How I Got Here - NLP, Geospatial, and beyond

Nara Institute of Science and Technology
Watanabe Laboratory - NAIST NLP
https://nlp.naist.jp/

Transcript

1. How I Got Here SCONJ PRON VERB ADV NLP, Geospatial, and beyond advmod nsubj advmod @sorami 2025-05-16 Nara Institute of Science and Technology
2. Hello NAIST NLP! ¥HISAMOTO Sorami ٱຊۭւ ¥NAIST, 2012-2014 Matsumoto Lab Predecessor of Watanabe Lab ¥MIERUNE Inc. Software Engineer Web, Geospatial, Visualization ¥Lives in Hokkaido
3. How I Got Here SCONJ PRON VERB ADV advmod nsubj advmod
4. Hokkaido ๺ւಓ 21st biggest island in the world 83,424 km² 22% of Japan 5.3 million people 4.2% of Japan
5. Higashikawa ౦઒ொ Sapporo ࡳຈࢢ
6. https:@Awww.openstreetmap.org/relation/4061071 Half of the town is in the National Park
8. Asahi-dake, the tallest mountain in Hokkaido 2024-06-16
9. 2025-04-12
10. btw,
11. Just got married!
12. ¥HISAMOTO Sorami ٱຊۭւ ¥NAIST, 2012-2014 Matsumoto Lab Predecessor of Watanabe Lab ¥MIERUNE Inc. Software Engineer Web, Geospatial, Visualization ¥Lives in Hokkaido
13. ¥SHIROMIZU Sorami നਫۭւ ¥NAIST, 2012-2014 Matsumoto Lab Predecessor of Watanabe Lab ¥MIERUNE Inc. Software Engineer Web, Geospatial, Visualization ¥Lives in Hokkaido
14. ¥SHIROMIZU Sorami നਫۭւ ¥NAIST, 2012-2014 Matsumoto Lab Predecessor of Watanabe Lab ¥MIERUNE Inc. Software Engineer Web, Geospatial, Visualization ¥Lives in Hokkaido → To Bangkok, Thailand (from next week É!)
15. ¥ 2014-2016 BrainPad Data Scientist ¥ 2017-2020 Works Applications R&D Engineer Tokushima Laboratory of AI and NLP ¥ 2018-2019 Johns Hopkins University Researcher Center for Language and Speech Processing ¥ 2020-2022 Legalscape Software Engineer ¥ 2022-2025 MIERUNE
16. How I Got Here SCONJ PRON VERB ADV advmod nsubj advmod
17. TodayÕs Agenda 17 ¥NLP & I ¥NLP → Geospatial / Visualization ¥Introduction to Geospatial ¥Geo x NLP
18. 2012-2014 NAIST Nara Institute of Science & Technology
22. Illustration generated by ChatGPT Time flies like an arrow É!
23. NLP Natural Language Processing
24. 24 https:@Awww.anlp.jp/en/
25. 25 https:@Aobservablehq.com/@sorami/nenji-stat
28. NLP 2013 @ Nagoya (my first Nenji-taikai) 246 talks, 625 participants
30. NLP 2025 @ Nagasaki 777 talks 2,309 participants
31. Why NLP ?
32. NLP & I
33. 33 ¥2012-2014 NAIST Parsing & Word Representations ¥2017-2020 WAP Tokushima NLP Sudachi & chiVe - tokenizer & embedding ¥2018-2019 Johns Hopkins University Machine Translation & Privacy ¥2020-2022 Legalscape LegalTech - Law & NLP
34. Parsing & Word Representations NAIST, 2012-2014 34 https:@Aspeakerdeck.com/sorami/parsing-the-web-with-deep-learning
35. Parsing & Word Representations NAIST, 2012-2014 35 Poster at the 13th International Conference on Parsing Technologies (IWPT 2013)
36. Parsing & Word Representations NAIST, 2012-2014 36 ݴޠॲཧֶձୈ19ճ೥࣍େձ (NLP2013) 
37. Sudachi & chiVe - tokenizer & embedding WAP Tokushima NLP, 2017-2020 37 https:@Aspeakerdeck.com/sorami/sudachi-elasticsearch
38. Sudachi & chiVe - tokenizer & embedding WAP Tokushima NLP, 2017-2020 38 https:@Aspeakerdeck.com/sorami/sudachi-elasticsearch
39. Sudachi & chiVe - tokenizer & embedding WAP Tokushima NLP, 2017-2020 39 https:@Azenn.dev/sorami/articles/c9a506000fd1fbd1cf98
40. Sudachi & chiVe - tokenizer & embedding WAP Tokushima NLP, 2017-2020 40 https:@Aspeakerdeck.com/sorami/chive-zhi-pin-li-yong-ke-neng-nari-ben-yu-dan-yu-bekutoruzi-yuan-falseshi-xian-hexiang-kete
41. Sudachi & chiVe - tokenizer & embedding WAP Tokushima NLP, 2017-2020 41 ୈ16ճςΩετΞφϦςΟΫεɾγϯϙδ΢Ϝ (2020)
 ݴޠཧղͱίϛϡχέʔγϣϯݚڀձ ༏लݚڀ৆
    (Best Paper Award) LREC 2018
42. Machine Translation & Privacy Johns Hopkins University, 2018-2019 42 https:@Aspeakerdeck.com/sorami/acl2020-membership-inference-attacks-on-sequence-to-sequence-models-is-my-data-in-your-machine-translation-system
43. Machine Translation & Privacy Johns Hopkins University, 2018-2019 43 https:@Aspeakerdeck.com/sorami/acl2020-membership-inference-attacks-on-sequence-to-sequence-models-is-my-data-in-your-machine-translation-system
44. Machine Translation & Privacy Johns Hopkins University, 2018-2019 44 TACL 2020
45. LegalTech - Law & NLP Legalscape, 2020-2022 45 https:@Awww.legalscape.jp/ https:@Awww.legalscape.co.jp/
46. LegalTech - Law & NLP Legalscape, 2020-2022 46 ݴޠॲཧֶձୈճ೥࣍େձ (NLP2022) Verification of machine-processed pseudonymization of legal precedents toward appropriate open data in civil cases https:@Anote.com/legalscape/n/nf6341940deaa
47. LegalTech - Law & NLP Legalscape, 2020-2022 47
48. LegalTech - Law & NLP Legalscape, 2020-2022 48
49. What I Do Now
51. https:@Awww.mierune.co.jp/
52. ¥Geospatial technology company ¥Since 2016, HQ in Hokkaido ¥36 members as of 2025.05
53. 53 https:@Ads.yahoo.co.jp/ Flow of People Yahoo! JAPAN MineCraft Sapporo, Takamatsu, etc. AIST 3DDB Client AIST ࢈૯ݚ www.digiarc.aist.go.jp/team/gsvrt/information/aist-3ddb-client/ https:@Atakamatsu-mymachi.jp/ Project PLATEAU MLIT ࠃ౔ަ௨ল https:@Agithub.com/Project-PLATEAU/mapengine-survey
54. Heritage Map ಸྑจԽࡒݚڀॴ Find cultural properties and ruins on the web map 54 https:@Aheritagemap.nabunken.go.jp/
55. Rekichizu Modern-style Historical Map 55 https:@Arekichizu.jp/
56. Rekichizu Modern-style Historical Map 56 https:@Aspeakerdeck.com/hjmkth/231129-foss4g-asia-2023-kato
57. ÒBirth of SapporoÓ Special Website Collaboration with a historical novel 57 ໳Ҫܚتʰࡳຈ஀ੜʱʢՏग़ॻ๪৽ࣾ, 2025ʣ https:@Asapporo-tanjo.rekichizu.jp/
58. ÒBirth of SapporoÓ special website Collaboration with a historical novel 58 ໳Ҫܚتʰࡳຈ஀ੜʱʢՏग़ॻ๪৽ࣾ, 2025ʣ https:@Asapporo-tanjo.rekichizu.jp/
59. Finding ancient tombs With geospatial data and machine learning models 59 https:@Awww.nikkei.com/article/DGXZQOUF134BP0T11C23A2000000/ https:@Awww.fujitv.co.jp/sekainonandakore/archive/20250423.html
60. Commercial Company, Born from OSS Community Founded by the committee members of ÒFOSS4G HokkaidoÓ 60 Image: https:@Atwitter.com/howmori/status/1150064830851645440
61. 61 FOSS4G Free and Open Source Software for Geospatial
62. 62 https:@A2025.foss4g.org/
63. We ❤ Open Source! 63 https:@Adev.to/mierune https:@Agithub.com/MIERUNE/ https:@Aqiita.com/organizations/MIERUNE
64. Financial & Operational Contributions If you want to go far, go together 64 QGIS Desktop geospatial software €32,000 2017~ + over €6,600 w/ certificate program In 2025, Medium→Large: €9,000/year MapLibre Web map library project $40,000 2022~ ※The first sponsor in the world Others - AWS, Meta, Microsoft, etc.
65. Financial & Operational Contributions If you want to go far, go together 65 ¥QGIS Country User Group, Voting Member ¥MapLibre, Voting Member ¥OSGeo Foundation, Charter Member ¥OSGeo.JP, Company Member ¥etc., ...
66. Internship If you are interested, come talk to us! 66 https:@Anote.com/mierune/m/m0352153f87dd
67. What I Do Now Personally
68. https:@Asorami.dev/
69. A People Map of Japan City names are replaced by their most WikipediaÕed person 69 https://sorami.dev/2021/people-map-japan/
70. Japanese from Hokkaido to Okinawa Population of the cities along latitude (north-south) 70 https://observablehq.com/@sorami/japanese-from-hokkaido-to-okinawa
71. HOKKAIDLE Spot the cities in Hokkaido - Inspired by WORDLE 71 https://sorami.dev/hokkaidle/
72. Country ÒShiritoriÓ Japanese word chain game, based on last sound 72 https://sorami.dev/city-shiritori/
73. The ÒRoundestÓ Lake in Japan Uses geospatial techniques and gradient descent 73 https://zenn.dev/mierune/articles/9f970dc3e61a66
74. N NLP Geo Vis
75. N NLP Geo Vis ???
76. 76 https:@Adailyportalz.jp/kiji/reading-dictionary
77. 77 https:@Adailyportalz.jp/kiji/reading-dictionary "What's interesting about reading a dictionary cover to cover?" Nishimura Masayuki, Daily Portal Z 2021-12-09
78. 78 https:@Adailyportalz.jp/kiji/reading-dictionary ਫ໺ɿ͋ͱɺࣙయ͕޷͖ͳਓ͸ɺ ɹɹɹՋͳ࣌ʹάʔάϧϚοϓݟΔઆɻ Mizuno: Also, people who like dictionaries probably look at Google Maps in their free time. ੢ଜɿݟ·͢Ͷʂ Nishimura: Yes!
79. 79 https:@Adailyportalz.jp/kiji/reading-dictionary Ҵ઒ɿ࣮͸ɺ஍ਤ΋ ɹɹʮݱ࣮ͷ΋ͷΛͳΔ͚ͩਖ਼֬ʹॻ͖ࣸͯ͠ɺ ͲͷΑ͏ʹ·ͱΊͯදݱ͢Δͷ͔ʯ ͱ͍͏ͱ͜Ζ͸ɺࣙయͱڞ௨͢Δͱ͜Ζ͕͋ΔΜͰ͢ΑͶɻ Inagawa: Actually, maps have something in common with dictionaries in that they involve copying real things as accurately as possible and then figuring out how to summarize and express them.
80. 80 ৫ా෢༤ʰ஍ਤͷྺ࢙ ੈքฤɾ೔ຊฤʱ, p.15 ߨஊֶࣾज़จݿ, 2018 ஍ਤͷྺ࢙͸ݹ͍ɻ ਓྨͷ࠷΋وॏͳจԽҨ࢈Ͱ͋ΔจࣈΑΓ΋ݹ͍ͱ͍ΘΕΔɻ “ ” The history of maps is old. It is said to be older than the most precious cultural heritage of mankind, writing.
81. 81 ʰσʔλࢹ֮Խͷਓྨ࢙ʱ ϑϨϯυϦʔˍ΢ΣΠφʔʢ൧ౢوࢠ༁ʣ ੨౔ࣾ, 2021 “A History of Data Visualization & Graphic Communication”
82. Wall paintings in the Lascaux Caves Estimated to be 17,300 years old 82 Image: Prof saxx / Wikimedia
83. 83 ྺ࢙Λ௨ͯ͡ɺਓʑͷߟ͑΍ݱ৅͸ ࡾͭͷܗଶͰදݱ͞Ε͖ͯͨɻ ͢ͳΘͪɺݴ༿ɺ਺ࣈɺͦͯ͠ਤ৅Ͱ͋Δɻ “ ” Throughout history, people's ideas and phenomena have been expressed in three forms: words, numbers, and images. p.21-22
84. Anaximander's World Map (Reconstruction) Circa 500 BCE 84 Image: Bibi Saint-Paul / Wikimedia
85. 85 ࣌୅Λ͞ΒʹਐΊΔͱɺࢹ֮Խͷ࣍ͷൃలظ͸ɺ ࠓͰ͸ਤදͱ͔ॳظͷ஍ਤͱݺ͹Ε͍ͯΔ΋ͷ ÑֆͰදݱ͞Ε͍ͯͳ͕Β΋ந৅తͳ৘ใΛද͍ͯ͠ΔÑ Ͱىͬͨ͜ɻ “ ” Moving further in time, the next development in visualization occurred with what we now call charts and early maps Ñpictorial yet abstract representations of information. p.27
86. 86 ¥Both ÒlanguageÓ and ÒmapÓ are Tools for Thought ¥Human cognition is limited by the structure of the human body ¥Still, tools and concepts allow us to think the unthinkable
87. N NLP Geo Vis
88. 3 years ago (around 2021) While doing NLP, visualization as a hobby... 88 My skectes while I was learning Amelia Wattenberger’s “Fullstack D3 and Data Visualization” course
89. Naturally, this leads to an interest in maps... 89
90. John Snow's Cholera Map 1854 A pioneer in ÒData VisualizationÓ 90 en.wikipedia.org/wiki/File:Snow-cholera-map-1.jpg (public domain) ¥ Cholera outbreak in 19th century London ¥ The mechanism of infection was unclear based on the scientific knowledge at the time ¥ Dr. Snow plotted the patient's homes on a map and discovered a correlation between the number of infected people around a certain well
91. Then, I joined a new online community ÒLearning Data Visualization with the CommunityÓ 91 https:@Adata-viz-manabiba.visualizing.jp
92. In that community É 92 FURUKAWA Yasuto MIERUNE co-founder We run a geospatial & visualization company called MIERUNE in Hokkaido
93. Then, I went to give a talk at their event Without knowing anyone, I jumped in and signed up to be a speaker 93 https:@Aspeakerdeck.com/sorami/mierune-meetup-mini-number-01
94. Then, I went to give a talk at their event Without knowing anyone, I jumped in and signed up to be a speaker 94 https:@Aspeakerdeck.com/sorami/mierune-meetup-mini-number-01 This is how I joined MIERUNE and moved to Hokkaido !!!
95. By coincidence, at the same time É I joined NAIST in 2012, Ouchi-san joined in 2013 95 https:@Ax.com/blankeyelephant/status/1498197522548150275 Awarded a KAKENHI B grant for our research on computational understanding of text and grounding it on real-world maps É “ ”
96. 96 https:@Anote.com/mierune/n/nbffaebc4b1d7 https:@Anote.com/mierune/n/n02e261e96f25
97. 97 https:@Aanlp.jp/nlp2025/
98. Crossed Pathes
99. WAP Tokushima NLP, 2017- 99 https:@Awww.nikkei.com/article/DGXLZO12701880Y7A200C1TI5000/ ¥ Lab opened in February 2017 ¥ Prof. Matsumoto as technical advisor ¥ At the time, I had just quit my job without deciding on a next one ¥ He introduced me to this new lab, then I joined in June 2017
100. Johns Hopkins University, 2018- 100 ¥ Kevin Duh was my supervisor at Matsumoto Lab (2012-2014) ¥ He then moved to JHU ¥ He saw our work on Sudachi and Elasticsearch (full-text search) ¥ He was doing a project involving Information Retrieval; Then he invited me to JHU https:@Awww.cs.jhu.edu/~kevinduh/
101. Legalscape, 2020- 101 I got to know this person via this Pull Request https:@Agithub.com/WorksApplications/elasticsearch-sudachi/pull/60
102. Legalscape, 2020- 102 https:@Agithub.com/WorksApplications/elasticsearch-sudachi/pull/60 He is the CTO and co-founder of Legalscape NLP at Tokyo U. / NII Aizawa Lab
103. Legalscape, 2020- 103 https:@Anote.com/legalscape/n/nf9f4c249f24e
104. Be grateful, Cherish relationships!
105. Geospatial A quick tour
106. Data
107. What is ÒGeospatialÓ Data? ஍ཧۭؒ৘ใ All kinds of data on (and under) the Earth (and other planets) 107
108. What is ÒGeospatialÓ Data? ஍ཧۭؒ৘ใɾҐஔ৘ใ All kinds of data on (and under) the Earth (and other planets) 108 Ò80% of Data is GeographicÓ
109. What is ÒGeospatialÓ Data? All kinds of data on (and under) the Earth (and other planets) 109 Location Point, Line, Polygon Properties + Name, Year, ID, É (Can be anything)
110. What is ÒGeospatialÓ Data? All kinds of data on (and under) the Earth (and other planets) 110 https:@Amaps.gsi.go.jp/ Satellite Imagery and Aerial Photos are also Geospatial Data (raster data)
111. Data: OpenStreetMap 2004~ Free Wiki world map (ÒWikipediaÓ of Map) 111 https:@Awww.openstreetmap.org/
112. Data: Mapillary 2013~ Street Images (like Google Street View) - acquired by Meta in 2020 112 https:@Awww.mapillary.com/
113. Data: Overture Maps Foundation 2022~ Company alliance, founded by AWS, Meta, Microsoft, & TomTom 113 https:@Aoverturemaps.org/ ¥ Addresses ¥ Base water, land, land use, infrastructure, land cover ¥ Buildings ¥ Divisions ¥ Places ¥ Transportation
114. Project PLATEAU 2020~ Create and utilize 3D city models of Japanese cities Led by the government (MLIT: Ministry of Land, Infrastructure, Transport & Tourism - ࠃ౔ަ௨ল) 114 https:@Aplateauview.mlit.go.jp/
115. Project PLATEAU 2020~ 115 https:@Aspeakerdeck.com/sorami/foss4g-asia-2024
116. GIS
117. What is ÒGISÓ? 117 Geographic Information System A system to view, create, edit, or analyze geospatial data ※May refer to ÒGeographic Information ScienceÓ instead
118. 118 Desktop GIS WebGIS ¥ Software to be installed on a PC ¥ Advanced analysis, etc. possible ¥ Via a web brower ¥ No need to install, easy to view
119. Dekstop GIS: ÒQGISÓ Free and open source software - since 2002 119 https:@Aqgis.org/
121. 121 Public Transport ÒGTFS-GOÓ plugin Earth Observation ÒJAXA Earth APIÓ plugin ※ Both plugins developed by MIERUNE
122. QGIS LAB by MIERUNE Lots of articles on various topics (in Japanese) 122 https:@Aqgis.mierune.co.jp/
123. WebGIS: MapLibre/Mapbox GL JS 2D/2.5D interactive maps 123 https:@Amaplibre.org/ https:@Awww.asahi.com/special/jiko-kosaten/ Map of 1 million traffic accidents nationwide The Asahi Shimbun
124. WebGIS: Cesium GL JS 3D geospatial visualization 124 https:@Acesium.com/platform/cesiumjs/ https:@Awww.yomiuri.co.jp/world/ukraine-reconstruction/ Ukraine: Reconstruction during the war - Report from the outskirts of Kyiv Yomiuri Shimbun
125. WebGIS: deck.gl A visualization framework 125 https:@Adeck.gl/ https:@Avdata.nikkei.com/newsgraphics/land-value-map/ What are the land prices in your town? Nikkei
126. The difficulty of geospatial data The Earth is not flat 126 Image: @vlandham, https:@Agist.githubusercontent.com/vlandham/raw/9216751/ https:@Ax.com/mourner/status/1458169016456032260/
127. The difficulty of geospatial data ÒMap Projection for the Interactive MediaÓ (in Japanese) 127 https:@Aspeakerdeck.com/sorami/mierune-she-nei-mian-qiang-hui-number-033
128. Example: Missile from North Korea 128 Fig. on a Newspaper Fixed Version ਤ: Esri, ArcUser Online, “Understanding Geodesic Buffering” Figure 1, Figure 2 https:@Awww.esri.com/news/arcuser/0111/geodesic.html
129. Recommended Books (in Japanese) Written by CTO of MIERUNE 129 ʰݱ৔ͷϓϩ͕Θ͔Γ΍͘͢ڭ͑ΔҐஔ৘ใΤϯδχΞཆ੒ߨ࠲ʱ Ҫޱ૗େ (ल࿨γεςϜ, 2023) ʰݱ৔ͷϓϩ͕Θ͔Γ΍͘͢ڭ͑ΔҐஔ৘ใσϕϩούʔཆ੒ߨ࠲ʱ Ҫޱ૗େ (ल࿨γεςϜ, 2024)
130. FYI: Book Review (in Japanese) 130 https:@Aqiita.com/sorami/items/bfeebba21216609ad576
131. NLP & Geospatial
132. Hiroki Ouchi ÒGeospatial in the TextÓ Toward the Integration of GIS and NLP 132 https:@Aspeakerdeck.com/hiroki13/wen-zhang-nonakanodi-li-kong-jian-di-li-kong-jian-qing-bao-ke-xue-gis-tozi-ran-yan-yu-chu- li-nlp-norong-he-hexiang-kete-c41f1d81-1397-4df0-888d-370f0583597e
133. Example: Geocoding Link addresses, facility names, etc. to location 133 https:@Aspeakerdeck.com/sorami/nlp2023 ݴޠॲཧֶձୈճ೥࣍େձ (NLP2023)
134. Example: Geocoding Link addresses, facility names, etc. to location 134 Named Entity Recognition Geotagging (Toponym Extraction) Entity Resolution Geocoding (Toponym Resolution) Entity Linking Geoparsing + = = + Retrieval Resolution Retrieval + Resolution General Geospatial
135. Geography & Language Project 135 https:@Asites.google.com/view/geography-and-language
136. Study Group (online events), Slack, etc. 136 https:@Asites.google.com/view/geography-and-language
137. Illustration generated by ChatGPT At the Intersection of Technology & Liberal Arts
138. Lastly ...
139. Words from Kevin Duh that still remains in my heart 139 ¥Always Ask Questions ¥Just Do It ¥Have Fun My supervisor at NAIST Matsumoto Lab.
140. Words from Me 140 ¥Be Curious ¥Be Grateful ¥Enjoy The Journey!
141. How I Got Here SCONJ PRON VERB ADV NLP, Geospatial, and beyond advmod nsubj advmod @sorami 2025-05-16 Nara Institute of Science and Technology