NLP APIs

8 NLP APIs

9 NLP APIs

10 NLP APIs

11 NLP APIs

12 Natural Language Processing

13 Natural Language Processing Natural language text typed recognized handwriting transcribed speech

14 Natural Language Processing Natural language text typed recognized handwriting Processing transcribed speech

15 Natural Language Processing Natural language text typed recognized handwriting transcribed speech Processing Extract information Confer intelligence

16 Natural Language Processing Natural language text typed recognized handwriting transcribed speech Processing Extract information Confer intelligence

17 Natural Language Processing Natural language text typed recognized handwriting Confer intelligence transcribed speech

18 Natural Language Processing Language identification Natural language text typed recognized handwriting Detect language transcribed speech

19 Natural Language Processing Language identification Natural language text typed recognized handwriting Detect language transcribed speech Welcome to our talk on Natural Language Processing 欢迎参加我们关于 自然语 言处理理的讲座 Bienvenidos a nuestra plática sobre Procesamiento de Lenguaje Natural हम र प र क तक भ ष प रस स करण भ षण म आपक स व गत ह Willkommen zu unserem Vortrag über Verarbeitung natürlicher Sprache

20 Natural Language Processing Language identification Natural language text typed recognized handwriting transcribed speech Tokenization Word Sentence Paragraph Tokenize text

21 Natural Language Processing Language identification Natural language text typed recognized handwriting transcribed speech Tokenization Word Sentence Paragraph Tokenize text Mr. Tim Cook presided over the earnings report of Apple Inc. on Tuesday.

22 Natural Language Processing Language identification Natural language text typed recognized handwriting transcribed speech Tokenization Word Sentence Paragraph Tokenize text 星期 二, 蒂姆 库克先 生主持了了苹果公司的财报会议

23 Natural Language Processing Language identification Natural language text typed recognized handwriting transcribed speech Tokenization Word Sentence Paragraph Tokenize text 星期 二, 蒂姆 库克先 生主持了了苹果公司的财报会议

24 Natural Language Processing Language identification Natural language text typed recognized handwriting transcribed speech Tokenization Word Sentence Paragraph Part of speech Assign parts of speech

25 Natural Language Processing Language identification Natural language text typed recognized handwriting transcribed speech Tokenization Word Sentence Paragraph Part of speech Assign parts of speech Mr. Tim Cook presided over the earnings report of Apple Inc. on Tuesday. Noun NNP NNP Verb PP DT Noun Noun PP Noun Noun PP Noun PUNCT

26 Natural Language Processing Language identification Natural language text typed recognized handwriting transcribed speech Tokenization Word Sentence Paragraph Part of speech Lemmatization Lemmatize text

27 Natural Language Processing Language identification Natural language text typed recognized handwriting transcribed speech Tokenization Word Sentence Paragraph Part of speech Lemmatization Lemmatize text Mr. Tim Cook presided over the earnings report of Apple. The stock was up 3% after hours.

28 Natural Language Processing Language identification Natural language text typed recognized handwriting transcribed speech Tokenization Word Sentence Paragraph Part of speech Lemmatization Lemmatize text Verb Verb Noun Mr. Tim Cook presided over the earnings report of Apple. The stock was up 3% after hours. preside be hour

29 Natural Language Processing Language identification Natural language text typed recognized handwriting transcribed speech Tokenization Word Sentence Paragraph Part of speech Lemmatization Named entity recognition Extract entities Mr. Tim Cook presided over the earnings report of Apple Inc. on Tuesday.

30 Natural Language Processing Language identification Natural language text typed recognized handwriting transcribed speech Tokenization Word Sentence Paragraph Part of speech Lemmatization Named entity recognition Extract entities PER ORG Mr. Tim Cook presided over the earnings report of Apple Inc. on Tuesday.

31 Natural Language Processing Language identification Natural language text typed recognized handwriting transcribed speech Tokenization Word Sentence Paragraph Part of speech Lemmatization Named entity recognition Confer intelligence

32 Natural Language Processing Language identification Natural language text typed recognized handwriting transcribed speech Tokenization Word Sentence Paragraph Part of speech Lemmatization Named entity recognition Confer intelligence Linguistics

33 Natural Language Processing Language identification Natural language text typed recognized handwriting transcribed speech Tokenization Word Sentence Paragraph Part of speech Lemmatization Named entity recognition Confer intelligence Linguistics Machine learning

34 Natural Language Processing NLP APIs Language identification Natural language text typed recognized handwriting transcribed speech Tokenization Word Sentence Paragraph Part of speech Lemmatization Named entity recognition Confer intelligence Linguistics Machine learning

35 Enough! Tell me how to use them

36 NLP APIs

37 NLP APIs NSLinguisticTagger

38 NSLinguisticTagger Class in foundation Segment and tag text Linguistic tasks : tagschemes

39 NSLinguisticTagger Class in foundation Segment and tag text Linguistic tasks : tagschemes

40 NSLinguisticTagger NEW

41 NSLinguisticTagger NEW Tagging units

42 NSLinguisticTagger NEW Tagging units public enum NSLinguisticTaggerUnit : Int { case word case sentence case paragraph case document }

43 NSLinguisticTagger NEW Tagging units public enum NSLinguisticTaggerUnit : Int { case word case sentence case paragraph case document }

44 NSLinguisticTagger NEW Tagging units public enum NSLinguisticTaggerUnit : Int { case word case sentence case paragraph case document }

45 NSLinguisticTagger NEW Tagging units public enum NSLinguisticTaggerUnit : Int { case word case sentence case paragraph case document }

46 NSLinguisticTagger NEW Tagging units public enum NSLinguisticTaggerUnit : Int { case word case sentence case paragraph case document }

47 NSLinguisticTagger NEW Tagging units public enum NSLinguisticTaggerUnit : Int { case word case sentence case paragraph case document } Units and schemes

48 NSLinguisticTagger NEW Tagging units public enum NSLinguisticTaggerUnit : Int { case word case sentence case paragraph case document } Units and schemes class func availabletagschemes (for unit: NSLinguisticTaggerUnit, language: String) -> [NSLinguisticTagScheme]

49 NSLinguisticTagger NEW Tagging units public enum NSLinguisticTaggerUnit : Int { case word case sentence case paragraph case document } Units and schemes class func availabletagschemes (for unit: NSLinguisticTaggerUnit, language: String) -> [NSLinguisticTagScheme]

50 NSLinguisticTagger NEW

51 NSLinguisticTagger NEW dominantlanguage

52 NSLinguisticTagger NEW dominantlanguage Swift 4 : named types for tags and tagschemes

53 NSLinguisticTagger NEW dominantlanguage Swift 4 : named types for tags and tagschemes Improved performance Higher accuracy Additional language support

54 Winnow and Whisk

55 Winnow and Whisk

56 Winnow and Whisk

57 Winnow Tag photos with descriptions

58 Winnow Tag photos with descriptions

59 Winnow Tag photos with descriptions We hiked to the top and then ran down. Lizzie won! Kids partied with all the goodies Die Kleinen haben friedlich zusammen gespielt. July 4th, 2016 summer parties at the Holden s begin!

60 Winnow Search for pictures

61 Winnow Search for pictures Hike

62 Winnow Search for pictures Hike No results

63 Winnow Improve search experience using NLP

64 Winnow Improve search experience using NLP Hike

65 Winnow Improve search experience using NLP Hike We hiked to the top and then ran down. Lizzie won! Great hikes make great pics! Holden family loves hiking and also posing for pictures The hiking boys!

66 Winnow

67 Winnow

68 Winnow NLP

69 Winnow NLP Language identification

70 Winnow NLP Language identification Tokenization Word Sentence Paragraph

71 Winnow NLP Language identification Tokenization Word Sentence Paragraph Part of speech

72 Winnow NLP Language identification Tokenization Word Sentence Paragraph Part of speech Lemmatization

73 Winnow NLP Language identification Tokenization Word Sentence Paragraph Part of speech Lemmatization

74 Language Identification import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.language], options: 0) tagger.string = "Die Kleinen haben friedlich zusammen gespielt." let language = tagger.dominantlanguage

75 Language Identification import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.language], options: 0) tagger.string = "Die Kleinen haben friedlich zusammen gespielt." let language = tagger.dominantlanguage

76 Language Identification import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.language], options: 0) tagger.string = "Die Kleinen haben friedlich zusammen gespielt." let language = tagger.dominantlanguage

77 Language Identification import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.language], options: 0) tagger.string = "Die Kleinen haben friedlich zusammen gespielt." let language = tagger.dominantlanguage

78 Language Identification import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.language], options: 0) tagger.string = "Die Kleinen haben friedlich zusammen gespielt." let language = tagger.dominantlanguage

79 Tokenization import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.tokentype], options: 0) let text = "NSLinguisticTagger provides text processing APIs.\n NSLinguisticTagger 是苹果的 文字处理理平台 " tagger.string = text let range = NSRange(location: 0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace] tagger.enumeratetags(in: range, unit:.word, scheme:.tokentype, options: options) { tag, tokenrange, stop in let token = (text as NSString).substring(with: tokenrange) // Do something with each token }

80 Tokenization import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.tokentype], options: 0) let text = "NSLinguisticTagger provides text processing APIs.\n NSLinguisticTagger 是苹果的 文字处理理平台 " tagger.string = text let range = NSRange(location: 0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace] tagger.enumeratetags(in: range, unit:.word, scheme:.tokentype, options: options) { tag, tokenrange, stop in let token = (text as NSString).substring(with: tokenrange) // Do something with each token }

81 Tokenization import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.tokentype], options: 0) let text = "NSLinguisticTagger provides text processing APIs.\n NSLinguisticTagger 是苹果的 文字处理理平台 " tagger.string = text let range = NSRange(location: 0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace] tagger.enumeratetags(in: range, unit:.word, scheme:.tokentype, options: options) { tag, tokenrange, stop in let token = (text as NSString).substring(with: tokenrange) // Do something with each token }

82 Tokenization import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.tokentype], options: 0) let text = "NSLinguisticTagger provides text processing APIs.\n NSLinguisticTagger 是苹果的 文字处理理平台 " tagger.string = text let range = NSRange(location: 0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace] tagger.enumeratetags(in: range, unit:.word, scheme:.tokentype, options: options) { tag, tokenrange, stop in let token = (text as NSString).substring(with: tokenrange) // Do something with each token }

83 Tokenization import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.tokentype], options: 0) let text = "NSLinguisticTagger provides text processing APIs.\n NSLinguisticTagger 是苹果的 文字处理理平台 " tagger.string = text let range = NSRange(location: 0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace] tagger.enumeratetags(in: range, unit:.word, scheme:.tokentype, options: options) { tag, tokenrange, stop in let token = (text as NSString).substring(with: tokenrange) // Do something with each token }

84 Tokenization import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.tokentype], options: 0) let text = "NSLinguisticTagger provides text processing APIs.\n NSLinguisticTagger 是苹果的 文字处理理平台 " tagger.string = text let range = NSRange(location: 0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace] tagger.enumeratetags(in: range, unit:.word, scheme:.tokentype, options: options) { tag, tokenrange, stop in let token = (text as NSString).substring(with: tokenrange) // Do something with each token }

85 Tokenization import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.tokentype], options: 0) let text = "NSLinguisticTagger provides text processing APIs.\n NSLinguisticTagger 是苹果的 文字处理理平台 " tagger.string = text let range = NSRange(location: 0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace] tagger.enumeratetags(in: range, unit:.word, scheme:.tokentype, options: options) { tag, tokenrange, stop in let token = (text as NSString).substring(with: tokenrange) // Do something with each token }

86 Lemmatization import Foundation let tagger = NSLinguisticTagger(tagSchemes:[.lemma], options: 0) let text = "Great hikes make great pics! Wonderful afternoon in Marin County." tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace] tagger.enumeratetags(in: range, unit:.word, scheme:.lemma, options: options) { tag, tokenrange, stop in if let lemma = tag?.rawvalue { // Do something with each lemma } }

87 Lemmatization import Foundation let tagger = NSLinguisticTagger(tagSchemes:[.lemma], options: 0) let text = "Great hikes make great pics! Wonderful afternoon in Marin County." tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace] tagger.enumeratetags(in: range, unit:.word, scheme:.lemma, options: options) { tag, tokenrange, stop in if let lemma = tag?.rawvalue { // Do something with each lemma } }

88 Lemmatization import Foundation let tagger = NSLinguisticTagger(tagSchemes:[.lemma], options: 0) let text = "Great hikes make great pics! Wonderful afternoon in Marin County." tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace] tagger.enumeratetags(in: range, unit:.word, scheme:.lemma, options: options) { tag, tokenrange, stop in if let lemma = tag?.rawvalue { // Do something with each lemma } }

89 Lemmatization import Foundation let tagger = NSLinguisticTagger(tagSchemes:[.lemma], options: 0) let text = "Great hikes make great pics! Wonderful afternoon in Marin County." tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace] tagger.enumeratetags(in: range, unit:.word, scheme:.lemma, options: options) { tag, tokenrange, stop in if let lemma = tag?.rawvalue { // Do something with each lemma } }

90 Lemmatization import Foundation let tagger = NSLinguisticTagger(tagSchemes:[.lemma], options: 0) let text = "Great hikes make great pics! Wonderful afternoon in Marin County." tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace] tagger.enumeratetags(in: range, unit:.word, scheme:.lemma, options: options) { tag, tokenrange, stop in if let lemma = tag?.rawvalue { // Do something with each lemma } }

91 Lemmatization import Foundation let tagger = NSLinguisticTagger(tagSchemes:[.lemma], options: 0) let text = "Great hikes make great pics! Wonderful afternoon in Marin County." tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace] tagger.enumeratetags(in: range, unit:.word, scheme:.lemma, options: options) { tag, tokenrange, stop in if let lemma = tag?.rawvalue { // Do something with each lemma } }

92 Lemmatization import Foundation let tagger = NSLinguisticTagger(tagSchemes:[.lemma], options: 0) let text = "Great hikes make great pics! Wonderful afternoon in Marin County." tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace] tagger.enumeratetags(in: range, unit:.word, scheme:.lemma, options: options) { tag, tokenrange, stop in if let lemma = tag?.rawvalue { // Do something with each lemma } }

93 Lemmatization import Foundation let tagger = NSLinguisticTagger(tagSchemes:[.lemma], options: 0) let text = "Great hikes make great pics! Wonderful afternoon in Marin County." tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace] tagger.enumeratetags(in: range, unit:.word, scheme:.lemma, options: options) { tag, tokenrange, stop in if let lemma = tag?.rawvalue { // Do something with each lemma } }

94 Demo

95 Whisk Collate social media feeds

96 Whisk Collate social media feeds

97 Whisk Collate social media feeds

98 Whisk Organize feeds by People, Organization, and Location using NLP

99 Whisk Organize feeds by People, Organization, and Location using NLP

100 Whisk Organize feeds by People, Organization, and Location using NLP Pharrell, NYU Tim Cook, Apple, Stevie Wonder, Cupertino, Angela Ahrendts Bruce Wayne, Thomas Vintberg

101 Whisk

102 Whisk

103 Whisk

104 Whisk NLP

105 Whisk NLP Language identification

106 Whisk NLP Language identification Tokenization Word Sentence Paragraph

107 Whisk NLP Language identification Tokenization Word Sentence Paragraph Named entity recognition

108 Whisk NLP Language identification Tokenization Word Sentence Paragraph Named entity recognition

109 Named Entity Recognition import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.nametype], options: 0) let text = "Tim Cook is the CEO of Apple Inc. which is located in Cupertino, California" tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace,.joinnames] let tags: [NSLinguisticTag] = [.personalname,.placename,.organizationname] tagger.enumeratetags(in: range, unit:.word, scheme:.nametype, options: options) { tag, tokenrange, stop in if let tag = tag, tags.contains(tag) { let name = (text as NSString).substring(with: tokenrange) } }

110 Named Entity Recognition import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.nametype], options: 0) let text = "Tim Cook is the CEO of Apple Inc. which is located in Cupertino, California" tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace,.joinnames] let tags: [NSLinguisticTag] = [.personalname,.placename,.organizationname] tagger.enumeratetags(in: range, unit:.word, scheme:.nametype, options: options) { tag, tokenrange, stop in if let tag = tag, tags.contains(tag) { let name = (text as NSString).substring(with: tokenrange) } }

111 Named Entity Recognition import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.nametype], options: 0) let text = "Tim Cook is the CEO of Apple Inc. which is located in Cupertino, California" tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace,.joinnames] let tags: [NSLinguisticTag] = [.personalname,.placename,.organizationname] tagger.enumeratetags(in: range, unit:.word, scheme:.nametype, options: options) { tag, tokenrange, stop in if let tag = tag, tags.contains(tag) { let name = (text as NSString).substring(with: tokenrange) } }

112 Named Entity Recognition import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.nametype], options: 0) let text = "Tim Cook is the CEO of Apple Inc. which is located in Cupertino, California" tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace,.joinnames] let tags: [NSLinguisticTag] = [.personalname,.placename,.organizationname] tagger.enumeratetags(in: range, unit:.word, scheme:.nametype, options: options) { tag, tokenrange, stop in if let tag = tag, tags.contains(tag) { let name = (text as NSString).substring(with: tokenrange) } }

113 Named Entity Recognition import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.nametype], options: 0) let text = "Tim Cook is the CEO of Apple Inc. which is located in Cupertino, California" tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace,.joinnames] let tags: [NSLinguisticTag] = [.personalname,.placename,.organizationname] tagger.enumeratetags(in: range, unit:.word, scheme:.nametype, options: options) { tag, tokenrange, stop in if let tag = tag, tags.contains(tag) { let name = (text as NSString).substring(with: tokenrange) } }

114 Named Entity Recognition import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.nametype], options: 0) let text = "Tim Cook is the CEO of Apple Inc. which is located in Cupertino, California" tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace,.joinnames] let tags: [NSLinguisticTag] = [.personalname,.placename,.organizationname] tagger.enumeratetags(in: range, unit:.word, scheme:.nametype, options: options) { tag, tokenrange, stop in if let tag = tag, tags.contains(tag) { let name = (text as NSString).substring(with: tokenrange) } }

115 Named Entity Recognition import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.nametype], options: 0) let text = "Tim Cook is the CEO of Apple Inc. which is located in Cupertino, California" tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace,.joinnames] let tags: [NSLinguisticTag] = [.personalname,.placename,.organizationname] tagger.enumeratetags(in: range, unit:.word, scheme:.nametype, options: options) { tag, tokenrange, stop in if let tag = tag, tags.contains(tag) { let name = (text as NSString).substring(with: tokenrange) } }

116 Named Entity Recognition import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.nametype], options: 0) let text = "Tim Cook is the CEO of Apple Inc. which is located in Cupertino, California" tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace,.joinnames] let tags: [NSLinguisticTag] = [.personalname,.placename,.organizationname] tagger.enumeratetags(in: range, unit:.word, scheme:.nametype, options: options) { tag, tokenrange, stop in if let tag = tag, tags.contains(tag) { let name = (text as NSString).substring(with: tokenrange) } }

117 Named Entity Recognition import Foundation let tagger = NSLinguisticTagger(tagSchemes: [.nametype], options: 0) let text = "Tim Cook is the CEO of Apple Inc. which is located in Cupertino, California" tagger.string = text let range = NSRange(location:0, length: text.utf16.count) let options: NSLinguisticTagger.Options = [.omitpunctuation,.omitwhitespace,.joinnames] let tags: [NSLinguisticTag] = [.personalname,.placename,.organizationname] tagger.enumeratetags(in: range, unit:.word, scheme:.nametype, options: options) { tag, tokenrange, stop in if let tag = tag, tags.contains(tag) { let name = (text as NSString).substring(with: tokenrange) } }

118 Demo

119 Why should you use the NLP APIs?

120 Homogeneous Text Processing

121 Homogeneous Text Processing NLP APIs Natural language text typed recognized handwriting transcribed speech Consistent text processing Consistent user experience

122 Privacy

123 Privacy On-device machine learning User data stays on-device

124 Performance

125 Performance Highly optimized on-device Multi-threaded Existing clients significant speedup

126 Performance Highly optimized on-device Multi-threaded Existing clients significant speedup Chinese tokenization is 30% faster on ios

127 Performance Highly optimized on-device Multi-threaded Existing clients significant speedup Named Entity Recognition 80% faster on ios

128 Performance

129 Performance Part of Speech Tagging tokens/sec tokens/sec 1 thread

130 Performance Part of Speech Tagging tokens/sec Named Entity Recognition tokens/sec tokens/sec tokens/sec 1 thread

131 Language Support

132 Language Support Language identification 29 scripts, 52 languages

133 Language Support Language identification 29 scripts, 52 languages Tokenization All ios/macos system languages

134 Language Support Language identification 29 scripts, 52 languages Tokenization All ios/macos system languages Lemmatization English French Italian German Spanish Portuguese Russian Turkish

135 Language Support Language identification 29 scripts, 52 languages Tokenization All ios/macos system languages Lemmatization Part of speech English French Italian German Spanish Portuguese Russian Turkish

136 Language Support Language identification 29 scripts, 52 languages Tokenization All ios/macos system languages Lemmatization Part of speech Named entity recognition English French Italian German Spanish Portuguese Russian Turkish

137 Language Support Language identification 29 scripts, 52 languages Tokenization All ios/macos system languages Lemmatization Part of speech Named entity recognition English French Italian German Spanish Portuguese Russian Turkish

138 Accuracy

139 Accuracy Part of speech 0 English Spanish Named entity recognition

140 Debugging Hints

141 Debugging Hints Tags are NSLinguisticTagOtherWord Model not downloaded

142 Debugging Hints Tags are NSLinguisticTagOtherWord Model not downloaded Downloaded over-the-air

143 Debugging Hints Tags are NSLinguisticTagOtherWord Model not downloaded Downloaded over-the-air Explicitly set language if known

144 Summary NLP APIs NSLinguisticTagger Support for new units Faster Higher accuracy More languages

145 More Information

146 Related Sessions What's New in Cocoa Touch WWDC 2017 Introducing Core ML WWDC 2017 Vision Framework: Building on Core ML Hall 2 Wednesday 3:10PM Core ML in Depth Hall 3 Thursday 9:00 AM Accelerate and Sparse Solvers Grand Ballroom A Thursday 10:00 AM

147 Labs Core ML & Natural Language Processing Lab Technology Lab D Thu 11:00AM-3:30PM Vision Lab Technology Lab A Fri 1:50PM-4:00PM Core ML & Natural Language Processing Lab Technology Lab D Fri 1:50PM-4:00PM Cocoa Lab Technology Lab B Fri 1:50PM-3:20PM

148