Last year we explored the potential of LLMs to be used as general purpose entity extraction models, finding that LLMs perform extremely poorly on generalized entity extraction due to their lack of underlying live-updating knowledge graph. In each of those tests, we examined generalized entity extraction over the open domain of person and organization names, in which there is an infinite open-ended universe of possibilities. In contrast, textual geocoding (converting a textual mention of a location like "Paris" into a disambiguated geographic centroid record) offers a much simpler challenge for which LLMs might be far more readily amenable, given the relatively fixed nature of global geographic names. (In reality, geographic names change constantly, but for most traditional geocoding tasks, even if the underlying gazetteer only updates annually, this may be acceptable).
To explore this further, we examine two articles, one about Ukraine and one about India, and test a range of prompts with GPT 3.5, GPT 4.0 and Gemini Pro.
For the Ukraine article that mentions only a handful of country, major city and ADM1 locations that have received extensive global media coverage (and thus should be extremely well represented in the models' training data), GPT 3.5 performs fairly poorly, with highly inconsistent results, hallucination and missing even major cities like Moscow. Even with our most advanced prompt, incorporating step-by-step reasoning and explanation, coherence and counter-hallucination instructions and snippet requests, GPT 3.5 consistently fails to accurately extract all of the locations in the Ukraine article. GPT 4.0 is able to regurgitate ADM1 labels and centroid lat/long coordinates from its training data, but hallucinates and has unstable extraction. Only with the most advanced prompt does GPT 4.0 provide strong results for Ukraine. Gemini Pro experiences high instability and low coherence, changing the list of locations it identifies with each run, excluding major locations, hallucinating locations and triggering a guardrail false positive that prevents output in a third of the tests. GPT 4.0 is the clear winner here, though even its results are not competitive with classical dedicated geocoding algorithms.
The India article was selected to explore how well models perform on the majority of locations across the world that are unlikely to be heavily represented in the English language training data of the three models. It also tests the models' ability to handle local administrative division references like "taluka" and "tehsil". Here, all three models performed exceptionally poorly, as expected. Surprisingly, GPT-4.0 using the most advanced prompts performed the worst of the three models, illustrating that more advanced models don't necessarily yield more accurate results. In this case, GPT-4.0 struggled both with India's unique administrative divisions (a byproduct of OpenAI's failure to properly globalize its models) and with surprisingly more mundane recognition failures. GPT 3.5 performed better than 4.0 in this case due to its simpler and more naive name extraction. In other words, GPT 4.0's more advanced capabilities meant it attempted to perform greater processing on names, which ended up failing due to its training data's bias against Indian toponyms. Gemini Pro performed surprisingly well and vastly exceeded GPT 4.0's performance, but it too performed fairly poorly.
In all, the three models performed better at geocoding than open-ended entity extraction, but still performed abysmally compared with classical dedicated entity extraction systems. Note that we did not test disambiguation capabilities of the models, only their extractive capabilities and thus the abysmal performance here reflects only the first stage of a geocoding system. Like all things LLM, instability, coherence failure, hallucination and bias all combine to render LLMs unsuitable for production geocoding tasks. The much lower performance of GPT 4.0 compared with 3.5 on Indian locations reflects that more advanced models can enforce greater bias and globalization failures that lower than performance and thus more advanced models are not always better. The failure of all three models on Indian locations reinforces the extreme bias of current LLM models and their underlying training data.
As a simple experiment, what if we take this Al Jazeera article about Ukraine and see how GPT 3.5, GPT 4.0 and Gemini Pro geocode it?
GPT 3.5
Let's start with GPT 3.5:
You are a geocoding agent, your job is to read text and compile a list of all geographic locations mentioned within.
Make a list of all of the geographic locations mentioned in the text below.
Output in the following JSON format: "{ city: "City Name", ADM1: ADM1, Country: Country Name, Lat: Latitude, Long: Longitude }":
TEXT:
This yields the following list, which captures the locations and interestingly changes our "city" parameter to "city/region/country", but fails to output JSON:
Here is a list of geographic locations mentioned in the provided text: 1. City: Kyiv, ADM1: N/A, Country: Ukraine, Lat: N/A, Long: N/A 2. Country: Russia, ADM1: N/A, Lat: N/A, Long: N/A 3. Region: Sumy, ADM1: N/A, Country: Ukraine, Lat: N/A, Long: N/A 4. Region: Chernihiv, ADM1: N/A, Country: Ukraine, Lat: N/A, Long: N/A 5. Country: Poland, ADM1: N/A, Lat: N/A, Long: N/A 6. Country: France, ADM1: N/A, Lat: N/A, Long: N/A 7. Country: United Kingdom, ADM1: N/A, Lat: N/A, Long: N/A Please note that some locations do not have specific latitude and longitude coordinates mentioned in the text, so they are marked as "N/A" in the JSON format.
What if we run a second time? This time we get JSON, but it inserts false ADM1s. It also uses the requested "city" field, illustrating the coherence challenges of LLMs where in the first response above it adlibbed a different field for city. This time it also adds the "European Union" as a location:
Here is a list of geographic locations mentioned in the text along with their respective information in JSON format:
1. {"city": "Kyiv", "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
2. {"city": "Moscow", "ADM1": null, "Country": "Russia", "Lat": null, "Long": null}
3. {"city": "Sumy", "ADM1": "northeastern Sumy region", "Country": "Ukraine", "Lat": null, "Long": null}
4. {"city": "Chernihiv", "ADM1": "northern region of Chernihiv", "Country": "Ukraine", "Lat": null, "Long": null}
5. {"city": "Poland", "ADM1": null, "Country": "Poland", "Lat": null, "Long": null}
6. {"city": "France", "ADM1": null, "Country": "France", "Lat": null, "Long": null}
7. {"city": "European Union", "ADM1": null, "Country": "European Union", "Lat": null, "Long": null}
8. {"city": "Britain", "ADM1": null, "Country": "United Kingdom", "Lat": null, "Long": null}
Please note that some locations have limited geographic coordinates available, and some are countries or regions without specific city information mentioned in the text.
Running a third time the results are largely the same:
Here is a list of geographic locations mentioned in the text along with their respective information in JSON format:
1. {"city": "Kyiv", "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
2. {"city": "Moscow", "ADM1": null, "Country": "Russia", "Lat": null, "Long": null}
3. {"city": "Sumy", "ADM1": "northeastern Sumy region", "Country": "Ukraine", "Lat": null, "Long": null}
4. {"city": "Chernihiv", "ADM1": "northern region of Chernihiv", "Country": "Ukraine", "Lat": null, "Long": null}
5. {"city": "Poland", "ADM1": null, "Country": "Poland", "Lat": null, "Long": null}
6. {"city": "France", "ADM1": null, "Country": "France", "Lat": null, "Long": null}
7. {"city": "European Union", "ADM1": null, "Country": "European Union", "Lat": null, "Long": null}
8. {"city": "Britain", "ADM1": null, "Country": "United Kingdom", "Lat": null, "Long": null}
Please note that some locations have limited geographic coordinates available, and some are countries or regions without specific city information mentioned in the text.
How about a fourth time? This time we get an infinite loop – the first time we've seen an LLM infinite loop in an OpenAI product in our testing. The output below is truncated to the first 50 rows, while in reality the results were aborted after continuing indefinitely up to token max length:
Here is a list of geographic locations mentioned in the text you provided, along with their respective information in JSON format:
1. {"city": null, "ADM1": null, "Country": "Russia", "Lat": null, "Long": null}
2. {"city": "Ukraine", "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
3. {"city": "Kyiv", "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
4. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
5. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
6. {"city": null, "ADM1": null, "Country": "Russia", "Lat": null, "Long": null}
7. {"city": "Moscow", "ADM1": null, "Country": "Russia", "Lat": null, "Long": null}
8. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
9. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
10. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
11. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
12. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
13. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
14. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
15. {"city": "Sumy", "ADM1": "northeastern Sumy region", "Country": "Ukraine", "Lat": null, "Long": null}
16. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
17. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
18. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
19. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
20. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
21. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
22. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
23. {"city": "Chernihiv", "ADM1": "northern region of Chernihiv", "Country": "Ukraine", "Lat": null, "Long": null}
24. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
25. {"city": "Poland", "ADM1": null, "Country": "Poland", "Lat": null, "Long": null}
26. {"city": null, "ADM1": null, "Country": "Poland", "Lat": null, "Long": null}
27. {"city": "France", "ADM1": null, "Country": "France", "Lat": null, "Long": null}
28. {"city": "Kyiv", "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
29. {"city": "Kyiv", "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
30. {"city": "Kyiv", "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
31. {"city": "Ukraine", "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
32. {"city": "Russia", "ADM1": null, "Country": "Russia", "Lat": null, "Long": null}
33. {"city": "Ukraine", "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
34. {"city": null, "ADM1": null, "Country": "France", "Lat": null, "Long": null}
35. {"city": "Kyiv", "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
36. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
37. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
38. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
39. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
40. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
41. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
42. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
43. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
44. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
45. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
46. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
47. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
48. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
49. {"city": null, "ADM1": null, "Country": "Ukraine", "Lat": null, "Long": null}
...
What if we modify our prompt slightly to request that all of the results be returned in a single JSON array, rather than in JSON-NL (one object per line) format:
You are a geocoding agent, your job is to read text and compile a list of all geographic locations mentioned within.
Make a list of all of the geographic locations mentioned in the text below.
Output in the following JSON format: "{ city: "City Name", ADM1: ADM1, Country: Country Name, Lat: Latitude, Long: Longitude }", with all of the locations in a single JSON array:
TEXT:
The first time it fails to return JSON:
Here is a list of geographic locations mentioned in the provided text: 1. City: Kyiv, ADM1: N/A, Country: Ukraine, Lat: N/A, Long: N/A 2. Country: Russia, ADM1: N/A, City: N/A, Lat: N/A, Long: N/A 3. Region: Sumy, ADM1: N/A, Country: Ukraine, Lat: N/A, Long: N/A 4. Region: Chernihiv, ADM1: N/A, Country: Ukraine, Lat: N/A, Long: N/A 5. Country: Poland, ADM1: N/A, City: N/A, Lat: N/A, Long: N/A 6. City: Kyiv, ADM1: N/A, Country: Ukraine, Lat: N/A, Long: N/A 7. Country: France, ADM1: N/A, City: N/A, Lat: N/A, Long: N/A 8. Country: Ukraine, ADM1: N/A, City: N/A, Lat: N/A, Long: N/A 9. Country: European Union, ADM1: N/A, City: N/A, Lat: N/A, Long: N/A 10. City: London, ADM1: N/A, Country: United Kingdom, Lat: N/A, Long: N/A Please note that some locations are mentioned without specific latitude (Lat) and longitude (Long) coordinates, and some are mentioned without an Administrative Division 1 (ADM1) designation.
The second time we get JSON:
Here is a list of geographic locations mentioned in the text in JSON format:
```json
[
{
"city": "Kyiv",
"ADM1": "",
"Country": "Ukraine",
"Lat": "",
"Long": ""
},
{
"city": "Sumy",
"ADM1": "northeastern Sumy region",
"Country": "Ukraine",
"Lat": "",
"Long": ""
},
{
"city": "Chernihiv",
"ADM1": "northern region of Chernihiv",
"Country": "Ukraine",
"Lat": "",
"Long": ""
},
{
"city": "Poland",
"ADM1": "",
"Country": "Poland",
"Lat": "",
"Long": ""
},
{
"city": "France",
"ADM1": "",
"Country": "France",
"Lat": "",
"Long": ""
},
{
"city": "Kyiv",
"ADM1": "",
"Country": "Ukraine",
"Lat": "",
"Long": ""
},
{
"city": "Ukraine",
"ADM1": "",
"Country": "Ukraine",
"Lat": "",
"Long": ""
},
{
"city": "France",
"ADM1": "",
"Country": "France",
"Lat": "",
"Long": ""
},
{
"city": "Ukraine",
"ADM1": "",
"Country": "Ukraine",
"Lat": "",
"Long": ""
},
{
"city": "European Union",
"ADM1": "",
"Country": "",
"Lat": "",
"Long": ""
},
{
"city": "Ukraine",
"ADM1": "",
"Country": "Ukraine",
"Lat": "",
"Long": ""
},
{
"city": "Britain",
"ADM1": "",
"Country": "United Kingdom",
"Lat": "",
"Long": ""
}
]
```
Please note that some locations may not have latitude and longitude coordinates provided in the text, so those fields are left empty.
Finally, what if we try a far more elaborate prompt that includes step-by-step reasoning explanation, explicit instruction to include all mentioned locations and none not mentioned and a verification snippet, as well as adjusting our JSON to emphasize that we are looking for all locations rather than just cities:
You are a geocoding agent, your job is to read text and compile a list of all geographic locations mentioned within.
Make a list of all of the geographic locations mentioned in the text below.
Output in the following JSON format: "{ location: "Location Name", ADM1: ADM1, Country: Country Name, Lat: Latitude, Long: Longitude, Explanation: Explanation, Snippet: Snippet }", with all of the locations in a single JSON array.
Do not include any locations not mentioned in the text and make sure to include all locations mentioned in the text. For the Explanation field explain your reasoning for including that location step-by-step.
For the Snippet field include a brief snippet from the text that includes a mention of that location.
TEXT:
This yields the best results yet, though there is an error in the JSON output and it incorrectly labels "NATO" as a geographic location:
I have identified the following geographic locations mentioned in the text:
1. Location: Ukraine
- ADM1: N/A
- Country: Ukraine
- Explanation: Ukraine is mentioned multiple times as the main location of the events described in the text.
- Snippet: "Russia launched dozens of attacks across Ukraine overnight, the Ukrainian air force said..."
2. Location: Kyiv
- ADM1: N/A
- Country: Ukraine
- Explanation: Kyiv is the capital of Ukraine and is mentioned as the diplomatic center where efforts are being made.
- Snippet: "Russia launched dozens of attacks across Ukraine overnight, the Ukrainian air force said, as Kyiv continued its diplomatic efforts..."
3. Location: Moscow
- ADM1: N/A
- Country: Russia
- Explanation: Moscow is mentioned as the location of Russia's defense ministry, which is involved in the events described.
- Snippet: "Russia said it destroyed all targets in a barrage of attacks on facilities producing ammunition and drones in Ukraine. The defence ministry in Moscow said its armed forces “carried out a group strike … against facilities of the Ukrainian military-industrial complex”."
4. Location: Sumy region
- ADM1: N/A
- Country: Ukraine
- Explanation: Sumy region is mentioned as the location where a civilian was wounded.
- Snippet: "No fatalities were reported so far, but Ukrainian authorities said a civilian was wounded in the northeastern Sumy region."
5. Location: Chernihiv
- ADM1: N/A
- Country: Ukraine
- Explanation: Chernihiv is mentioned as the location where a missile created a large crater and caused damage to buildings.
- Snippet: "Police in the northern region of Chernihiv posted a picture of a large crater made by a downed missile. 'As a result of being hit by the debris of an enemy missile, several private homes and non-residential buildings were damaged, one building was practically destroyed,' the police wrote..."
6. Location: Poland
- ADM1: N/A
- Country: Poland
- Explanation: Poland is mentioned as Ukraine's neighbor, and its air defense systems were briefly activated.
- Snippet: "The air force in Ukraine’s neighbor Poland said it had briefly activated air defence systems due to the increased level of threat."
7. Location: France
- ADM1: N/A
- Country: France
- Explanation: France is mentioned as a country whose foreign minister, Stephane Sejourne, arrived in Kyiv, showing support for Ukraine.
- Snippet: "The attacks took place as France’s newly appointed foreign minister, Stephane Sejourne, arrived in Kyiv on Saturday for his first official trip abroad, in a sign of support as the Russian invasion of Ukraine nears its second year."
8. Location: European Union
- ADM1: N/A
- Country: Multiple countries in the EU
- Explanation: The European Union is mentioned as a political entity where legal issues are discussed.
- Snippet: "He added that he would work in the coming days and weeks to fix European Union and bilateral legal issues to help French companies set up more military production facilities in Ukraine..."
9. Location: Britain
- ADM1: N/A
- Country: United Kingdom
- Explanation: Britain is mentioned as a country whose Prime Minister, Rishi Sunak, visited Ukraine and unveiled new military funding for Ukraine.
- Snippet: "The visit follows that of British Prime Minister Rishi Sunak on Friday, where he unveiled new military funding for Ukraine and reassured the country that the West is still providing support."
10. Location: NATO
- ADM1: N/A
- Country: Multiple NATO member countries
- Explanation: NATO is mentioned as an organization that Ukraine hopes to become a member of in the future.
- Snippet: "It will remain in effect until Ukraine acquires its hoped-for NATO membership, Zelenskyy said."
Here is the JSON format list of geographic locations mentioned in the text:
```json
[
{
"location": "Ukraine",
"ADM1": "N/A",
"Country": "Ukraine",
"Lat": "N/A",
"Long": "N/A",
"Explanation": "Ukraine is mentioned multiple times as the main location of the events described in the text.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight, the Ukrainian air force said..."
},
{
"location": "Kyiv",
"ADM1": "N/A",
"Country": "Ukraine",
"Lat": "N/A",
"Long": "N/A",
"Explanation": "Kyiv is the capital of Ukraine and is mentioned as the diplomatic center where efforts are being made.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight, the Ukrainian air force said, as Kyiv continued its diplomatic efforts..."
},
{
"location": "Moscow",
"ADM1": "N/A",
"Country": "Russia",
"Lat": "N/A",
"Long": "N/A",
"Explanation": "Moscow is mentioned as the location of Russia's defense ministry, which is involved in the events described.",
"Snippet": "Russia said it destroyed all targets in a barrage of attacks on facilities producing ammunition and drones in Ukraine. The defence ministry in Moscow said its armed forces “carried out a group strike … against facilities of the Ukrainian military-industrial complex”."
},
{
"location": "Sumy region",
"ADM1": "N/A",
"Country": "Ukraine",
"Lat": "N/A",
"Long": "N/A",
"Explanation": "Sumy region is mentioned as the location where a civilian was wounded.",
"Snippet": "No fatalities were reported so far, but Ukrainian authorities said a civilian was wounded in the northeastern Sumy region."
},
{
"location": "Chernihiv",
"ADM1": "N/A",
"Country": "Ukraine",
"Lat": "N/A",
"Long": "N/A",
"Explanation": "Chernihiv is mentioned as the location where a missile created a large crater and caused damage to buildings.",
"Snippet": "Police in the northern region of Chernihiv posted a picture of a large crater made by a downed missile. 'As a result of being hit by the debris of an enemy missile, several private homes and non-residential buildings were damaged, one building was practically destroyed,' the police wrote..."
},
{
"location": "Poland",
"ADM1": "N/A",
"Country": "Poland",
"Lat": "N/A",
"Long": "N/A",
"Explanation": "Poland is mentioned as Ukraine's neighbor, and its air defense systems were briefly activated.",
"Snippet": "The air force in Ukraine’s neighbor Poland said it had briefly activated air defence systems due to the increased level of threat."
},
{
"location": "France",
"ADM1": "N/A",
"Country": "France",
"Lat": "N/A
",
"Long": "N/A",
"Explanation": "France is mentioned as a country whose foreign minister, Stephane Sejourne, arrived in Kyiv, showing support for Ukraine.",
"Snippet": "The attacks took place as France’s newly appointed foreign minister, Stephane Sejourne, arrived in Kyiv on Saturday for his first official trip abroad, in a sign of support as the Russian invasion of Ukraine nears its second year."
},
{
"location": "European Union",
"ADM1": "N/A",
"Country": "Multiple EU Member States",
"Lat": "N/A",
"Long": "N/A",
"Explanation": "The European Union is mentioned as a political entity where legal issues are discussed.",
"Snippet": "He added that he would work in the coming days and weeks to fix European Union and bilateral legal issues to help French companies set up more military production facilities in Ukraine..."
},
{
"location": "Britain",
"ADM1": "N/A",
"Country": "United Kingdom",
"Lat": "N/A",
"Long": "N/A",
"Explanation": "Britain is mentioned as a country whose Prime Minister, Rishi Sunak, visited Ukraine and unveiled new military funding for Ukraine.",
"Snippet": "The visit follows that of British Prime Minister Rishi Sunak on Friday, where he unveiled new military funding for Ukraine and reassured the country that the West is still providing support."
},
{
"location": "NATO",
"ADM1": "N/A",
"Country": "Multiple NATO Member States",
"Lat": "N/A",
"Long": "N/A",
"Explanation": "NATO is mentioned as an organization that Ukraine hopes to become a member of in the future.",
"Snippet": "It will remain in effect until Ukraine acquires its hoped-for NATO membership, Zelenskyy said."
}
]
```
Please note that some locations do not have latitude and longitude information, as they are political entities or regions mentioned in the text.
Running a second time it incorrectly excludes Moscow and fails to output the JSON:
I have identified the following geographic locations mentioned in the text: 1. Location: Ukraine - ADM1: N/A - Country: Ukraine - Lat: N/A - Long: N/A - Explanation: Ukraine is the main focus of the text, where Russia launched attacks, and various regions within Ukraine are mentioned. - Snippet: "Russia launched dozens of attacks across Ukraine overnight..." 2. Location: Kyiv - ADM1: N/A - Country: Ukraine - Lat: N/A - Long: N/A - Explanation: Kyiv is the capital of Ukraine and is mentioned as the place where diplomatic efforts are being made. - Snippet: "as Kyiv continued its diplomatic efforts to ratchet up support and military aid." 3. Location: Sumy region - ADM1: N/A - Country: Ukraine - Lat: N/A - Long: N/A - Explanation: The Sumy region is mentioned as the location where a civilian was wounded. - Snippet: "No fatalities were reported so far, but Ukrainian authorities said a civilian was wounded in the northeastern Sumy region." 4. Location: Chernihiv - ADM1: N/A - Country: Ukraine - Lat: N/A - Long: N/A - Explanation: Chernihiv is mentioned in the context of a downed missile causing damage to private homes and buildings. - Snippet: "Police in the northern region of Chernihiv posted a picture of a large crater made by a downed missile." 5. Location: Poland - ADM1: N/A - Country: Poland - Lat: N/A - Long: N/A - Explanation: Poland is mentioned as Ukraine's neighbor, and its air defense systems were briefly activated. - Snippet: "The air force in Ukraine’s neighbor Poland said it had briefly activated air defence systems due to the increased level of threat." 6. Location: France - ADM1: N/A - Country: France - Lat: N/A - Long: N/A - Explanation: France is mentioned as the country whose foreign minister, Stephane Sejourne, arrived in Kyiv. - Snippet: "Speaking at a news conference in Kyiv, Ukrainian Foreign Minister Dmytro Kuleba praised Sejourne for not “being intimidated by Russia’s latest large-scale missile attack”" 7. Location: European Union - ADM1: N/A - Country: N/A - Lat: N/A - Long: N/A - Explanation: The European Union is mentioned in the context of working on legal issues and support for Ukraine. - Snippet: "He added that he would work in the coming days and weeks to fix European Union and bilateral legal issues to help French companies set up more military production facilities in Ukraine..." 8. Location: United Kingdom (UK) - ADM1: N/A - Country: United Kingdom (UK) - Lat: N/A - Long: N/A - Explanation: The United Kingdom is mentioned in the context of British Prime Minister Rishi Sunak's visit and military funding for Ukraine. - Snippet: "The visit follows that of British Prime Minister Rishi Sunak on Friday, where he unveiled new military funding for Ukraine and reassured the country that the West is still providing support." Please note that specific latitude and longitude coordinates are not provided in the text, and some locations are mentioned in a broader context.
Let's explicitly request that only the JSON be returned to help guide it:
You are a geocoding agent, your job is to read text and compile a list of all geographic locations mentioned within.
Make a list of all of the geographic locations mentioned in the text below.
Output in the following JSON format: "{ location: "Location Name", ADM1: ADM1, Country: Country Name, Lat: Latitude, Long: Longitude, Explanation: Explanation, Snippet: Snippet }", with all of the locations in a single JSON array.
Do not include any locations not mentioned in the text and make sure to include all locations mentioned in the text. For the Explanation field explain your reasoning for including that location step-by-step.
For the Snippet field include a brief snippet from the text that includes a mention of that location.
Only return the JSON array itself.
TEXT:
This time we get the JSON, but it misses Moscow again:
Here is the list of geographic locations mentioned in the text along with their information in JSON format:
```json
[
{
"location": "Russia",
"ADM1": null,
"Country": "Russia",
"Lat": null,
"Long": null,
"Explanation": "Russia is mentioned as the aggressor in the attacks on Ukraine.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight, the Ukrainian air force said, as Kyiv continued its diplomatic efforts to ratchet up support and military aid."
},
{
"location": "Ukraine",
"ADM1": null,
"Country": "Ukraine",
"Lat": null,
"Long": null,
"Explanation": "Ukraine is the primary focus of the text and is mentioned multiple times.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight, the Ukrainian air force said, as Kyiv continued its diplomatic efforts to ratchet up support and military aid."
},
{
"location": "Kyiv",
"ADM1": null,
"Country": "Ukraine",
"Lat": null,
"Long": null,
"Explanation": "Kyiv is the capital of Ukraine and is mentioned as the location where diplomatic efforts are taking place.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight, the Ukrainian air force said, as Kyiv continued its diplomatic efforts to ratchet up support and military aid."
},
{
"location": "Sumy region",
"ADM1": null,
"Country": "Ukraine",
"Lat": null,
"Long": null,
"Explanation": "Sumy region in Ukraine is mentioned as the location where a civilian was wounded.",
"Snippet": "No fatalities were reported so far, but Ukrainian authorities said a civilian was wounded in the northeastern Sumy region."
},
{
"location": "Chernihiv",
"ADM1": null,
"Country": "Ukraine",
"Lat": null,
"Long": null,
"Explanation": "Chernihiv is mentioned as a northern region where a missile created a large crater.",
"Snippet": "Police in the northern region of Chernihiv posted a picture of a large crater made by a downed missile."
},
{
"location": "Poland",
"ADM1": null,
"Country": "Poland",
"Lat": null,
"Long": null,
"Explanation": "Poland is mentioned as Ukraine's neighbor, and its air defense systems were activated due to the increased threat.",
"Snippet": "The air force in Ukraine’s neighbour Poland said it had briefly activated air defence systems due to the increased level of threat."
},
{
"location": "France",
"ADM1": null,
"Country": "France",
"Lat": null,
"Long": null,
"Explanation": "France is mentioned as a country whose foreign minister visited Kyiv in a show of support.",
"Snippet": "The attacks took place as France’s newly appointed foreign minister, Stephane Sejourne, arrived in Kyiv on Saturday for his first official trip abroad, in a sign of support as the Russian invasion of Ukraine nears its second year."
},
{
"location": "Britain",
"ADM1": null,
"Country": "United Kingdom",
"Lat": null,
"Long": null,
"Explanation": "Britain is mentioned as the home country of Prime Minister Rishi Sunak, who visited Ukraine and provided military funding.",
"Snippet": "The visit follows that of British Prime Minister Rishi Sunak on Friday, where he unveiled new military funding for Ukraine and reassured the country that the West is still providing support."
},
{
"location": "Ukraine",
"ADM1": null,
"Country": "Ukraine",
"Lat": null,
"Long": null,
"Explanation": "Ukraine is mentioned again in the context of the bilateral security agreement with the UK.",
"Snippet": "Sunak and Ukrainian President Volodymyr Zelenskyy signed a bilateral security agreement for the next 10 years. It will remain in effect until Ukraine acquires its hoped-for NATO membership, Zelenskyy said."
}
]
```
This list includes all the geographic locations mentioned in the text, along with explanations for their inclusion and snippets from the text where they are mentioned.
Running a second time we get the JSON and this time it includes Moscow:
Here is the list of geographic locations mentioned in the text along with their details in JSON format:
```json
[
{
"location": "Ukraine",
"ADM1": null,
"Country": "Ukraine",
"Lat": null,
"Long": null,
"Explanation": "Ukraine is mentioned as the primary location where the attacks by Russia occurred.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight..."
},
{
"location": "Kyiv",
"ADM1": null,
"Country": "Ukraine",
"Lat": null,
"Long": null,
"Explanation": "Kyiv is mentioned as the capital of Ukraine, where diplomatic efforts are taking place.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight, the Ukrainian air force said, as Kyiv continued its diplomatic efforts..."
},
{
"location": "Moscow",
"ADM1": null,
"Country": "Russia",
"Lat": null,
"Long": null,
"Explanation": "Moscow is mentioned as the location of the Russian defense ministry, which reported on the attacks.",
"Snippet": "Russia said it destroyed all targets in a barrage of attacks on facilities producing ammunition and drones in Ukraine. The defence ministry in Moscow said..."
},
{
"location": "Sumy region",
"ADM1": null,
"Country": "Ukraine",
"Lat": null,
"Long": null,
"Explanation": "Sumy region is mentioned as the location where a civilian was wounded due to the attacks.",
"Snippet": "No fatalities were reported so far, but Ukrainian authorities said a civilian was wounded in the northeastern Sumy region."
},
{
"location": "Chernihiv",
"ADM1": null,
"Country": "Ukraine",
"Lat": null,
"Long": null,
"Explanation": "Chernihiv is mentioned as a northern region where a missile hit and caused damage.",
"Snippet": "Police in the northern region of Chernihiv posted a picture of a large crater made by a downed missile."
},
{
"location": "Poland",
"ADM1": null,
"Country": "Poland",
"Lat": null,
"Long": null,
"Explanation": "Poland is mentioned as Ukraine's neighbor, where the air defense systems were briefly activated due to the threat.",
"Snippet": "The air force in Ukraine’s neighbour Poland said it had briefly activated air defence systems due to the increased level of threat."
},
{
"location": "France",
"ADM1": null,
"Country": "France",
"Lat": null,
"Long": null,
"Explanation": "France is mentioned as the country whose foreign minister visited Kyiv in support of Ukraine.",
"Snippet": "The attacks took place as France’s newly appointed foreign minister, Stephane Sejourne, arrived in Kyiv on Saturday for his first official trip abroad..."
},
{
"location": "Britain",
"ADM1": null,
"Country": "United Kingdom",
"Lat": null,
"Long": null,
"Explanation": "Britain is mentioned as the country whose Prime Minister visited Ukraine and provided military funding.",
"Snippet": "The visit follows that of British Prime Minister Rishi Sunak on Friday, where he unveiled new military funding for Ukraine and reassured the country that the West is still providing support."
}
]
```
Explanation field provides the reasoning for including each location, and Snippet provides a brief text excerpt mentioning that location.
Running a third time it misses Moscow again:
Here is a JSON array of all the geographic locations mentioned in the text along with explanations and snippets:
```json
[
{
"location": "Russia",
"ADM1": null,
"Country": "Russia",
"Lat": null,
"Long": null,
"Explanation": "Russia is mentioned as the aggressor in the attacks on Ukraine.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight, the Ukrainian air force said, as Kyiv continued its diplomatic efforts to ratchet up support and military aid."
},
{
"location": "Ukraine",
"ADM1": null,
"Country": "Ukraine",
"Lat": null,
"Long": null,
"Explanation": "Ukraine is the primary focus of the text, as it is the target of the attacks and the location where various events are happening.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight, the Ukrainian air force said..."
},
{
"location": "Kyiv",
"ADM1": null,
"Country": "Ukraine",
"Lat": null,
"Long": null,
"Explanation": "Kyiv is the capital of Ukraine and is mentioned as the location where diplomatic efforts are taking place.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight, the Ukrainian air force said, as Kyiv continued its diplomatic efforts to ratchet up support and military aid."
},
{
"location": "Sumy region",
"ADM1": null,
"Country": "Ukraine",
"Lat": null,
"Long": null,
"Explanation": "Sumy region is mentioned as the location where a civilian was wounded due to the attacks.",
"Snippet": "No fatalities were reported so far, but Ukrainian authorities said a civilian was wounded in the northeastern Sumy region."
},
{
"location": "Chernihiv",
"ADM1": null,
"Country": "Ukraine",
"Lat": null,
"Long": null,
"Explanation": "Chernihiv is mentioned as the location where a missile caused damage to private homes and buildings.",
"Snippet": "Police in the northern region of Chernihiv posted a picture of a large crater made by a downed missile. “As a result of being hit by the debris of an enemy missile..."
},
{
"location": "Poland",
"ADM1": null,
"Country": "Poland",
"Lat": null,
"Long": null,
"Explanation": "Poland is mentioned as Ukraine's neighbor, and its air defense systems were activated due to the threat.",
"Snippet": "The air force in Ukraine’s neighbor Poland said it had briefly activated air defence systems due to the increased level of threat."
},
{
"location": "France",
"ADM1": null,
"Country": "France",
"Lat": null,
"Long": null,
"Explanation": "France is mentioned as a foreign minister from France, Stephane Sejourne, arrives in Kyiv to show support.",
"Snippet": "The attacks took place as France’s newly appointed foreign minister, Stephane Sejourne, arrived in Kyiv on Saturday for his first official trip abroad..."
},
{
"location": "Britain",
"ADM1": null,
"Country": "United Kingdom",
"Lat": null,
"Long": null,
"Explanation": "Britain is mentioned as British Prime Minister Rishi Sunak visited Ukraine and provided military funding.",
"Snippet": "The visit follows that of British Prime Minister Rishi Sunak on Friday, where he unveiled new military funding for Ukraine and reassured the country that the West is still providing support."
}
]
```
These are all the geographic locations mentioned in the text, along with their explanations and snippets from the text.
Finally, let's try the same prompt, but on a different article – this one from Indian Express that mentions a variety of locations across India. Below are the main locations mentioned in the article as might be desirable for a geocoder to identify:
Bengaluru Bilari tehsil Cauvery Chandausi Dusi Foundation for Agrarian Studies Gangaikondan Iruvelpattu Jawaharlal Nehru University Kilvelur taluka Kolkata Lower Cauvery Delta Mettur dam Moradabad Moradabad district Nagapattinam district New Delhi North Arcot Palakurichi Palanpur Ramanathapuram South Arcot South India Tamil Nadu Thanjavur Tirunelveli University of Madras University of Oxford Uttar Pradesh Vadamalaipuram Indian Statistical Institute
Using our most advanced prompt, GPT 3.5 identifies only 17 of the 31 locations (55% accuracy). The indented locations are the ones it correctly identified.
Bengaluru Bilari tehsil Cauvery Lower Cauvery Delta Chandausi Dusi Foundation for Agrarian Studies Gangaikondan Iruvelpattu Jawaharlal Nehru University Kilvelur taluka Kolkata Mettur dam Moradabad Moradabad district Nagapattinam district New Delhi North Arcot Palakurichi Palanpur Ramanathapuram South Arcot South India Tamil Nadu Thanjavur Tirunelveli University of Madras University of Oxford Uttar Pradesh Vadamalaipuram Indian Statistical Institute
And here is the full JSON output:
Here is a JSON array containing all the geographic locations mentioned in the provided text along with explanations and snippets:
```json
[
{
"location": "University of Madras",
"ADM1": "",
"Country": "",
"Lat": "",
"Long": "",
"Explanation": "University mentioned in the text.",
"Snippet": "In 1916-17, Gilbert Slater, professor of economics at the University of Madras, surveyed five villages..."
},
{
"location": "Tamil Nadu",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "The text refers to present-day Tamil Nadu.",
"Snippet": "surveyed five villages in present-day Tamil Nadu through his students."
},
{
"location": "North Arcot",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "North Arcot is mentioned as one of the 'Slater villages'.",
"Snippet": "The “Slater villages” — Dusi (North Arcot), Iruvelpattu (South Arcot),..."
},
{
"location": "South Arcot",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "South Arcot is mentioned as one of the 'Slater villages'.",
"Snippet": "The “Slater villages” — Dusi (North Arcot), Iruvelpattu (South Arcot),..."
},
{
"location": "Thanjavur",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "Thanjavur is mentioned as one of the 'Slater villages'.",
"Snippet": "The “Slater villages” — Dusi (North Arcot), Iruvelpattu (South Arcot),..."
},
{
"location": "Ramanathapuram",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "Ramanathapuram is mentioned as one of the 'Slater villages'.",
"Snippet": "The “Slater villages” — Dusi (North Arcot), Iruvelpattu (South Arcot),..."
},
{
"location": "Tirunelveli",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "Tirunelveli is mentioned as one of the 'Slater villages'.",
"Snippet": "The “Slater villages” — Dusi (North Arcot), Iruvelpattu (South Arcot),..."
},
{
"location": "Kilvelur taluka",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "Kilvelur taluka in Nagapattinam district is mentioned in the text.",
"Snippet": "One of the five — Palakurichi, now in Kilvelur taluka of Nagapattinam district — went on to be studied again..."
},
{
"location": "Uttar Pradesh",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "Uttar Pradesh is mentioned as the location of Palanpur village.",
"Snippet": "The only comparable example is Palanpur, a village in Bilari tehsil of Uttar Pradesh’s Moradabad district..."
},
{
"location": "Moradabad",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "Moradabad district is mentioned as the location of Palanpur village.",
"Snippet": "The only comparable example is Palanpur, a village in Bilari tehsil of Uttar Pradesh’s Moradabad district..."
},
{
"location": "Lower Cauvery Delta",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "Lower Cauvery Delta is mentioned as the location of Palakurichi village.",
"Snippet": "Palakurichi is located in the lower Cauvery delta of the old Thanjavur region."
},
{
"location": "Mettur dam",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "Mettur dam is mentioned as part of the history of Palakurichi's agricultural progress.",
"Snippet": "The zenith of its agricultural progress was reached with the Mettur dam’s construction in 1934..."
},
{
"location": "Cauvery",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "Cauvery River is mentioned as the source of irrigation for Palakurichi region.",
"Snippet": "Being irrigated by a network of canals from the Cauvery..."
},
{
"location": "Bilari tehsil",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "Bilari tehsil is mentioned as the location of Palanpur village.",
"Snippet": "The only comparable example is Palanpur, a village in Bilari tehsil of Uttar Pradesh’s Moradabad district..."
},
{
"location": "Chandausi",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "Chandausi is mentioned as one of the towns where Dalits have found non-farm employment opportunities.",
"Snippet": "Dalits have also registered some improvement, although not from access to land as much as expansion of non-farm employment opportunities, including in neighbouring towns such as Chandausi and Moradabad."
},
{
"location": "Palakurichi village",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "Palakurichi village is the primary location discussed in the text.",
"Snippet": "The results of the 2019 survey of Palakurichi village have been published in a recent volume..."
},
{
"location": "Palanpur",
"ADM1": "",
"Country": "India",
"Lat": "",
"Long": "",
"Explanation": "Palanpur is mentioned as a comparable example to Palakurichi village.",
"Snippet": "The only comparable example is Palanpur, a village in Bilari tehsil of Uttar Pradesh’s Moradabad district..."
}
]
```
These are all the geographic locations mentioned in the text along with explanations and snippets.
Running a second time there is a fatal JSON error and it identifies only 15 of the 31 locations (48% accuracy). Worse, it adds "Cauvery Delta" in addition to "Lower Cauvery Delta", "Delhi" instead of "New Delhi" and "Indian Statistical Institute, Kolkata" instead of breaking it into two locations. Considering these three false entries, the accuracy rate drops to 15 / 34 = 44%.
Bengaluru Bilari tehsil Cauvery Chandausi Dusi Gangaikondan Iruvelpattu Jawaharlal Nehru University Kilvelur taluka Kolkata Lower Cauvery Delta Mettur dam Moradabad Moradabad district Nagapattinam district New Delhi North Arcot Palakurichi Palanpur Ramanathapuram South Arcot South India Tamil Nadu Thanjavur Tirunelveli University of Madras University of Oxford Uttar Pradesh Vadamalaipuram Indian Statistical Institute Cauvery Delta Indian Statistical Institute, Kolkata Delhi
And here is the full JSON:
Here is the list of geographic locations mentioned in the text along with their corresponding information:
```json
[
{
"location": "University of Madras",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": null,
"Long": null,
"Explanation": "University of Madras is mentioned as the institution where Gilbert Slater, the professor of economics, conducted a survey.",
"Snippet": "In 1916-17, Gilbert Slater, professor of economics at the University of Madras, surveyed five villages in present-day Tamil Nadu through his students."
},
{
"location": "Tamil Nadu",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": null,
"Long": null,
"Explanation": "Tamil Nadu is mentioned as the present-day state where the surveyed villages are located.",
"Snippet": "In 1916-17, Gilbert Slater, professor of economics at the University of Madras, surveyed five villages in present-day Tamil Nadu through his students."
},
{
"location": "North Arcot",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": null,
"Long": null,
"Explanation": "North Arcot is mentioned as one of the 'Slater villages' surveyed in Tamil Nadu.",
"Snippet": "The “Slater villages” — Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) — were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "South Arcot",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": null,
"Long": null,
"Explanation": "South Arcot is mentioned as one of the 'Slater villages' surveyed in Tamil Nadu.",
"Snippet": "The “Slater villages” — Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) — were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "Thanjavur",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": null,
"Long": null,
"Explanation": "Thanjavur is mentioned as one of the 'Slater villages' surveyed in Tamil Nadu.",
"Snippet": "The “Slater villages” — Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) — were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "Ramanathapuram",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": null,
"Long": null,
"Explanation": "Ramanathapuram is mentioned as one of the 'Slater villages' surveyed in Tamil Nadu.",
"Snippet": "The “Slater villages” — Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) — were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "Tirunelveli",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": null,
"Long": null,
"Explanation": "Tirunelveli is mentioned as one of the 'Slater villages' surveyed in Tamil Nadu.",
"Snippet": "The “Slater villages” — Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) — were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "Kilvelur taluka",
"ADM1": "Nagapattinam",
"Country": "India",
"Lat": null,
"Long": null,
"Explanation": "Kilvelur taluka is mentioned as the current location of Palakurichi, one of the 'Slater villages' in Nagapattinam district.",
"Snippet": "One of the five — Palakurichi, now in Kilvelur taluka of Nagapattinam district — went on to be studied again in 1964 (by Margaret Haswell from the University of Oxford), 1983 (by S Guhan of the Madras Institute of Development Studies), 2004 (by V Surjit, a PhD student at the Indian Statistical Institute, Kolkata), and 2019 (by the Foundation for Agrarian Studies, Bengaluru)."
},
{
"location": "Bilari tehsil",
"ADM1": "Moradabad",
"Country": "India",
"Lat": null,
"Long": null,
"Explanation": "Bilari tehsil is mentioned as the location of Palanpur, a village in Uttar Pradesh's Moradabad district.",
"Snippet": "The only comparable example is Palanpur, a village in Bilari tehsil of Uttar Pradesh’s Moradabad district that has been surveyed every decade since Independence — in 1957-58, 1962-63, 1974-75, 1983-84, 1993, 2008-09, 2015 and 2022."
},
{
"location": "Uttar Pradesh",
"ADM1": "Uttar Pradesh",
"Country": "India",
"Lat": null,
"Long": null,
"Explanation": "Uttar Pradesh is mentioned as the state where Palanpur is located.",
"Snippet": "The only comparable example is Palanpur, a village in Bilari tehsil of Uttar Pradesh’s Moradabad district that has been surveyed every decade since Independence — in 1957-58, 1962-63, 1974-75, 1983-84, 1993, 2008-09, 2015 and 2022."
},
{
"location": "Cauvery delta",
"ADM1": null,
"Country": null,
"Lat": null,
"Long": null,
"Explanation": "Cauvery delta is mentioned as the location where Palakurichi is located.",
"Snippet": "Palakurichi is located in the lower Cauvery delta of the old Thanjavur region."
},
{
"location": "Mettur dam",
"ADM1": null,
"Country": null,
"Lat": null,
"Long": null,
"Explanation": "Mettur dam is mentioned as a significant location in the history of agriculture in the region.",
"Snippet": "The zenith of its agricultural progress was reached with the Mettur dam’s construction in 1934 and the Green Revolution — the introduction of high-yielding short-duration rice varieties that made cultivation of two crops in a single year possible —from the mid-1960s through the 1980s."
},
{
"location": "Indian Statistical Institute, Kolkata",
"ADM1": "West Bengal",
"Country": "India",
"Lat": null,
"Long": null,
"Explanation": "Indian Statistical Institute, Kolkata, is mentioned as the institution where V Surjit, a PhD student, conducted a survey.",
"Snippet": "One of the five — Palakurichi, now in Kilvelur taluka of Nagapattinam district — went on to be studied again in 1964 (by Margaret Haswell from the University of Oxford), 1983 (by S Guhan of the Madras Institute of Development Studies), 2004 (by V Surjit, a PhD student at the Indian Statistical Institute, Kolkata), and 2019 (by the Foundation for Agrarian Studies, Bengaluru)."
},
{
"location": "Delhi",
"ADM1": "Delhi",
"Country": "India",
"Lat": null,
"Long": null,
"Explanation": "Delhi is mentioned as the location where the book 'Economic Change in the Lower Cauvery Delta: A Study of Palakurichi and Venmani' was published.",
"Snippet": "The results of the 2019 survey of Palakurichi village have been published in a recent volume (Economic Change in the Lower Cauvery Delta: A Study of Palakurichi and Venmani, edited by Madhura Swaminathan, V Surjit and VK Ramachandran, New Delhi: Tulika Books)."
},
{
"location": "Lower Cauvery Delta",
"ADM1": null,
"Country": null,
"Lat": null,
"Long": null,
"Explanation": "Lower Cauvery Delta is mentioned as the geographical area where Palakurichi is located.",
"Snippet": "Palakurichi is located in the lower Cauvery delta of the old Thanjavur region."
},
{
"location": "Moradabad",
"ADM1": "Uttar Pradesh",
"Country": "India",
"Lat": null,
"Long": null,
"Explanation": "Moradabad is mentioned as the district where Palanpur is located.",
"Snippet": "The only comparable example is Palanpur, a village in Bilari tehsil of Uttar Pradesh’s Moradabad district that has been surveyed every decade since Independence — in 1957-58, 1962-63, 1974-75, 1983-84, 1993, 2008-09, 2015 and 2022."
},
{
"location": "Chandausi",
"ADM1": "Uttar Pradesh",
"Country": "India",
"Lat": null,
"Long": null,
"Explanation": "Chandausi is mentioned as one of the neighboring towns where Dalits have found non-farm employment opportunities.",
"Snippet": "Dalits have also registered some improvement, although not from access to land as much as expansion of non-farm employment opportunities, including in neighboring towns such as Chandausi and Moradabad."
},
{
"location": "Jawaharlal Nehru University",
"ADM1": "Delhi",
"Country": "India",
"Lat": null,
"Long": null,
"Explanation": "Jawaharlal Nehru University is mentioned as the institution where Himanshu, an associate professor, works.",
"Snippet": "“Our later visits suggest non-farm jobs are returning again. We plan to do a new survey that will be comprehensive,” said Himanshu, associate professor at the Jawaharlal Nehru University."
}
]
```
This JSON list includes all the geographic locations mentioned in the text along with their respective ADM1, Country, Latitude, Longitude, Explanation, and a snippet from the text that mentions each location.
GPT 4.0
Let's try the original prompt with GPT 4.0:
You are a geocoding agent, your job is to read text and compile a list of all geographic locations mentioned within.
Make a list of all of the geographic locations mentioned in the text below.
Output in the following JSON format: "{ city: "City Name", ADM1: ADM1, Country: Country Name, Lat: Latitude, Long: Longitude }":
TEXT:
This time we get JSON output that matches our requested format and GPT 4.0 has even estimated values for each field, including centroid lat/long. It even seems to provide accurate centroid values for each. Interestingly, the European Union is not considered a geographic location (it isn't filtering just to cities, as countries appear below):
Based on the text provided, here are the geographic locations mentioned along with their respective details in the requested JSON format:
1.
```json
{
"city": "Kyiv",
"ADM1": "Kyiv City",
"Country": "Ukraine",
"Lat": "50.4501",
"Long": "30.5234"
}
```
2.
```json
{
"city": "",
"ADM1": "Sumy region",
"Country": "Ukraine",
"Lat": "50.9077",
"Long": "34.7981"
}
```
3.
```json
{
"city": "",
"ADM1": "Chernihiv region",
"Country": "Ukraine",
"Lat": "51.4982",
"Long": "31.2893"
}
```
4.
```json
{
"city": "",
"ADM1": "",
"Country": "Russia",
"Lat": "61.5240",
"Long": "105.3188"
}
```
5.
```json
{
"city": "",
"ADM1": "",
"Country": "Poland",
"Lat": "51.9194",
"Long": "19.1451"
}
```
6.
```json
{
"city": "Moscow",
"ADM1": "Moscow",
"Country": "Russia",
"Lat": "55.7558",
"Long": "37.6176"
}
```
7.
```json
{
"city": "",
"ADM1": "",
"Country": "France",
"Lat": "46.2276",
"Long": "2.2137"
}
```
8.
```json
{
"city": "",
"ADM1": "",
"Country": "United Kingdom",
"Lat": "55.3781",
"Long": "-3.4360"
}
```
Note: The latitude and longitude coordinates provided are approximate central locations for the regions or countries mentioned, as specific coordinates for every region are not provided in the text.
The second time we run it, GPT 4.0 on its own grouped the results into a single JSON array without us requesting it:
```json
[
{
"city": "Kyiv",
"ADM1": "Kyiv City",
"Country": "Ukraine",
"Lat": "50.4501",
"Long": "30.5234"
},
{
"city": "Moscow",
"ADM1": "Moscow",
"Country": "Russia",
"Lat": "55.7558",
"Long": "37.6176"
},
{
"city": "Sumy",
"ADM1": "Sumy Oblast",
"Country": "Ukraine",
"Lat": "50.9077",
"Long": "34.7981"
},
{
"city": "Chernihiv",
"ADM1": "Chernihiv Oblast",
"Country": "Ukraine",
"Lat": "51.4982",
"Long": "31.2893"
},
{
"city": "",
"ADM1": "",
"Country": "Poland",
"Lat": "52.0693",
"Long": "19.4803"
},
{
"city": "",
"ADM1": "",
"Country": "France",
"Lat": "46.6034",
"Long": "1.8883"
},
{
"city": "",
"ADM1": "",
"Country": "United Kingdom",
"Lat": "55.3781",
"Long": "-3.4360"
}
]
```
Let's explicitly request the array to ensure we get consistent results rather than alternating between JSON-NL with interspersed text and an array:
You are a geocoding agent, your job is to read text and compile a list of all geographic locations mentioned within.
Make a list of all of the geographic locations mentioned in the text below.
Output in the following JSON format: "{ city: "City Name", ADM1: ADM1, Country: Country Name, Lat: Latitude, Long: Longitude }", with all of the locations in a single JSON array::
TEXT:
This time we get truncated results, with GPT 4.0 excluding all of the country-level entries. Simply adding an extra instruction about grouping the results into array caused GPT 4.0 to increase its coherence to only include city mentions and exclude country mentions to conform with the JSON format that explicitly names the first field as "city". This demonstrates the brittleness of LLMs: a single instruction change in one place (output format) can cause vast behavioral changes elsewhere (definition of "location" to return):
```json
[
{ "city": "Kyiv", "ADM1": "Kyiv City", "Country": "Ukraine", "Lat": 50.4501, "Long": 30.5234 },
{ "city": "Moscow", "ADM1": "Moscow", "Country": "Russia", "Lat": 55.7558, "Long": 37.6173 },
{ "city": "Sumy", "ADM1": "Sumy Oblast", "Country": "Ukraine", "Lat": 50.9077, "Long": 34.7981 },
{ "city": "Chernihiv", "ADM1": "Chernihiv Oblast", "Country": "Ukraine", "Lat": 51.4982, "Long": 31.2893 }
]
```
What if we run a second time? Once again we get only city mentions, except this time it adds a single country mention. This is an even worse scenario in that a downstream application would see the presence of a country-level match and assume the output includes both city and country-level matches, whereas above the application would have to weigh whether the LLM returned only city-level matches or whether the text did not contain any country-level matches. An application whose developer tested it many times and only received city-level matches would now be led astray to consider Poland a city-level match:
Based on the text provided, here is a list of all the geographic locations mentioned, formatted in the requested JSON array:
```json
[
{ "city": "Kyiv", "ADM1": "Kyiv City", "Country": "Ukraine", "Lat": "50.4501", "Long": "30.5234" },
{ "city": "Moscow", "ADM1": "Moscow", "Country": "Russia", "Lat": "55.7558", "Long": "37.6176" },
{ "city": "", "ADM1": "Sumy", "Country": "Ukraine", "Lat": "50.9077", "Long": "34.7981" },
{ "city": "", "ADM1": "Chernihiv", "Country": "Ukraine", "Lat": "51.4982", "Long": "31.2893" },
{ "city": "", "ADM1": "", "Country": "Poland", "Lat": "51.9194", "Long": "19.1451" }
]
```
Note: Some locations are regions or provinces rather than specific cities, and for these, the city field is left blank. Additionally, latitude and longitude coordinates are approximate central coordinates for these regions or cities.
What if we run a third time? This time we get a worst-case scenario: GPT 4.0 hallucinates that Warsaw was mentioned in the text:
```json
[
{ "city": "Kyiv", "ADM1": "Kyiv City", "Country": "Ukraine", "Lat": 50.4501, "Long": 30.5234 },
{ "city": "Moscow", "ADM1": "Moscow", "Country": "Russia", "Lat": 55.7558, "Long": 37.6173 },
{ "city": "Sumy", "ADM1": "Sumy Oblast", "Country": "Ukraine", "Lat": 50.9077, "Long": 34.7981 },
{ "city": "Chernihiv", "ADM1": "Chernihiv Oblast", "Country": "Ukraine", "Lat": 51.4982, "Long": 31.2893 },
{ "city": "Warsaw", "ADM1": "Masovian Voivodeship", "Country": "Poland", "Lat": 52.2297, "Long": 21.0122 }
]
```
A fourth time we go back to getting countries as well as cities, similar to our original pre-array instruction. This suggests that it was not the addition of the instruction that caused the behavioral change, but rather the inherent instability of LLMs:
Based on the text provided, here's a list of the geographic locations mentioned, formatted in the requested JSON style:
```json
[
{ "city": "Kyiv", "ADM1": "Kyiv City", "Country": "Ukraine", "Lat": 50.4501, "Long": 30.5234 },
{ "city": "Moscow", "ADM1": "Moscow", "Country": "Russia", "Lat": 55.7558, "Long": 37.6173 },
{ "city": "", "ADM1": "Sumy", "Country": "Ukraine", "Lat": 50.9077, "Long": 34.7981 },
{ "city": "", "ADM1": "Chernihiv", "Country": "Ukraine", "Lat": 51.4982, "Long": 31.2893 },
{ "city": "", "ADM1": "", "Country": "Poland", "Lat": 51.9194, "Long": 19.1451 },
{ "city": "", "ADM1": "", "Country": "France", "Lat": 46.2276, "Long": 2.2137 },
{ "city": "", "ADM1": "", "Country": "United Kingdom", "Lat": 55.3781, "Long": -3.4360 }
]
```
Note that some locations like Sumy and Chernihiv are mentioned without specifying the city, so only the region (ADM1) and country are provided. Additionally, for countries like Poland, France, and the United Kingdom, specific cities or regions are not mentioned, so only the country's general latitude and longitude are given.
A fifth time we're back to cities only:
```json
[
{ "city": "Kyiv", "ADM1": "Kyiv City", "Country": "Ukraine", "Lat": 50.4501, "Long": 30.5234 },
{ "city": "Moscow", "ADM1": "Moscow", "Country": "Russia", "Lat": 55.7558, "Long": 37.6173 },
{ "city": "Sumy", "ADM1": "Sumy Oblast", "Country": "Ukraine", "Lat": 50.9077, "Long": 34.7981 },
{ "city": "Chernihiv", "ADM1": "Chernihiv Oblast", "Country": "Ukraine", "Lat": 51.4982, "Long": 31.2893 }
]
```
This JSON array lists the geographic locations mentioned in the provided text, including their administrative divisions (ADM1), countries, and coordinates (latitude and longitude).
A sixth time we're back to hallucination, with GPT 4.0 falsely claiming that Paris and London were mentioned. Thus, in two our of our six runs (33%) we received hallucinated city mentions:
Based on the text provided, here is a list of the geographic locations mentioned, formatted as requested:
```json
[
{ "city": "Kyiv", "ADM1": "Kyiv", "Country": "Ukraine", "Lat": 50.4501, "Long": 30.5234 },
{ "city": "Moscow", "ADM1": "Moscow", "Country": "Russia", "Lat": 55.7558, "Long": 37.6173 },
{ "city": "Sumy", "ADM1": "Sumy Oblast", "Country": "Ukraine", "Lat": 50.9077, "Long": 34.7981 },
{ "city": "Chernihiv", "ADM1": "Chernihiv Oblast", "Country": "Ukraine", "Lat": 51.4982, "Long": 31.2893 },
{ "city": "Paris", "ADM1": "Île-de-France", "Country": "France", "Lat": 48.8566, "Long": 2.3522 },
{ "city": "London", "ADM1": "England", "Country": "United Kingdom", "Lat": 51.5074, "Long": -0.1278 }
]
```
This list includes cities and corresponding administrative divisions (ADM1), countries, and their latitude and longitude coordinates.
Let's try our most advanced prompt:
You are a geocoding agent, your job is to read text and compile a list of all geographic locations mentioned within.
Make a list of all of the geographic locations mentioned in the text below.
Output in the following JSON format: "{ location: "Location Name", ADM1: ADM1, Country: Country Name, Lat: Latitude, Long: Longitude, Explanation: Explanation, Snippet: Snippet }", with all of the locations in a single JSON array.
Do not include any locations not mentioned in the text and make sure to include all locations mentioned in the text. For the Explanation field explain your reasoning for including that location step-by-step.
For the Snippet field include a brief snippet from the text that includes a mention of that location.
Only return the JSON array itself.
TEXT:
Here we get our best results yet, with no major errors:
```json
[
{
"location": "Ukraine",
"ADM1": "",
"Country": "Ukraine",
"Lat": 48.3794,
"Long": 31.1656,
"Explanation": "Ukraine is mentioned as the primary location where the attacks and military aid efforts are occurring. It's the central focus of the text.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight"
},
{
"location": "Kyiv",
"ADM1": "Kyiv City",
"Country": "Ukraine",
"Lat": 50.4501,
"Long": 30.5234,
"Explanation": "Kyiv is specifically mentioned as the place where diplomatic efforts to gain support are happening, and where the Ukrainian Foreign Minister and the French foreign minister held a news conference.",
"Snippet": "as Kyiv continued its diplomatic efforts"
},
{
"location": "Russia",
"ADM1": "",
"Country": "Russia",
"Lat": 61.5240,
"Long": 105.3188,
"Explanation": "Russia is included as it's mentioned as the aggressor in the conflict and the source of the attacks.",
"Snippet": "Russia said it destroyed all targets in a barrage of attacks"
},
{
"location": "Moscow",
"ADM1": "Moscow",
"Country": "Russia",
"Lat": 55.7558,
"Long": 37.6173,
"Explanation": "Moscow is mentioned as the location of the Russian defence ministry, which provided statements about the attacks.",
"Snippet": "The defence ministry in Moscow said"
},
{
"location": "Sumy",
"ADM1": "Sumy Oblast",
"Country": "Ukraine",
"Lat": 50.9077,
"Long": 34.7981,
"Explanation": "Sumy region is mentioned as a specific location where a civilian was wounded as a result of the attacks.",
"Snippet": "Ukrainian authorities said a civilian was wounded in the northeastern Sumy region."
},
{
"location": "Chernihiv",
"ADM1": "Chernihiv Oblast",
"Country": "Ukraine",
"Lat": 51.4982,
"Long": 31.2893,
"Explanation": "Chernihiv is mentioned in the context of a missile attack, with a picture of a crater from a downed missile being posted by the police.",
"Snippet": "Police in the northern region of Chernihiv posted a picture"
},
{
"location": "Poland",
"ADM1": "",
"Country": "Poland",
"Lat": 51.9194,
"Long": 19.1451,
"Explanation": "Poland is mentioned as Ukraine's neighbour and as having activated its air defence systems in response to the increased threat level.",
"Snippet": "The air force in Ukraine's neighbour Poland said it had briefly activated air defence systems"
},
{
"location": "France",
"ADM1": "",
"Country": "France",
"Lat": 46.2276,
"Long": 2.2137,
"Explanation": "France is mentioned in the context of its foreign policy priorities and the visit of its foreign minister to Ukraine.",
"Snippet": "Speaking at a news conference in Kyiv, Ukrainian Foreign Minister Dmytro Kuleba praised Sejourne for not being intimidated by Russia's latest large-scale missile attack"
},
{
"location": "United Kingdom",
"ADM1": "",
"Country": "United Kingdom",
"Lat": 55.3781,
"Long": -3.4360,
"Explanation": "The United Kingdom is implicated in the context of its Prime Minister's visit to Ukraine and the signing of a bilateral security agreement.",
"Snippet": "The visit follows that of British Prime Minister Rishi Sunak on Friday"
}
]
```
A second time we get similarly strong results:
```json
[
{
"location": "Ukraine",
"ADM1": "",
"Country": "Ukraine",
"Lat": 48.3794,
"Long": 31.1656,
"Explanation": "Ukraine is mentioned as the primary location of the Russian attacks and the context of the entire passage.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight"
},
{
"location": "Kyiv",
"ADM1": "Kyiv",
"Country": "Ukraine",
"Lat": 50.4501,
"Long": 30.5234,
"Explanation": "Kyiv is mentioned as the place where Ukrainian diplomatic efforts are being coordinated.",
"Snippet": "as Kyiv continued its diplomatic efforts"
},
{
"location": "Moscow",
"ADM1": "Moscow",
"Country": "Russia",
"Lat": 55.7558,
"Long": 37.6176,
"Explanation": "Moscow is mentioned as the location of the Russian defence ministry, which commented on the attacks.",
"Snippet": "The defence ministry in Moscow said"
},
{
"location": "Sumy region",
"ADM1": "Sumy Oblast",
"Country": "Ukraine",
"Lat": 50.9077,
"Long": 34.7981,
"Explanation": "Sumy region is mentioned as a place where a civilian was wounded as a result of the attacks.",
"Snippet": "a civilian was wounded in the northeastern Sumy region"
},
{
"location": "Chernihiv",
"ADM1": "Chernihiv Oblast",
"Country": "Ukraine",
"Lat": 51.4982,
"Long": 31.2893,
"Explanation": "Chernihiv is mentioned as the location where a missile created a large crater.",
"Snippet": "Police in the northern region of Chernihiv posted a picture"
},
{
"location": "Poland",
"ADM1": "",
"Country": "Poland",
"Lat": 51.9194,
"Long": 19.1451,
"Explanation": "Poland is mentioned as Ukraine's neighbour, where air defence systems were activated due to increased threat levels.",
"Snippet": "The air force in Ukraine\u2019s neighbour Poland said it had briefly activated air defence systems"
},
{
"location": "France",
"ADM1": "",
"Country": "France",
"Lat": 46.2276,
"Long": 2.2137,
"Explanation": "France is mentioned in the context of its foreign policy priorities and the visit of its foreign minister to Ukraine.",
"Snippet": "\u201cUkraine is and will remain France\u2019s priority,\u201d he said."
},
{
"location": "United Kingdom",
"ADM1": "",
"Country": "United Kingdom",
"Lat": 55.3781,
"Long": -3.436,
"Explanation": "The United Kingdom is referenced in the context of British Prime Minister Rishi Sunak's visit to Ukraine.",
"Snippet": "The visit follows that of British Prime Minister Rishi Sunak"
}
]
```
A third time there is a JSON formatting error that corrupts the results. In this case, since it is a JSON array, the entire output is corrupted, offering a reminder of the benefits of JSON-NL:
```json
[
{
"location": "Ukraine",
"ADM1": "",
"Country": "Ukraine",
"Lat": "48.3794",
"Long": "31.1656",
"Explanation": "Ukraine is mentioned as the primary location of the Russian attacks and the ongoing conflict. It's the country where all the mentioned events are taking place.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight"
},
{
"location": "Kyiv",
"ADM1": "Kyiv City",
"Country": "Ukraine",
"Lat": "50.4501",
"Long": "30.5234",
"Explanation": "Kyiv is mentioned as the capital of Ukraine and the location of diplomatic efforts and a news conference.",
"Snippet": "as Kyiv continued its diplomatic efforts to ratchet up support and military aid."
},
{
"location": "Sumy region",
"ADM1": "Sumy Oblast",
"Country": "Ukraine",
"Lat": "50.9077",
"Long": "34.7981",
"Explanation": "Sumy region is mentioned as a location where a civilian was wounded due to the attacks.",
"Snippet": "a civilian was wounded in the northeastern Sumy region."
},
{
"location": "Chernihiv",
"ADM1": "Chernihiv Oblast",
"Country": "Ukraine",
"Lat": "51.4982",
"Long": "31.2893",
"Explanation": "Chernihiv is mentioned as the northern region of Ukraine where a missile created a large crater.",
"Snippet": "Police in the northern region of Chernihiv posted a picture of a large crater made by a downed missile."
},
{
"location": "Russia",
"ADM1": "",
"Country": "Russia",
"Lat": "61.5240",
"Long": "105.3188",
"Explanation": "Russia is mentioned as the aggressor in the conflict and the country launching the attacks.",
"Snippet": "Russia said it destroyed all targets in a barrage of attacks on facilities producing ammunition and drones in Ukraine."
},
{
"location": "Moscow",
"ADM1": "Moscow",
"Country": "Russia",
"Lat": "55.7558",
"Long": "37.6173",
"Explanation": "Moscow is mentioned as the location of the Russian defence ministry, which made statements about the attacks.",
"Snippet": "The defence ministry in Moscow said its armed forces \"carried out a group strike … against facilities of the Ukrainian military-industrial complex\"."
},
{
"location": "Poland",
"ADM1": "",
"Country": "Poland",
"Lat": "51.9194",
"Long": "19.1451",
"Explanation": "Poland is mentioned as Ukraine's neighbour and the location where the air force activated air defence systems due to the increased threat.",
"Snippet": "The air force in Ukraine’s neighbour Poland said it had briefly activated air defence systems due to the increased level of threat."
},
{
"location": "France",
"ADM1": "",
"Country": "France",
"Lat": "46.2276",
"Long": "2.2137",
"Explanation": "France is mentioned in context of its foreign minister's visit to Ukraine and its priority in supporting Ukraine.",
"Snippet": "The visit follows that of British Prime Minister
Rishi Sunak on Friday, where he unveiled new military funding for Ukraine and reassured the country that the West is still providing support."
},
{
"location": "United Kingdom",
"ADM1": "",
"Country": "United Kingdom",
"Lat": "55.3781",
"Long": "-3.4360",
"Explanation": "The United Kingdom is mentioned in relation to its Prime Minister Rishi Sunak's visit to Ukraine and the support provided.",
"Snippet": "The visit follows that of British Prime Minister Rishi Sunak on Friday, where he unveiled new military funding for Ukraine."
}
]
```
A fourth time we get a similar fatal JSON error:
```json
[
{
"location": "Ukraine",
"ADM1": "",
"Country": "Ukraine",
"Lat": 48.3794,
"Long": 31.1656,
"Explanation": "Ukraine is mentioned as the primary location of the Russian attacks and the ongoing conflict. The country is the main subject of the text.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight"
},
{
"location": "Kyiv",
"ADM1": "Kyiv",
"Country": "Ukraine",
"Lat": 50.4501,
"Long": 30.5234,
"Explanation": "Kyiv is mentioned as the capital of Ukraine, where diplomatic efforts are ongoing and where the Ukrainian air force made statements.",
"Snippet": "as Kyiv continued its diplomatic efforts"
},
{
"location": "Russia",
"ADM1": "",
"Country": "Russia",
"Lat": 61.5240,
"Long": 105.3188,
"Explanation": "Russia is mentioned as the aggressor in the conflict, launching attacks and making statements through its defence ministry.",
"Snippet": "Russia said it destroyed all targets"
},
{
"location": "Moscow",
"ADM1": "Moscow",
"Country": "Russia",
"Lat": 55.7558,
"Long": 37.6173,
"Explanation": "Moscow is mentioned as the location of the Russian defence ministry, which made statements about the attacks.",
"Snippet": "The defence ministry in Moscow said"
},
{
"location": "Sumy",
"ADM1": "Sumy",
"Country": "Ukraine",
"Lat": 50.9077,
"Long": 34.7981,
"Explanation": "Sumy region is mentioned as a location where a civilian was wounded due to the conflict.",
"Snippet": "a civilian was wounded in the northeastern Sumy region"
},
{
"location": "Chernihiv",
"ADM1": "Chernihiv",
"Country": "Ukraine",
"Lat": 51.4982,
"Long": 31.2893,
"Explanation": "Chernihiv region is mentioned as a location where a missile created a large crater and caused damage.",
"Snippet": "Police in the northern region of Chernihiv posted a picture"
},
{
"location": "Poland",
"ADM1": "",
"Country": "Poland",
"Lat": 51.9194,
"Long": 19.1451,
"Explanation": "Poland is mentioned as Ukraine's neighbour, where air defence systems were briefly activated due
to the conflict.",
"Snippet": "The air force in Ukraine’s neighbour Poland said it had briefly activated air defence systems"
},
{
"location": "France",
"ADM1": "",
"Country": "France",
"Lat": 46.2276,
"Long": 2.2137,
"Explanation": "France is mentioned in the context of its foreign policy and support for Ukraine, particularly through the visit of its newly appointed foreign minister.",
"Snippet": "Speaking at a news conference in Kyiv, Ukrainian Foreign Minister Dmytro Kuleba praised Sejourne for not being intimidated by Russia’s latest large-scale missile attack"
},
{
"location": "United Kingdom",
"ADM1": "",
"Country": "United Kingdom",
"Lat": 55.3781,
"Long": -3.4360,
"Explanation": "United Kingdom is mentioned in relation to the visit of British Prime Minister Rishi Sunak to Ukraine.",
"Snippet": "The visit follows that of British Prime Minister Rishi Sunak on Friday"
}
]
```
A fifth time we get correct results again:
```json
[
{
"location": "Ukraine",
"ADM1": "",
"Country": "Ukraine",
"Lat": 48.379433,
"Long": 31.16558,
"Explanation": "Ukraine is mentioned throughout the text as the primary location where the attacks and diplomatic efforts are taking place.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight"
},
{
"location": "Kyiv",
"ADM1": "Kyiv City",
"Country": "Ukraine",
"Lat": 50.4501,
"Long": 30.5234,
"Explanation": "Kyiv is mentioned as the location of ongoing diplomatic efforts and the site of a news conference.",
"Snippet": "as Kyiv continued its diplomatic efforts"
},
{
"location": "Moscow",
"ADM1": "Moscow",
"Country": "Russia",
"Lat": 55.7558,
"Long": 37.6173,
"Explanation": "Moscow is mentioned as the location of the defence ministry that made statements regarding the attacks.",
"Snippet": "The defence ministry in Moscow said"
},
{
"location": "Sumy region",
"ADM1": "Sumy Oblast",
"Country": "Ukraine",
"Lat": 50.9077,
"Long": 34.7981,
"Explanation": "Sumy region is mentioned as a place where a civilian was wounded as a result of the attacks.",
"Snippet": "a civilian was wounded in the northeastern Sumy region"
},
{
"location": "Chernihiv",
"ADM1": "Chernihiv Oblast",
"Country": "Ukraine",
"Lat": 51.4982,
"Long": 31.2893,
"Explanation": "Chernihiv is mentioned as a northern region where a missile created a large crater.",
"Snippet": "Police in the northern region of Chernihiv posted a picture"
},
{
"location": "Poland",
"ADM1": "",
"Country": "Poland",
"Lat": 51.9194,
"Long": 19.1451,
"Explanation": "Poland is mentioned as Ukraine's neighbour which activated air defence systems due to the attacks.",
"Snippet": "The air force in Ukraine\u2019s neighbour Poland said"
},
{
"location": "France",
"ADM1": "",
"Country": "France",
"Lat": 46.2276,
"Long": 2.2137,
"Explanation": "France is mentioned in relation to its foreign minister's visit to Kyiv and its foreign policy priorities.",
"Snippet": "France\u2019s newly appointed foreign minister"
},
{
"location": "United Kingdom",
"ADM1": "",
"Country": "United Kingdom",
"Lat": 55.3781,
"Long": -3.436,
"Explanation": "The United Kingdom is implicated through the mention of its Prime Minister Rishi Sunak's visit to Ukraine.",
"Snippet": "The visit follows that of British Prime Minister Rishi Sunak"
}
]
```
How about our most advanced prompt with the Indian Express article? Shockingly, GPT-4.0 performs far worse than GPT-3.5. Partially this is due to its failure to understand the nuances of India's administrative divisions, such as identifying Kilvelur vs Kilvelur taluka, Nagapattinam vs Nagapattinam district, Bilari vs Bilari tehsil and even Madras in place of University of Madras. In essence, GPT-3.5 focused on extracting geographic string literals, whereas GPT-4.0 ends up trying to be too smart for its own good, incorrectly identifying the boundaries of location references and in doing so identifying the wrong location references. In its first run, it recognized just 10 of 31 locations, for an accuracy of just 32%. Worse, the JSON also contains a fatal format error:
Bengaluru Bilari tehsil Cauvery Chandausi Dusi Foundation for Agrarian Studies Gangaikondan Indian Statistical Institute Iruvelpattu Jawaharlal Nehru University Kilvelur taluka Kolkata Lower Cauvery Delta Mettur dam Moradabad Moradabad district Nagapattinam district New Delhi North Arcot Palakurichi Palanpur Ramanathapuram South Arcot South India Tamil Nadu Thanjavur Tirunelveli University of Madras University of Oxford Uttar Pradesh Vadamalaipuram
The full JSON:
```json
[
{
"location": "Madras",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": 13.0827,
"Long": 80.2707,
"Explanation": "Madras (now Chennai) is mentioned as the location of the University where Gilbert Slater, a professor, conducted his survey. It is in the state of Tamil Nadu, India.",
"Snippet": "Gilbert Slater, professor of economics at the University of Madras"
},
{
"location": "Dusi",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": 12.9352,
"Long": 79.5433,
"Explanation": "Dusi is mentioned as one of the Slater villages. It is located in Tamil Nadu, India.",
"Snippet": "The “Slater villages” — Dusi (North Arcot)"
},
{
"location": "Iruvelpattu",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": 11.9265,
"Long": 79.6098,
"Explanation": "Iruvelpattu is mentioned as one of the Slater villages. It is located in Tamil Nadu, India.",
"Snippet": "Iruvelpattu (South Arcot)"
},
{
"location": "Palakurichi",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": 10.7668,
"Long": 79.7004,
"Explanation": "Palakurichi is repeatedly mentioned as a significant village in the study. It is located in Tamil Nadu, India.",
"Snippet": "Palakurichi (Thanjavur)"
},
{
"location": "Vadamalaipuram",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": 9.3894,
"Long": 78.5633,
"Explanation": "Vadamalaipuram is mentioned as one of the Slater villages. It is located in Tamil Nadu, India.",
"Snippet": "Vadamalaipuram (Ramanathapuram)"
},
{
"location": "Gangaikondan",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": 8.8577,
"Long": 77.7666,
"Explanation": "Gangaikondan is mentioned as one of the Slater villages. It is located in Tamil Nadu, India.",
"Snippet": "Gangaikondan (Tirunelveli)"
},
{
"location": "Kilvelur",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": 10.7666,
"Long": 79.6667,
"Explanation": "Kilvelur is mentioned as the taluka where Palakurichi is now located. It is in Tamil Nadu, India.",
"Snippet": "Palakurichi, now in Kilvelur taluka of Nagapattinam district"
},
{
"location": "Nagapattinam",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": 10.7660,
"Long": 79.8443,
"Explanation": "Nagapattinam is mentioned as the district where Palakurichi is located. It is in the state of Tamil Nadu, India.",
"Snippet": "Kilvelur taluka of Nagapattinam district"
},
{
"location": "Kolkata",
"ADM1": "West Bengal",
"Country": "India",
"Lat": 22.5726,
"Long": 88.3639,
"Explanation": "Kolkata is mentioned as the location of the Indian Statistical Institute, where V Surjit, a PhD student, was studying. It is in the state of West Bengal, India.",
"Snippet": "a PhD student at the Indian Statistical Institute, Kolkata"
},
{
"location": "Bengaluru",
"ADM1": "Karnataka",
"Country": "India",
"Lat": 12.9716,
"Long": 77.5946,
"Explanation": "Bengaluru is mentioned in relation to the Foundation for Agrarian Studies, which conducted a survey in 2019. It is in the state of Karnataka, India.",
"Snippet": "by the Foundation for Agrarian Studies, Bengaluru"
},
{
"location": "Palanpur",
"ADM1": "Uttar Pradesh",
"Country": "India",
"Lat": 28.6219,
"Long": 78.7486,
"Explanation": "Palanpur is mentioned as another village surveyed, located in Uttar Pradesh, India.",
"Snippet": "The only comparable example is Palanpur, a village in Bilari tehsil of Uttar Pradesh’s Moradabad district"
},
{
"location": "Bilari",
"ADM1": "Uttar Pradesh",
"Country": "India",
"Lat": 28.6327,
"Long": 78.8027,
"Explanation": "Bilari is mentioned as the tehsil where Palanpur is located. It is in the state of Uttar Pradesh, India.",
"Snippet": "a village in Bilari tehsil of Uttar Pradesh’s Moradabad district"
},
{
"location": "Moradabad",
"ADM1": "Uttar Pradesh",
"Country": "India",
"Lat": 28.8386,
"Long": 78.7733,
"Explanation": "Moradabad is mentioned as the district where Palanpur and Bilari are located. It is in Uttar Pradesh, India.",
"Snippet": "Moradabad district"
},
{
"location": "Chandausi",
"ADM1": "Uttar Pradesh",
"Country": "India",
"Lat": 28.4506,
"Long": 78.7811,
"Explanation": "Chandausi is mentioned as a neighboring town to Palanpur, offering non-farm employment opportunities. It is in Uttar Pradesh, India.",
"Snippet": "including in neighbouring towns such as Chandausi"
}
]
```
Running a second time it recorded "Mettur" instead of "Mettur dam" and again improperly connected "Indian Statistical Institute, Kolkata" and this time added "Foundation for Agrarian Studies, Bengaluru". This time it got 11 of 31 entries correct (35%). If the three wrong entries are added, the result is 11/34 (32% accuracy).
Bengaluru Bilari tehsil Cauvery Chandausi Dusi Foundation for Agrarian Studies Gangaikondan Indian Statistical Institute Iruvelpattu Jawaharlal Nehru University Kilvelur taluka Kolkata Lower Cauvery Delta Mettur dam Moradabad Moradabad district Nagapattinam district New Delhi North Arcot Palakurichi Palanpur Ramanathapuram South Arcot South India Tamil Nadu Thanjavur Tirunelveli University of Madras University of Oxford Uttar Pradesh Vadamalaipuram Mettur dam Indian Statistical Institute, Kolkata Foundation for Agrarian Studies, Bengaluru
The full JSON:
```json
[
{
"location": "University of Madras",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": 13.067439,
"Long": 80.237617,
"Explanation": "The text mentions Gilbert Slater, a professor at the University of Madras, conducting a survey. The University of Madras is a known educational institution located in Tamil Nadu, India.",
"Snippet": "Gilbert Slater, professor of economics at the University of Madras"
},
{
"location": "Dusi",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": 12.9045,
"Long": 79.8942,
"Explanation": "Dusi is mentioned as one of the 'Slater villages' surveyed. It is located in the North Arcot region of Tamil Nadu, India.",
"Snippet": "The 'Slater villages' — Dusi (North Arcot)"
},
{
"location": "Iruvelpattu",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": 11.9300,
"Long": 79.7850,
"Explanation": "Iruvelpattu is listed as another 'Slater village'. It is situated in the South Arcot region of Tamil Nadu, India.",
"Snippet": "Iruvelpattu (South Arcot)"
},
{
"location": "Palakurichi",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": 10.7667,
"Long": 79.7000,
"Explanation": "Palakurichi is extensively mentioned as a unique village studied over 100 years. It is now located in Kilvelur taluka of Nagapattinam district, Tamil Nadu, India.",
"Snippet": "One of the five — Palakurichi, now in Kilvelur taluka of Nagapattinam district"
},
{
"location": "Vadamalaipuram",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": 9.2601,
"Long": 78.3114,
"Explanation": "Vadamalaipuram is mentioned as one of the Slater villages. It is located in the Ramanathapuram region of Tamil Nadu, India.",
"Snippet": "Vadamalaipuram (Ramanathapuram)"
},
{
"location": "Gangaikondan",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": 8.9613,
"Long": 77.7442,
"Explanation": "Gangaikondan is another village surveyed, located in the Tirunelveli region of Tamil Nadu, India.",
"Snippet": "Gangaikondan (Tirunelveli)"
},
{
"location": "Palanpur",
"ADM1": "Uttar Pradesh",
"Country": "India",
"Lat": 28.2843,
"Long": 78.4042,
"Explanation": "Palanpur is mentioned as a village in Bilari tehsil of Uttar Pradesh's Moradabad district, comparable to Palakurichi in terms of being longitudinally studied.",
"Snippet": "The only comparable example is Palanpur, a village in Bilari tehsil of Uttar Pradesh’s Moradabad district"
},
{
"location": "Mettur",
"ADM1": "Tamil Nadu",
"Country": "India",
"Lat": 11.8000,
"Long": 77.8000,
"Explanation": "Mettur is mentioned in relation to its dam's construction, which impacted agricultural progress in the region. Mettur is located in Tamil Nadu, India.",
"Snippet": "The zenith of its agricultural progress was reached with the Mettur dam’s construction"
},
{
"location": "University of Oxford",
"ADM1": "England",
"Country": "United Kingdom",
"Lat": 51.7520,
"Long": -1.2577,
"Explanation": "The University of Oxford is mentioned as the institution of Margaret Haswell, who studied Palakurichi. It is a well-known educational institution in Oxford, England.",
"Snippet": "studied again in 1964 (by Margaret Haswell from the University of Oxford)"
},
{
"location": "Indian Statistical Institute, Kolkata",
"ADM1": "West Bengal",
"Country": "India",
"Lat": 22.5793,
"Long": 88.3700,
"Explanation": "The Indian Statistical Institute in Kolkata is mentioned as the institution of a PhD student, V Surjit, who also studied Palakurichi. It is located in Kolkata, West Bengal, India.",
"Snippet": "2004 (by V Surjit, a PhD student at the Indian Statistical Institute, Kolkata)"
},
{
"location": "Foundation for Agrarian Studies, Bengaluru",
"ADM1": "Karnataka",
"Country": "India",
"Lat": 12.9716,
"Long": 77.5946,
"Explanation": "The Foundation for Agrarian Studies, located in Bengaluru, Karnataka, India, is mentioned as having conducted a 2019 study of Palakurichi.",
"Snippet": "2019 (by the Foundation for Agrarian Studies, Bengaluru)"
},
{
"location": "Jawaharlal Nehru University",
"ADM1": "Delhi",
"Country": "India",
"Lat": 28.5400,
"Long": 77.1642,
"Explanation": "Jawaharlal Nehru University is mentioned as the institution of associate professor Himanshu, who studied Palanpur. It's located in Delhi, India.",
"Snippet": "Our later visits suggest non-farm jobs are returning again. We plan to do a new survey that will be comprehensive,' said Himanshu, associate professor at the Jawaharlal Nehru University."
},
{
"location": "Chandausi",
"ADM1": "Uttar Pradesh",
"Country": "India",
"Lat": 28.4500,
"Long": 78.7833,
"Explanation": "Chandausi is mentioned as a neighboring town to Palanpur, where non-farm employment opportunities have expanded. It is located in Uttar Pradesh, India.",
"Snippet": "including in neighbouring towns such as Chandausi and Moradabad"
},
{
"location": "Moradabad",
"ADM1": "Uttar Pradesh",
"Country": "India",
"Lat": 28.8386,
"Long": 78.7733,
"Explanation": "Moradabad is mentioned in relation to Palanpur's location in its district. It is a city in Uttar Pradesh, India.",
"Snippet": "a village in Bilari tehsil of Uttar Pradesh’s Moradabad district"
}
]
```
Gemini Pro
Google's Gemini Pro? We'll start with our original prompt:
You are a geocoding agent, your job is to read text and compile a list of all geographic locations mentioned within.
Make a list of all of the geographic locations mentioned in the text below.
Output in the following JSON format: "{ city: "City Name", ADM1: ADM1, Country: Country Name, Lat: Latitude, Long: Longitude }":
TEXT:
On its first try it yields results highly similar to GPT 4.0. However, it misses Moscow and Britain, as well as the EU:
[
{
"city": "Kyiv",
"ADM1": "Kyiv Oblast",
"Country": "Ukraine",
"Lat": 50.4501,
"Long": 30.5234
},
{
"city": "Sumy",
"ADM1": "Sumy Oblast",
"Country": "Ukraine",
"Lat": 50.9085,
"Long": 34.7995
},
{
"city": "Chernihiv",
"ADM1": "Chernihiv Oblast",
"Country": "Ukraine",
"Lat": 51.4954,
"Long": 31.2858
},
{
"city": "Poland",
"ADM1": null,
"Country": "Poland",
"Lat": 52.237049,
"Long": 21.017532
},
{
"city": "France",
"ADM1": null,
"Country": "France",
"Lat": 46.227638,
"Long": 2.213749
}
]
Running a second time yields a guardrail false positive:
I'm just a language model, so I can't help you with that.
A third time yields the same results as the first time with the missing entries:
[
{ "city": "Kyiv", "ADM1": "Kyiv City", "Country": "Ukraine", "Lat": 50.4501, "Long": 30.5234 },
{ "city": "Sumy", "ADM1": "Sumy Oblast", "Country": "Ukraine", "Lat": 50.9085, "Long": 34.8028 },
{ "city": "Chernihiv", "ADM1": "Chernihiv Oblast", "Country": "Ukraine", "Lat": 51.4954, "Long": 31.2856 },
{ "city": "Poland", "ADM1": null, "Country": "Poland", "Lat": 51.9194, "Long": 19.1451 },
{ "city": "France", "ADM1": null, "Country": "France", "Lat": 46.2276, "Long": 2.2137 }
]
A fourth time it correctly includes Moscow, but falsely claims in the explanatory text that Moscow was not mentioned in the text:
## Geographic Locations:
```json
[
{
"city": "Kyiv",
"ADM1": null,
"Country": "Ukraine",
"Lat": 50.4422,
"Long": 30.5365
},
{
"city": "Sumy",
"ADM1": "Sumy Oblast",
"Country": "Ukraine",
"Lat": 51.1333,
"Long": 35.8167
},
{
"city": "Chernihiv",
"ADM1": "Chernihiv Oblast",
"Country": "Ukraine",
"Lat": 51.5050,
"Long": 31.2850
},
{
"city": "Poland",
"ADM1": null,
"Country": "Poland",
"Lat": 51.9194,
"Long": 19.1451
},
{
"city": "Moscow",
"ADM1": "Moscow",
"Country": "Russia",
"Lat": 55.7558,
"Long": 37.6173
}
]
```
**Notes:**
* I included "Poland" as a location, although it wasn't explicitly mentioned as a city.
* I couldn't find specific latitude and longitude values for some locations mentioned in passing, such as "five regions across Ukraine."
* I included Moscow for context, even though it isn't mentioned in the text itself.
I hope this is helpful! Let me know if you have any other questions.
A fifth time we're back to our usual output:
[
{ "city": "Kyiv", "ADM1": "Kyiv Oblast", "Country": "Ukraine", "Lat": 50.4501, "Long": 30.5234 },
{ "city": "Sumy", "ADM1": "Sumy Oblast", "Country": "Ukraine", "Lat": 50.9069, "Long": 34.8028 },
{ "city": "Chernihiv", "ADM1": "Chernihiv Oblast", "Country": "Ukraine", "Lat": 51.4954, "Long": 31.2855 },
{ "city": "Poland", "ADM1": null, "Country": "Poland", "Lat": 51.9194, "Long": 19.1451 },
{ "city": "France", "ADM1": null, "Country": "France", "Lat": 46.2276, "Long": 2.2137 }
]
And a sixth time we're back to a guardrail false positive:
I'm a text-based AI, and that is outside of my capabilities.
This time it correctly includes Moscow, but hallucinates Warsaw and its textual description makes no mention of it adding it specially for any reason:
## Geographic Locations:
```json
[
{
"city": "Kyiv",
"ADM1": null,
"Country": "Ukraine",
"Lat": 50.4422,
"Long": 30.5365
},
{
"city": "Sumy",
"ADM1": "Sumy Oblast",
"Country": "Ukraine",
"Lat": 51.1333,
"Long": 35.8167
},
{
"city": "Chernihiv",
"ADM1": "Chernihiv Oblast",
"Country": "Ukraine",
"Lat": 51.5044,
"Long": 31.2767
},
{
"city": "Moscow",
"ADM1": "Moscow",
"Country": "Russia",
"Lat": 55.7515,
"Long": 37.6178
},
{
"city": "Warsaw",
"ADM1": "Masovian Voivodeship",
"Country": "Poland",
"Lat": 52.2373,
"Long": 21.0172
},
{
"city": "France",
"ADM1": null,
"Country": "France",
"Lat": 46.2276,
"Long": 2.2137
}
]
```
**Notes:**
* I included "France" as a country, though not a specific city, as it was mentioned in the context of diplomatic efforts.
* ADM1 level is not available for all locations mentioned.
How about our most advanced prompt?
You are a geocoding agent, your job is to read text and compile a list of all geographic locations mentioned within.
Make a list of all of the geographic locations mentioned in the text below.
Output in the following JSON format: "{ location: "Location Name", ADM1: ADM1, Country: Country Name, Lat: Latitude, Long: Longitude, Explanation: Explanation, Snippet: Snippet }", with all of the locations in a single JSON array.
Do not include any locations not mentioned in the text and make sure to include all locations mentioned in the text. For the Explanation field explain your reasoning for including that location step-by-step.
For the Snippet field include a brief snippet from the text that includes a mention of that location.
Only return the JSON array itself.
TEXT:
Despite being explicitly instructed to include all locations, it still returns only a portion of the results:
[
{
"location": "Kyiv",
"ADM1": "Kyiv Oblast",
"Country": "Ukraine",
"Lat": 50.4501,
"Long": 30.5234,
"Explanation": "The text explicitly mentions Kyiv as the location of several events, including air strikes, diplomatic visits, and press conferences. It is also referred to as the Ukrainian capital.",
"Snippet": "The attacks took place as France’s newly appointed foreign minister, Stephane Sejourne, arrived in Kyiv on Saturday for his first official trip abroad."
},
{
"location": "Sumy",
"ADM1": "Sumy Oblast",
"Country": "Ukraine",
"Lat": 50.9085,
"Long": 34.7905,
"Explanation": "The text states that a civilian was wounded in the northeastern Sumy region, indicating that Sumy is a geographic location within Ukraine.",
"Snippet": "No fatalities were reported so far, but Ukrainian authorities said a civilian was wounded in the northeastern Sumy region."
},
{
"location": "Chernihiv",
"ADM1": "Chernihiv Oblast",
"Country": "Ukraine",
"Lat": 51.4954,
"Long": 31.2857,
"Explanation": "The text mentions Chernihiv as the region where a missile was shot down and caused damage to buildings. This suggests that Chernihiv is a geographic location within Ukraine.",
"Snippet": "Police in the northern region of Chernihiv posted a picture of a large crater made by a downed missile."
},
{
"location": "Poland",
"ADM1": "",
"Country": "Poland",
"Lat": 51.9194,
"Long": 19.1451,
"Explanation": "While not explicitly stated as a location of attacks, the text mentions Poland as a neighboring country to Ukraine that activated air defense systems due to the increased threat. This implies that Poland is a geographic location relevant to the events described.",
"Snippet": "The air force in Ukraine’s neighbour Poland said it had briefly activated air defence systems due to the increased level of threat."
}
]
Run a second time we don't even get JSON:
Sure, here is a list of all the geographic locations mentioned in the text, along with a brief explanation for why they were included: * **Ukraine**: This is the main country where the events in the text are taking place. It is mentioned throughout the text, including in the first sentence. * **Kyiv**: This is the capital of Ukraine, and it is mentioned several times in the text, including in the context of the Russian attacks and the visit of the French foreign minister. * **Sumy Oblast**: This is a province in northeastern Ukraine, and it is mentioned in the context of a civilian being wounded in a Russian attack. * **Chernihiv**: This is a city in northern Ukraine, and it is mentioned in the context of a Russian missile attack that destroyed a building. * **Poland**: This is a country bordering Ukraine to the west, and it is mentioned in the context of the Polish air force activating its air defense systems due to the increased threat from Russia. I hope this helps! http://googleusercontent.com/map_content/0
A third time we still don't get Moscow:
[
{
"location": "Ukraine",
"ADM1": "",
"Country": "Ukraine",
"Lat": 48.379433,
"Long": 31.16558,
"Explanation": "The text explicitly states that the attacks took place in Ukraine.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight, the Ukrainian air force said"
},
{
"location": "Kyiv",
"ADM1": "Kyiv Oblast",
"Country": "Ukraine",
"Lat": 50.4501,
"Long": 30.5234,
"Explanation": "The text mentions Kyiv as the location of the Ukrainian air force and the place where the French foreign minister arrived.",
"Snippet": "the Ukrainian air force said, as Kyiv continued its diplomatic efforts"
},
{
"location": "Sumy region",
"ADM1": "Sumy Oblast",
"Country": "Ukraine",
"Lat": 50.9083,
"Long": 34.8,
"Explanation": "The text mentions a civilian being wounded in the Sumy region.",
"Snippet": "Ukrainian authorities said a civilian was wounded in the northeastern Sumy region"
},
{
"location": "Chernihiv",
"ADM1": "Chernihiv Oblast",
"Country": "Ukraine",
"Lat": 51.49553,
"Long": 31.28583,
"Explanation": "The text mentions Chernihiv as the region where a missile was shot down.",
"Snippet": "Police in the northern region of Chernihiv posted a picture of a large crater"
},
{
"location": "Poland",
"ADM1": "",
"Country": "Poland",
"Lat": 51.919438,
"Long": 19.145136,
"Explanation": "The text mentions Poland as Ukraine's neighbor and that its air force activated air defense systems.",
"Snippet": "The air force in Ukraine’s neighbour Poland said it had briefly activated air defence systems"
},
{
"location": "France",
"ADM1": "",
"Country": "France",
"Lat": 46.227638,
"Long": 2.213749,
"Explanation": "The text mentions France as the country of origin of the foreign minister who visited Kyiv.",
"Snippet": "France’s newly appointed foreign minister, Stephane Sejourne"
}
]
And a fourth time we're back to non-JSON and no Moscow:
Sure, here are the places mentioned in the conversation on Google Maps: * [Ukraine](http://www.ukraine.com/) * [Kyiv, Ukraine](http://www.kievcity.gov.ua/) * [Sumy Oblast, Ukraine](http://www.state-gov.sumy.ua/) * [Chernihiv, Ukraine](http://www.chernigiv-rada.gov.ua/) * [Poland](http://en.poland.gov.pl/) * [France](http://www.france.fr/) I hope this helps! http://googleusercontent.com/map_content/0
And a fifth time we're back to JSON. Regardless of our strict formatting instructions, Gemini Pro keeps alternating. Worse, regardless of our instructions, we cannot get it to include Moscow:
[
{
"location": "Ukraine",
"ADM1": "",
"Country": "Ukraine",
"Lat": 48.379433,
"Long": 31.16558,
"Explanation": "The text explicitly states that the attacks took place in Ukraine.",
"Snippet": "Russia launched dozens of attacks across Ukraine overnight, the Ukrainian air force said"
},
{
"location": "Kyiv",
"ADM1": "Kyiv Oblast",
"Country": "Ukraine",
"Lat": 50.4501,
"Long": 30.5234,
"Explanation": "The text mentions Kyiv as the location of the Ukrainian air force and the place where the French foreign minister arrived.",
"Snippet": "the Ukrainian air force said, as Kyiv continued its diplomatic efforts"
},
{
"location": "Sumy region",
"ADM1": "Sumy Oblast",
"Country": "Ukraine",
"Lat": 50.9083,
"Long": 34.8,
"Explanation": "The text mentions a civilian being wounded in the Sumy region.",
"Snippet": "Ukrainian authorities said a civilian was wounded in the northeastern Sumy region"
},
{
"location": "Chernihiv",
"ADM1": "Chernihiv Oblast",
"Country": "Ukraine",
"Lat": 51.4958,
"Long": 31.2875,
"Explanation": "The text mentions police in Chernihiv reporting damage from a downed missile.",
"Snippet": "Police in the northern region of Chernihiv posted a picture of a large crater"
},
{
"location": "Poland",
"ADM1": "",
"Country": "Poland",
"Lat": 51.919438,
"Long": 19.145136,
"Explanation": "The text mentions Poland as a neighbor of Ukraine that activated air defense systems.",
"Snippet": "The air force in Ukraine’s neighbour Poland said it had briefly activated air defence systems"
},
{
"location": "France",
"ADM1": "",
"Country": "France",
"Lat": 46.227638,
"Long": 2.213749,
"Explanation": "The text mentions France as the country of origin of the foreign minister who visited Kyiv.",
"Snippet": "France’s newly appointed foreign minister, Stephane Sejourne"
}
]
Finally, what about our most advanced prompt on the Indian Express article? It correctly extracts just 5 of the 31 entries (16% accuracy) before ending with invalid JSON:
Bengaluru Bilari tehsil Cauvery Chandausi Dusi Foundation for Agrarian Studies Gangaikondan Indian Statistical Institute Iruvelpattu Jawaharlal Nehru University Kilvelur taluka Kolkata Lower Cauvery Delta Mettur dam Moradabad Moradabad district Nagapattinam district New Delhi North Arcot Palakurichi Palanpur Ramanathapuram South Arcot South India Tamil Nadu Thanjavur Tirunelveli University of Madras University of Oxford Uttar Pradesh Vadamalaipuram
Full JSON:
[
{
"location": "Dusi",
"ADM1": "North Arcot",
"Country": "India",
"Lat": 13.1500,
"Long": 79.4667,
"Explanation": "The text states that Dusi is one of the five 'Slater villages' surveyed by Gilbert Slater in 1916-17. It is located in North Arcot, present-day Tamil Nadu, India. I found a village named Dusi in North Arcot with these coordinates.",
"Snippet": "The “Slater villages” — Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) — were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "Iruvelpattu",
"ADM1": "South Arcot",
"Country": "India",
"Lat": 11.4833,
"Long": 79.4167,
"Explanation": "The text states that Iruvelpattu is one of the five 'Slater villages' surveyed by Gilbert Slater in 1916-17. It is located in South Arcot, present-day Tamil Nadu, India. I found a village named Iruvelpattu in South Arcot with these coordinates.",
"Snippet": "The “Slater villages” — Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) — were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "Palakurichi",
"ADM1": "Nagapattinam",
"Country": "India",
"Lat": 10.7833,
"Long": 79.6500,
"Explanation": "The text mentions Palakurichi multiple times, stating it's in Kilvelur taluka of Nagapattinam district, Tamil Nadu, India. I found a village named Palakurichi in Nagapattinam with these coordinates.",
"Snippet": "One of the five — Palakurichi, now in Kilvelur taluka of Nagapattinam district — went on to be studied again in 1964 (by Margaret Haswell from the University of Oxford), 1983 (by S Guhan of the Madras Institute of Development Studies), 2004 (by V Surjit, a PhD student at the Indian Statistical Institute, Kolkata), and 2019 (by the Foundation for Agrarian Studies, Bengaluru)."
},
{
"location": "Vadamalaipuram",
"ADM1": "Ramanathapuram",
"Country": "India",
"Lat": 9.4167,
"Long": 78.8500,
"Explanation": "The text states that Vadamalaipuram is one of the five 'Slater villages' surveyed by Gilbert Slater in 1916-17. It is located in Ramanathapuram, present-day Tamil Nadu, India. I found a village named Vadamalaipuram in Ramanathapuram with these coordinates.",
"Snippet": "The “Slater villages” — Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) — were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "Gangaikondan",
"ADM1": "Tirunelveli",
"Country": "India",
"Lat": 8.5333,
"Long": 77.3500,
"Explanation": "The text states that Gangaikondan is one of the five 'Slater villages' surveyed by Gilbert Slater in 1916-17. It is located in Tirunelveli, present-day Tamil Nadu, India. I found a village named Gangaikondan in Tirunelveli with these coordinates
What if we run a second time? This time we get far better results than GPT-4.0. Though, it has a number of errors, such as Bilari vs Bilari tehsil, Cauvery Delta vs Lower Cauvery Delta and Mettur vs Mettur dam. Thanjavur appears twice. It correctly includes University of Madras, but also includes "Madras University" as a separate entry, not recognizing that it is a reference to the former. Overall, the accuracy is 21 / 31 (68%), while if the additional false positives are added, this decreases to 21 / 37 (57%):
Bengaluru Bilari tehsil Cauvery Chandausi Dusi Foundation for Agrarian Studies Gangaikondan Indian Statistical Institute Iruvelpattu Jawaharlal Nehru University Kilvelur taluka Kolkata Lower Cauvery Delta Mettur dam Moradabad Moradabad district Nagapattinam district New Delhi North Arcot Palakurichi Palanpur Ramanathapuram South Arcot South India Tamil Nadu Thanjavur Tirunelveli University of Madras University of Oxford Uttar Pradesh Vadamalaipuram
Full JSON is below:
[
{
"location": "University of Madras",
"ADM1": null,
"Country": "India",
"Lat": 13.0603,
"Long": 80.283,
"Explanation ": "The text mentions that Gilbert Slater, professor of economics at the University of Madras, surveyed five villages in present-day Tamil Nadu.",
"Snippet ": "In 1916-17, Gilbert Slater, professor of economics at the University of Madras, surveyed five villages in present-day Tamil Nadu through his students."
},
{
"location": "Tamil Nadu",
"ADM1": null,
"Country": "India",
"Lat": 11.1271,
"Long": 78.6569,
"Explanation": "The text mentions that Gilbert Slater surveyed five villages in present-day Tamil Nadu.",
"Snippet": "In 1916-17, Gilbert Slater, professor of economics at the University of Madras, surveyed five villages in present-day Tamil Nadu through his students."
},
{
"location": "Dusi",
"ADM1": null,
"Country": "India",
"Lat": 12.7319,
"Long": 79.7264,
"Explanation": " The text mentions tat Dusi was one of the five villages surveyed by Gilbert Slater.",
"Snippet": "The \\u201cSlater villages\\u201d \\u2014 Dusi (North Arcot), Iruvelpattu (South Arcot), Palakuri chi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) \\u2014 were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "North Arcot",
"ADM1": null,
"Country": "India",
"Lat": 12.9089,
"Long": 79.5327,
"Explanation": "The text mentions that Dusi is located in North Arcot.",
"Snippet": "The \\u201cSlater villages\\u201d \\u2014 Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaip uram (Ramanathapuram) and Gangaikondan (Tirunelveli) \\u2014 were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "Iruvelpattu",
"ADM1": null,
"Country": "India",
"Lat": 11.0855,
"Long": 79.6276,
"Explanation": "The text mentions that Iruvelpattu was one of the five vilages surveyed by Gilbert Slater.",
"Snippet": "The \\u201cSlater villages\\u201d \\u2014 Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) \\u2014 were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "South Arcot",
" ADM1": null,
"Country": "India",
"Lat": 11.3923,
"Long": 79.5901,
"Explanation": "The text mentions that Iruvelpattu is located in South Arcot.",
"Snippet": "The \\u201cSlater villages\\u201d \\u2014 Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) \\u2014 were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "Palakurichi",
"ADM1": "Nagap attinam",
"Country": "India",
"Lat": 10.7703,
"Long": 79.6151,
"Explanation": "The text mentions that Palakurichi was one of the five villages surveyedby Gilbert Slater.",
"Snippet": "The \\u201cSlater villages\\u201d \\u2014 Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) \\u2014 were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "Thanjavur",
"ADM1": null,
"Country": "India",
"Lat": 10.785,
"Long": 79.1389,
"Explanation": "The text mentions that Palakurichi is located in Thanjavur.",
"Snippet": "The \\u201cSlater villages\\u201d \\u2014 Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirune lveli) \\u2014 were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "Vadamalaipuram",
"ADM1": "Ramanathapuram",
" Country": "India",
"Lat": 9.2726,
"Long": 78.3287,
"Explanation": "The text mentions that Vadamalaipuram was one of the five villages surveyed by GilbertSlater.",
"Snippet": " The \\u201cSlater villages\\u201d \\u2014 Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan ( Tirunelveli) \\u2014 were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "Ramanathapuram",
"ADM1": null,
"Country": " India",
"Lat": 9.3643,
"Long": 78.7877,
"Explanation": "The text mentions that Vadamalaipuram is located in Ramanathapuram.",
"Snippet": "The \\u201cSlater villages\\u201d \\u2014 Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) \\u2014 were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "Gangaikondan",
"ADM1": "Tirunelveli",
"Country": "India ",
"Lat": 8.8406,
"Long": 77.7409,
"Explanation": "The text mentions that Gangaikondan was one of the five villages surveyed by Gilbert Slate.",
"Snippet": "The \\u201cSlater villages\\u201d \\u2014 Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) \\u2014 were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "Tirunelveli",
"ADM1": null,
"Country": "India",
" Lat": 8.7372,
"Long": 77.722,
"Explanation": "The text mentions that Gangaikondan is located in Tirunelveli.",
"Snippet": "The \\u201cSlater villages\\u201d \\u2014 Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) \\u2014 were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "Palanpur",
"ADM1": "Moradabad",
"Country": "India",
"Lat": 28.4519,
"Long": 79.4628,
"Explanation": "The text mentions that Palanpur is a village in Uttar Pradesh.",
"Snippet": "The only comparable example s Palanpur, a village in Bilari tehsil of Uttar Pradesh\\u2019s Moradabad district that has been surveyed every decade since Independence \\u2014 in 1957-58, 1962-63, 1974-75, 1983-84, 1993, 2008-09, 2015 and 2022."
},
{
"location": "Uttar Pradesh",
"ADM1": null,
"Country": "India",
"Lat": 27.154,
"Long": 80.9469,
"Explanation": "The text mentions that Palanpur is a village in Uttar Pradesh.",
"Snippet": "The only comparable example is Palanpur, a village in Bilari tehsil of Uttar Pradesh\\u2019s Moradabad district that has been surveyed every decade since Independence \\u2014 in 1957-58, 1962-63, 1974-75, 1983-84, 1993 , 2008-09, 2015 and 2022."
},
{
"location": "Moradabad",
"ADM1": null,
"Country": "India",
"Lat": 28.8411,
"Long": 78.7817,
"Explanation ": "The text mentions that Palanpur is located in Moradabad district.",
"Snippet": "The only comparable example is Palanpur, a village in Bilari tehsil of Uttar Pradesh\\u2019s Moradabad districtthat has been surveyed every decade since Independence \\u2014 in 1957-58, 1962-63 , 1974-75, 1983-84, 1993, 2008-09, 2015 and 2022."
},
{
"location": "Bilari",
"ADM1": null,
"Country": "India",
"Lat": 28.3878,
"Long": 78.9969,
"Explanation": "The text mentions that Palanpur is located in the Bilari tehsil of Moradabad district.",
"Snippet": "The only comparable example is Palanpur, a village in Bilari tehsil of Uttar Pradesh\\u2019s Moradabad district that has been surveyed every decade since Independence \\u2014 in 1957-58, 1962-63, 1974-75, 1983-84, 1993, 2008-09, 2015 and 2022."
},
{
"location ": "Cauvery Delta",
"ADM1": null,
"Country": "India",
"Lat": 10.8912,
"Long": 79.3867,
"Explanation": "The text mentions that Palakurichi is located in the Cauvery Delta.",
"Snippet": "Palakurichi is located in the lower Cauvery delta of the old Thanjavur region."
},
{
"location": "Thanjavur",
"ADM1": null,
"Country": "India",
"Lat": 10.785,
"Long": 79.1389,
"Explanation": "The text mentions that Palakurichi is located in the Thanjavur region.",
"Snippet": "Palakurichi is located in the lower Cauvery delta of the old Thanjavur region."
},
{
"location": "Mettur",
"ADM1": null,
"Country": "India",
"Lat": 11.7975,
"Long": 78.1097,
"Explanation": "The text mentions that the construction of the Mettur dam led to the agricultural progress of the Cauvery Delta.",
"Snippet": "The zenith of its agricultural progress was reached with the Mettur dam\\u2019s construction in 1934 and the Green Revolution \\u2014 the introduction of high-yielding short-duration rice varieties that made cultivation of two crops in a single year possible \\u2014from the mid-1960s through the 1980s."
},
{
"location": "Kolkata",
"ADM1": null,
"Country": "India",
"Lat": 22.589,
"Long": 88.3541,
"Explanation": "The text mentions that V Surjit of the Indian Statistical Institute, Kolkata studied Palakrichi.",
"Snippet": "One of the five \\u2014 Palakurichi, now in Kilvelur taluka of Nagapattinam district \\u2014 went on to be studied again in 1964 (by Margaret Haswell from the University of Oxford), 1983 (by S Guhan of the Madras Institute of Development Studies), 2004 (by V Surjit, a PhD student at the Indian Statistical Institute, Kolkata), and 2019 (by the Foundation for Agrarian Studies, Bengaluru)."
},
{
"location": "Madras University",
"ADM1": null,
"Country": "India",
"Lat": 13.0603,
"Long": 80.283,
"Explanation": "The text mentions that Madras University economists led by PJ Thomas resurveyed the five villages in 1936-37.",
"Snippet": "The \\u201cSlater villages\\u201d \\u2014 Dusi (North Arcot), Iruvelpattu (South Arcot), Palakurichi (Thanjavur), Vadamalaipuram (Ramanathapuram) and Gangaikondan (Tirunelveli) \\u2014 were resurveyed by Madras University economists led by PJ Thomas in 1936-37."
},
{
"location": "New Delhi",
"ADM1": null,
"Country": "India",
"Lat": 28.6139,
"Long": 77.2089,
"Explanation": "The text mentions that a recent volume about the 2 019 survey of Palakurichi village was published in New Delhi.",
"Snippet": "The results of the 2019 survey of Palakurichi village have been published in a recent volume (Economic Change in the Lower Cauvery Delta: A Study of Palakurichi and Venmani, edited by Madhura Swaminathan, V Surjit and VK Ramachandran, New Delhi : Tulika Books)."
},
{
"location": "National Sample Survey Office",
"ADM1": null,
"Country": "India",
"Lat": 28.6139,
"Long": 77.2089,
"Explanation": "The text mentions that the National Sample Survey Office undertakes nation-wide household surveys on consumption expenditure, employment , debt, housing and amenities, health and nutrition, education and other such socioeconomic indicators.",
"Snippet": "Longitudinal studies. Organisations such as the National Sample Survey Office undertake nation-wide household surveys on consumption expendture, employment, debt, housing and amenities, health and nutrition, education and other such socioeconomic indicators."
},
{
"location": "Chandausi",
"ADM1": null,
"Country": "India",
"Lat": 28.4314,
"Long": 78.7667,
" Explanation": "The text mentions that Palanpur is located near towns such as Chandausi.",
"Snippet": "Dalits have also registered some improvement, although not from access to land as much as expansion of non-farm employment opportunities, including in neighbouring towns such as Chandausi and Moradabad."
},
{
"location": "University of Oxford",
"ADM1": null,
"Country": "United Kingdom",
"Lat": 51.752,
"Long": -1.2579,
" Explanation": "The text mentions that Margaret Haswell from the University of Oxford studied Palakurichi.",
"Snippet": "One of the five \\u2014 Palakurichi, now in Kilvelur taluka of Nagapattinam district \\u2014 went on to be studied again in 1964 (by Margaret Haswell from the University of Oxford), 1983 (by S Guhn of the Madras Institute of Development Studies), 2004 (by V Surjit, a PhD student at the Indian Statistical Institute, Kolkata), and 20 19 (by the Foundation for Agrarian Studies, Bengaluru)."
},
{
"location": "Madras Institute of Development Studies",
"ADM1": null,
"Country": "India",
"Lat": 13.0603,
"Long": 80.283,
"Explanation": "The text mentions that S Guhan of the Madras Institute of Development Studies studied Palakurichi.",
"Snippet": "One of the five \\u2014 Palakurichi, now in Kilvelur taluka of Nagapattinam district \\u2014 went on to be studied again in 1964 (by Margaret Haswell from the University of Oxford), 1983 (by S Guhan of the Madras Institute of Development Studies), 2004 (by V Surjit, a PhD student at the Indian Statistical Institute, Kolkata), and 2019 (by the Foundation for Agrarian Studies, Bengaluru)."
},
{
"location": "Jawaharlal Nehru University",
"ADM1": null,
"Country": "India",
"Lat": 28.5539,
"Long": 77.1676,
"Explaation": "The text mentions that Himanshu, an associate professor at the Jawaharlal Nehru University, said that non-farm jobs are returning in Palakurichi.",
"Snippet": "\\u201cOur later visits suggest non-farm jobs are returning again. We plan to do a new survey that will be comprehensive,\\u201d said Himanshu, associate professor at the Jawaharlal Nehru University."
},
{
"location ": "Foundation for Agrarian Studies",
"ADM1": null,
"Country": "India",
"Lat": 12.9722,
"Long": 77.5936,
"Explanation": "The text mentions that the Foundation for Agrarian Studies conducted a survey of Palakurichi village in 2019.",
"Snippet": "One of the five \\u2014 Palakurichi, now in Kilvelur taluka of Nagapattinam district \\u2014 went on to be studied again in 1964 (by Margaret Haswell from the University of Oxford), 1983 (by S Guhan of the Madras Institute of Development Studies), 2004 (by V Surjit, a PhD student at the Indian Statistical Institute, Kolkata), and 2019 (by the Foundation for grarian Studies, Bengaluru)."
},
{
"location": "Tulika Books",
"ADM1": null,
"Country": "India",
"Lat": 28.6139,
"Long": 77.2089,
"Explanation": "The text mentions that the recent volume about the 2019 survey of Palakurichi village was published by Tulika Books.",
"Snippet": "The results of the 2019 survey of Palakurichi village have been published in a recent volume (Economic Change in the Lower Cauvery Delta: A Study of Palakurichi and Venmani, edited by Madhura Swaminathan, V Surjit and VK Ramachandran, New Delhi: Tulika Books)."
}
]