Search

ABSTRACT Geographic knowledge graphs (KGs) mainly describe static facts and have difficulty representing changes, greatly limiting their application in geographic information retrieval and geographic spatio‐temporal processes. By analyzing the spatio‐temporal features and evolution of geographic elements, this paper measures the degree of correlation between spatio‐temporal information and tuples and further accurately characterizes the semantic knowledge of tuples to support accurate computation and inference of KGs. This paper proposes a novel knowledge‐guided quantitative measure framework for spatio‐temporal correlation by considering rules and dependency syntax from natural language texts. Firstly, the natural language processing (NLP) stage preprocess the texts and extracts the candidate tuples by dependency syntactic analysis and rule matching. Secondly, we model the spatio‐temporal correlation measures by considering semantic (entity types and tuple predicate) and syntactic features (dependency distance and dependency path). Finally, we establish a specific threshold value with the extracted candidates and performing multiple levels of categorization to form the final spatio‐temporal correlation strength (strong, moderate, and weak). The experimental results with a large dataset indicate that the proposed method achieves an F‐score of over 0.73, which is better than those of the existing methods. The proposed spatio‐temporal correlation framework has more advantages in representing geographic evolutionary knowledge, revealing the evolution mechanism of geographic elements and the evolutionary reasons. [ABSTRACT FROM AUTHOR]

A Minimum Bounding Box (MBB) is a rectangle which bounds a geographic feature or dataset. It is commonly used in spatial information systems as a simplified way of describing the spatial extent of a resource. MBBs are typically indexed for searching and discovering resources relevant to a given geographic area of interest. However, this simplification leads to a loss of precision in the description of the extent and can affect the overall precision of the search results. We propose an alternative technique for describing the spatial extent based on the use of DGGS tiles. To measure the precision improvements offered by our method, we designed and implemented an empirical method for evaluating the average precision, and applied it to three different systems: one based on MBB, another on Convex Hull, and ours based on DGGS. The three methods were evaluated with the same test collection obtained from some of the main European geospatial data catalogues compliant with the INSPIRE directive. The results showed that our method outperformed the other two. Where the catalogue average precision of the MBB search scenarios is between 73% and 97%, the DGGS is between 96% and 99%. Additionally, we propose a realistic method of transitioning from the current technologies to the technology we are proposing, considering the current state of the spatial data infrastructures. [ABSTRACT FROM AUTHOR]

ABSTRACTMultiple tasks within the field of geographical information retrieval and geographical information sciences necessitate toponym matching, which involves the challenge of aligning toponyms that share a common referent. The multiple string similarity approaches struggle when confronted with the complexities associated with unofficial and/or historical variants of identical toponyms. Also, current state-of-the-art approaches/tools to supervised machine learning rely on labeled samples, and they do not adequately address the intricacies of character replacements either from transliterations or historical shifts in linguistic and cultural norms. To address these issues, this paper proposes a novel matching approach that leverages a deep neural network model empowered by geographic language representation model, known as GeoBERT, which stands for geographic Bidirectional Encoder Representations from Transformers (BERT). This model harnesses the groundbreaking capabilities of the GeoBERT framework by extending a generalized Enhanced Sequential Inference Model architecture and integrating multiple features to enhance the accuracy and robustness of the toponym matching. We present a comprehensive evaluation of the proposed method’s performance using three extensive datasets. The findings clearly illustrate that our approach outperforms the individual similarity metrics used in previous studies.

Describing where a place is situated is an innate communication skill that relies on spatial cognition, spatial reasoning, and linguistic systems. Accordingly, textual geolocation, a task for retrieving the coordinates of a place from linguistic descriptions, requires computerized spatial inference and natural language understanding. Yet, machine-based textual geolocation is currently limited, mainly due to the lack of rich geo-textual datasets necessitated to train natural language models that, in-turn, cannot adequately interpret the language-based expressions. These limitations are intensified in morphologically rich and resource-poor languages, such as Hebrew. This study aims to analyze and understand the linguistic systems used for place descriptions in Hebrew, later to be used to train machine learning natural language models. A novel crowdsourced geo-textual dataset is developed, composed of 5,695 written place descriptions provided by 1,554 native Hebrew speakers. All place descriptions rely on memory only, which increases spatial vagueness and requires referring expression resolution. Qualitative linguistic analysis of place descriptions shows that geospatial reasoning is greatly used in Hebrew, while empirical analysis with textual geolocation engines indicates that literal descriptions pose challenges for existing methods, as they require real understanding of space and geospatial references and cannot simply be geolocated by matching gazetteer with textual geo-entity extractions. The findings offer improved understanding of the challenges entailed in natural language processing of Hebrew geolocation, contributing to formalizing computerized systems used in future machine learning models for complex geographic information retrieval tasks.

Describing where a place is situated is an innate communication skill that relies on spatial cognition, spatial reasoning, and linguistic systems. Accordingly, textual geolocation, a task for retrieving the coordinates of a place from linguistic descriptions, requires computerized spatial inference and natural language understanding. Yet, machine-based textual geolocation is currently limited, mainly due to the lack of rich geo-textual datasets necessitated to train natural language models that, in-turn, cannot adequately interpret the language-based expressions. These limitations are intensified in morphologically rich and resource-poor languages, such as Hebrew. This study aims to analyze and understand the linguistic systems used for place descriptions in Hebrew, later to be used to train machine learning natural language models. A novel crowdsourced geo-textual dataset is developed, composed of 5,695 written place descriptions provided by 1,554 native Hebrew speakers. All place descriptions rely on memory only, which increases spatial vagueness and requires referring expression resolution. Qualitative linguistic analysis of place descriptions shows that geospatial reasoning is greatly used in Hebrew, while empirical analysis with textual geolocation engines indicates that literal descriptions pose challenges for existing methods, as they require real understanding of space and geospatial references and cannot simply be geolocated by matching gazetteer with textual geo-entity extractions. The findings offer improved understanding of the challenges entailed in natural language processing of Hebrew geolocation, contributing to formalizing computerized systems used in future machine learning models for complex geographic information retrieval tasks. [ABSTRACT FROM AUTHOR]

Case-based reasoning (CBR) simplifies knowledge acquisition and is suitable for researching complex geographical problems. However, CBR analyses of land-use changes are difficult to apply in the study of urban growth due to shortcomings in the case structure and model algorithms. In response, this study proposes a three-step urban-growth simulation model based on CBR (UGSCBR). First, to adapt the CBR to an urban-growth simulation process, the characteristics of regional differentiation in geographical spaces are determined. Second, a comprehensive retrieval method is developed that improves upon traditional case-retrieval methods by giving full play to the comprehensive function of each component of the case. Third, a quantity demand constraint indirectly adds a time factor to solve the initial blurriness of the traditional CBR-inference cycle. Taking Jixi city as the research area, we test the accuracy of the proposed model. The total accuracy of simulation results is 95.4%, and the Kappa is 87.4%. The figure of merit and Mathews correlation coefficient are 0.151 and 0.23, respectively, indicating that the model can meet the application requirements. The results show that the UGSCBR model has strong flexibility and simplicity, and it provides an effective prediction method for urban growth.

This paper is concerned with interpretation of natural language place descriptions, as they are a rich source of geographic information. A place description is interpreted by matching geographic entities occurring in the text against the OpenStreetMap (OSM) database. This paper is mainly concerned with interpretation of paraphrased places, i.e., entities for which no name is given and which may only by described. Our objective is to determine suitable entity types that allow querying the OpenStreetMap database for the respective place. For example, if we wish to identify a place to eat, we have to check for entities of an a-priori unknown type (cafe, restaurant, etc.). Challenges arise from the open-endedness of language, its ambiguity, and context-sensitivity as well as from mismatches between human conceptualization of place and database ontologies. The contributions of this paper are, first, to present a hard problem that is key to geo-information retrieval beyond named entities. Second, we propose context-sensitive methods for identifying place types based on semantic word similarity. We evaluate the methods on text extracted from Wikipedia and travel blogs, revealing their contribution to advancing automated interpretation of place descriptions to paraphrased places. [ABSTRACT FROM AUTHOR]

Case-based reasoning (CBR) simplifies knowledge acquisition and is suitable for researching complex geographical problems. However, CBR analyses of land-use changes are difficult to apply in the study of urban growth due to shortcomings in the case structure and model algorithms. In response, this study proposes a three-step urban-growth simulation model based on CBR (UGSCBR). First, to adapt the CBR to an urban-growth simulation process, the characteristics of regional differentiation in geographical spaces are determined. Second, a comprehensive retrieval method is developed that improves upon traditional case-retrieval methods by giving full play to the comprehensive function of each component of the case. Third, a quantity demand constraint indirectly adds a time factor to solve the initial blurriness of the traditional CBR-inference cycle. Taking Jixi city as the research area, we test the accuracy of the proposed model. The total accuracy of simulation results is 95.4%, and the Kappa is 87.4%. The figure of merit and Mathews correlation coefficient are 0.151 and 0.23, respectively, indicating that the model can meet the application requirements. The results show that the UGSCBR model has strong flexibility and simplicity, and it provides an effective prediction method for urban growth. [ABSTRACT FROM AUTHOR]

Spatial language incorporates descriptions of locations, routes, and landscapes, and is used by humans daily. Research has addressed a wide range of aspects of spatial language, including its form; the ways in which it is selected and applied; and cognitive, geometric, and functional factors affecting its use. Furthermore, much work has been done on the automation of spatial language extraction, analysis, interpretation, and generation. To introduce the Special Issue on this broad topic, this paper reviews spatial language research framed by an extension to the well-known semantic triangle, the "spatial semantic pyramid," which represents both human spatial language and relevant computational research. By introducing it, we hope to stimulate discussion about gaps and future directions in this important research field. [ABSTRACT FROM AUTHOR]

The availability of videos has grown rapidly in recent years. Finding and browsing relevant information to be automatically extracted from videos is not an easy task, but today it is an indispensable feature due to the immense number of digital products available. In this paper, we present a system which provides a process to automatically extract information from videos. We describe a system solution that uses a re-trained OpenNLP model to locate all the places and famous people included in a specific video. The system obtains information from the Google Knowledge Graph related to relevant named entities such as places or famous people. In this paper we will also present the Automatic Georeferencing Video (AGV) system developed by RAI (Radiotelevisione italiana, which is the national public broadcasting company of Italy, owned by the Ministry of Economy and Finance) Teche for the European Project "La Città Educante" (The Educating City: teaching and learning processes in cross-media ecosystem) Our system contributes to The Educating City project by providing the technological environment to create statistical models for automatic named entity recognition (NER), and has been implemented in the field of education, in Italian initially. The system has been applied to the learning challenges facing the world of educational media and has demonstrated how beneficial combining topical news content with scientific content can be in education. [ABSTRACT FROM AUTHOR]

Many natural language tasks related to geographic information retrieval (GIR) require toponym recognition, and identifying Chinese toponyms from social media messages to share real-time information is a critical problem for many practical applications, such as natural disaster response and geolocating. In this article, we focused on toponym recognition from social media messages in Chinese. While existing off-the-shelf Chinese named entity recognition (NER) tools could be applied to identify toponyms, these approaches cannot address a variety of language irregularities taken from social media messages, including location name abbreviations, informal sentence structures and combination toponyms. We present a deep neural network named BERT-BiLSTM-CRF, which extends a basic bidirectional recurrent neural network model (BiLSTM) with the pretraining bidirectional encoder representation from transformers (BERT) representation to handle the toponym recognition task in Chinese text. Using three datasets taken from lists of alternative location names, the experimental results showed that the proposed model can significantly outperform previous Chinese NER models/algorithms and a set of state-of-the-art deep learning models. [ABSTRACT FROM AUTHOR]

References to geographic locations are common in text data sources including social media and web pages. They take different forms from simple place names to relative expressions that describe location through a spatial relationship to a reference object (e.g. the house beside the Waikato River). Often complex, multi-word phrases are employed (e.g. the road and railway cross at right angles; the road in line with the canal) where spatial relationships are communicated with various parts of speech including prepositions, verbs, adverbs and adjectives. We address the problem of automatically detecting relative geospatial location descriptions, which we define as those that include spatial relation terms referencing geographic objects, and distinguishing them from non-geographical descriptions of location (e.g. the book on the table). We experiment with several methods for automated classification of text expressions, using features for machine learning that include bag of words that detect distinctive words, word embeddings that encode meanings of words and manually identified language patterns that characterise geospatial expressions. Using three data sets created for this study, we find that ensemble and meta-classifier approaches, that variously combine predictions from several other classifiers with data features, provide the best F-measure of 0.90 for detecting geospatial expressions. [ABSTRACT FROM AUTHOR]

We present a new dataset for the task of toponym resolution in digitized historical newspapers in English. It consists of 343 annotated articles from newspapers based in four different locations in England (Manchester, Ashton-under-Lyne, Poole and Dorchester), published between 1780 and 1870. The articles have been manually annotated with mentions of places, which are linked—whenever possible—to their corresponding entry on Wikipedia. The dataset consists of 3,364 annotated toponyms, of which 2,784 have been provided with a link to Wikipedia. The dataset is published in the British Library shared research repository, and is especially of interest to researchers working on improving semantic access to historical newspaper content.

We present a new dataset for the task of toponym resolution in digitized historical newspapers in English. It consists of 343 annotated articles from newspapers based in four different locations in England (Manchester, Ashton-under-Lyne, Poole and Dorchester), published between 1780 and 1870. The articles have been manually annotated with mentions of places, which are linked—whenever possible—to their corresponding entry on Wikipedia. The dataset consists of 3,364 annotated toponyms, of which 2,784 have been provided with a link to Wikipedia. The dataset is published in the British Library shared research repository, and is especially of interest to researchers working on improving semantic access to historical newspaper content. [ABSTRACT FROM AUTHOR]

Several tasks related to geographical information retrieval and to the geographical information sciences involve toponym matching, that is, the problem of matching place names that share a common referent. In this article, we present the results of a wide-ranging evaluation on the performance of different string similarity metrics over the toponym matching task. We also report on experiments involving the usage of supervised machine learning for combining multiple similarity metrics, which has the natural advantage of avoiding the manual tuning of similarity thresholds. Experiments with a very large dataset show that the performance differences for the individual similarity metrics are relatively small, and that carefully tuning the similarity threshold is important for achieving good results. The methods based on supervised machine learning, particularly when considering ensembles of decision trees, can achieve good results on this task, significantly outperforming the individual similarity metrics.

In recent years, researches related to Geographic Information Retrieval Systems as a specific field of Information Retrieval has continued to attract the attention of the research community by holding several assessing forums. However, these forums provide sets of tests comprised of text documents and queries that are ready to evaluate non‐interactive systems. This framework reduces the possibilities of carrying out a more thorough evaluation of these systems because it is not considering several important features such as the diversity provided by different information sources or the human‐computer interaction. The aim of this paper is to describe a new approach to evaluate interactive Geographic Information Retrieval Systems, which main novelty is to consider the user’s knowledge generated by the human‐computer interaction as well as the spatial information provided by different data sources. The proposed method will require generating a set of tests from three main data sources (Geonames, Wikipedia, and OpenStreetMap), as well as a set of queries that will consist of a tuple of three components: the object type, the spatial relationship and the geographic object. As a result, the proposed evaluation approach integrates the two most commonly used strategies to evaluate IR systems, which are focused on the system and the end user, by applying several user satisfaction techniques and usability tests. As a main conclusion, we pointed out that the evaluation process of Geographic Information Retrieval systems should consider the user’s knowledge generated by the human-computer interaction as well as the spatial information provided by different and heterogeneous data sources.

Many solutions for coarse geolocating of users at the time they post a message exist. However, for many important applications, like traffic monitoring and event detection, finer geolocation at the level of city neighborhoods, i.e., at a sub-city level, is needed. Data-driven approaches often do not guarantee good accuracy and efficiency due to the higher number of sub-city level positions to be estimated and the low availability of balanced and large training sets. We claim that external information sources overcome limitations of data-driven approaches in achieving good accuracy for sub-city level geolocation and we present a knowledge-driven approach achieving good results once the reference area of a message is known. Our algorithm, called Sherloc, exploits toponyms in the message, extracts their semantic from a geographic gazetteer, and embeds them into a metric space that captures the semantic distance among them. We identify the semantically closest toponyms to a message and then cluster them with respect to their spatial locations. Sherloc requires no prior training, it can infer the location at sub-city level with high accuracy, and it is not limited to geolocating on a fixed spatial grid. [ABSTRACT FROM AUTHOR]

The two-and-a-half day workshop in Santa Barbara (December 2006) focused on the role of digital gazetteers in geo-referencing applications, starting with overviews of the state-of-the-art and current activities and leading to a consensus on the opportunities and directions for collaboration and the advancement of a research and practice agenda. The workshop was convened by the National Center for Geographic Information & Analysis (NCGIA) and the Redlands Institute, with sponsorship from the National Geospatial-Intelligence Agency.

The two-and-a-half day workshop (Dec 2006, in Santa Barbara CA) focused on the role of digital gazetteers in georeferencing applications, with overviews of the state-of-the-art and current activities and discussions on opportunities and directions for collaboration and the advancement of a research and practice agenda. The workshop was convened by the National Center for Geographic Information & Analysis (NCGIA) at the University of California at Santa Barbara and the Redlands Institute. It was sponsored by the National Geospatial-Intelligence Agency. This document contains the list of participants, submitted position papers, and the meeting agenda.

Geographic text analysis (GTA) research in the digital humanities has focused on projects analyzing modern English-language corpora. These projects depend on temporally specific lexicons and gazetteers that enable place name identification and georesolution. Scholars working on the early modern period (1400–1800) lack temporally appropriate geoparsers and gazetteers and have been reliant on general purpose linked open data services like Geonames. These anachronistic resources introduce significant information retrieval and ethical challenges for early modernists. Using the geography entries of the canonical eighteenth-century Encyclopédie, we evaluate rule-based named entity recognition (NER) systems to pinpoint areas where they would benefit from adjustments for processing historical corpora. As we demonstrate, annotating nested and extended place information is one way to improve early modern GTA. Working with Enlightenment sources also motivates a critique of the landscape of digital geospatial data. [ABSTRACT FROM AUTHOR]

People often communicate with reference to informally agreed places, such as "the city centre". However, views of the spatial extent of such areas may vary, resulting in imprecise regions. We compare perceptions of Sheffield's City Centre from a street survey to extents derived from various web-based sources. Such automated approaches have advantages of speed, cost and repeatability. We show that footprints from web sources are often in concordance with models derived from more labour-intensive methods. Notable exceptions however were found with sources advertising or selling residential property. Agreement between sources was measured by aggregating them to identify locations of consensus. [ABSTRACT FROM AUTHOR]

Geoparser is a fundamental component of a Geographic Information Retrieval (GIR) geoparser, which performs toponym recognition, disambiguation, and geographic coordinate resolution from unstructured text domain. However, geoparsing of news articles which report several events across many place-mentions in the document are not yet adequately handled by regular geoparser, where the scope of resolution is either toponym-level or document-level. The capacity to detect multiple events and geolocate their true coordinates along with their numerical arguments is still missing from modern geoparsers, much less in Indonesian news corpora domain. We propose an event geoparser model with three stages of processing, which tightly integrates event extraction model into geoparsing and provides precise event-level resolution scope. The model casts the geotagging and event extraction as sequence labeling and uses LSTM-CRF inferencer equipped with features derived using Aggregated Topic Model from a large corpus to increase the generalizability. Throughout the proposed workflow and features, the geoparser is able to significantly improve the identification of pseudo-location entities, resulting in a 23.43% increase for weighted F1 score compared to baseline gazetteer and POS Tag features. As a side effect of event extraction, various numerical arguments are also extracted, and the output is easily projected to a rich choropleth map from a single news document.

Geographic information retrieval (GIR) is concerned with returning information in response to an information need, typically expressed in terms of a thematic and spatial component linked by a spatial relationship. However, evaluation initiatives have often failed to show significant differences between simple text baselines and more complex spatially enabled GIR approaches. We explore the effectiveness of three systems (a text baseline, spatial query expansion, and a full GIR system utilizing both text and spatial indexes) at retrieving documents from a corpus describing mountaineering expeditions, centred around fine grained toponyms. To allow evaluation, we use user generated content (UGC) in the form of metadata associated with individual articles to build a test collection of queries and judgments. The test collection allowed us to demonstrate that a GIR-based method significantly outperformed a text baseline for all but very specific queries associated with very small query radii. We argue that such approaches to test collection development have much to offer in the evaluation of GIR.

Abstract Today most of the users on the move require contextualized local and georeferenced information. Several solutions aim to meet these trends, thus assisting users and satisfying their needs and preferences, such as virtual assistants and Location-Aware Recommender Systems (LARS), both in commercial and research literature. However, general purpose virtual assistants usually have to manage large domains, dealing with big amounts of data and online resources, losingfocus on more specific requirements and local information. On the other hand, traditional recommender systems are based on filtering techniques and contextual knowledge, and they usually do not rely on Natural Language Processing (NLP) features on users' queries, which are useful to understand and contextualize users' necessities on the spot. Therefore, comprehending the actual users' information needs and other key information that can be included in the user query, such as geographical references, is a challenging task which is not yet fully accomplished by current state-of-the-art solutions. In this paper, we propose Paval (Location-Aware Virtual Personal Assistant 2 2 The name Paval is chosen as a permutation of the initials of "Location-aware virtual personal assistant".), a semantic assisting engine for suggesting local points of interest (POIs) and services by analyzing users' natural language queries, in order to estimate the information need and potential geographic references expressed by the users. The system exploits NLP and semantic techniques providing as output recommendations on local geolocated POIs and services which best match the users' requests, retrieved by querying our semantic Km4City Knowledge Base. The proposed system is validated against the most popular virtual assistants, such as Google Assistant, Apple Siri and Microsoft Cortana, focusing the assessment on the request of geolocated POIs and services, showing very promising capabilities in successfully estimating the users' information needs and multiple geographic references. [ABSTRACT FROM AUTHOR]

In this paper a system providing an efficient integration between Content-Based Image Retrieval (CBIR) and Geographic Information Retrieval (GIR) is presented. Over the years, many CBIR systems have been proposed to give a solution for an efficient use of multimedia/visual contents and other issues as performance, quality of retrieval, data heterogeneity, and multimodal information integration. The aim of the proposed approach is to prove that the use of geographic data can improve the results obtained by an image matching system based only on visual data. Our framework is composed of three parts, each of them described in detail in this paper: the first part is dedicated to CBIR, with an experimental comparison of a large number of different multimedia features to choose the one to use in the system implementation; in the second part the methodology to integrate geographic and multimedia data is showed; in the last part is presented a GIR system implementation using a “points of interest” search. An Android application has been developed for the client-side using Apache Solr as server side provider for the information retrieval functionalities. An experimental evaluation is carried out to demonstrate the effective improvement given by the combination of geographic and multimedia data. Our results have been obtained using a real dataset composed of artworks located in Naples’s museums. [ABSTRACT FROM AUTHOR]

Several tasks related to geographical information retrieval and to the geographical information sciences involve toponym matching, that is, the problem of matching place names that share a common referent. In this article, we present the results of a wide-ranging evaluation on the performance of different string similarity metrics over the toponym matching task. We also report on experiments involving the usage of supervised machine learning for combining multiple similarity metrics, which has the natural advantage of avoiding the manual tuning of similarity thresholds. Experiments with a very large dataset show that the performance differences for the individual similarity metrics are relatively small, and that carefully tuning the similarity threshold is important for achieving good results. The methods based on supervised machine learning, particularly when considering ensembles of decision trees, can achieve good results on this task, significantly outperforming the individual similarity metrics. [ABSTRACT FROM AUTHOR]

How do people perceive landscapes? What elements of the landscape do they identify as characteristic of a landscape? And how can we arrive at descriptions, and ultimately representations that better reflect people’s notions of landscapes? In this study, we collected landscape descriptions from five landscape types at 10 study sites in Switzerland. For each site, we collected data from three sources: free lists with participants, hiking blogs, and Flickr tags. Free lists were obtained through on-site interviews with visitors, hiking blogs were gathered by focused crawling of web content, and Flickr tags were selected based on spatial footprints obtained from the hiking blogs. We quantitatively compared landscape descriptions between data sources and landscape types using cosine similarity. We found that descriptions from the same data source were significantly more similar, irrespective of landscape type. Descriptions from the same landscape type were more similar, but only within the same data source. Through a qualitative analysis of different aspects of landscape in our content, we found that each data source offered a different distribution of landscape aspects. For example, while Flickr tags contained high proportions of toponyms, they contained little content relating to sense of place. In contrast, hiking blogs contained more information about sense of place. Our approach combining these varied textual sources thus offers a more holistic view on landscapes. This study constitutes a step toward extracting semantically rich descriptions of landscapes from a variety of sources and using this information to distinguish different landscapes, with potential applications for landscape monitoring and management. [ABSTRACT FROM AUTHOR]

Data-driven GIScience shows a growing interest in making spatial information from large text data. In this paper, we quantify and thus evaluate the relation between text frequency and properties of the outer-text, geographic setting by comparing text frequencies of mountain names to the respective geomorphometric characteristics. We focus on some 2000 unique mountain names that appear some 50,000 times in a large compilation of texts on Swiss alpine history. The results on the full data set suggest only a weak relation: only 5-10% of the variation in the text frequency being explained by the respective geomorphometric characteristics. However, an analysis of multiple scales allows us to identify a Simpson’s Paradox. What appears to be ‘noise’ in the analysis of all mountains in the whole of Switzerland shows significant local signals. Small spatial extents, found all over Switzerland, can show considerably strong correlations between text frequency and spatial prominence, with up to 90% of the total variation explained. We argue that our findings have practical implications for data-driven GIScience. Retrieving meaningful spatial information from text might only be possible if the spatial scale of analysis reflects the spatial scale described in the input text documents. [ABSTRACT FROM AUTHOR]