<!DOCTYPE article PUBLIC "-//NLM//DTD JATS (Z39.96) Journal Archiving and Interchange DTD v1.0 20120330//EN" "JATS-archivearticle1.dtd">
<article xmlns:xlink="http://www.w3.org/1999/xlink">
  <front>
    <journal-meta>
      <journal-title-group>
        <journal-title>Enschede, The
Netherlands
* Corresponding author.
†These authors contributed equally.
$ george.a.wright@qmul.ac.uk (G. A. Wright); m.purver@qmul.ac.uk (M. Purver)
 http://georgeaw.com/ (G. A. Wright); http://www.eecs.qmul.ac.uk/~mpurver (M. Purver)</journal-title>
      </journal-title-group>
    </journal-meta>
    <article-meta>
      <title-group>
        <article-title>The Geography of Temperature Space</article-title>
      </title-group>
      <contrib-group>
        <contrib contrib-type="author">
          <string-name>George A. Wright</string-name>
          <xref ref-type="aff" rid="aff1">1</xref>
          <xref ref-type="aff" rid="aff2">2</xref>
        </contrib>
        <contrib contrib-type="author">
          <string-name>Matthew Purver</string-name>
          <xref ref-type="aff" rid="aff0">0</xref>
          <xref ref-type="aff" rid="aff2">2</xref>
        </contrib>
        <aff id="aff0">
          <label>0</label>
          <institution>Department of Knowledge Technologies, Jožef Stefan Institute</institution>
          ,
          <addr-line>Ljubljana</addr-line>
          ,
          <country country="SI">Slovenia</country>
        </aff>
        <aff id="aff1">
          <label>1</label>
          <institution>School of Computing and Mathematical Sciences</institution>
          ,
          <addr-line>Birkbeck</addr-line>
          ,
          <institution>University of London</institution>
          ,
          <country country="UK">United Kingdom</country>
        </aff>
        <aff id="aff2">
          <label>2</label>
          <institution>School of Electronic Engineering and Computer Science, Queen Mary University of London</institution>
          ,
          <country country="UK">United Kingdom</country>
        </aff>
      </contrib-group>
      <pub-date>
        <year>2024</year>
      </pub-date>
      <volume>000</volume>
      <fpage>0</fpage>
      <lpage>0002</lpage>
      <abstract>
        <p>This paper takes as its basis, Gärdenfors' [1] description of conceptual spaces and sketches an initial geography of the conceptual space for temperature. It highlights important diferences between the space's constituents, including: how category membership difers for concepts along diferent parts of the dimension; how comparatives and compounds are formed; how context afects concept location and behaviour; and how non-literal usage allows concepts to be used outside of their conceptual space or in a diferent way within their own conceptual space. The purpose of this tour of temperature space is to aid the practical implementation of semantic representations based upon conceptual spaces.</p>
      </abstract>
      <kwd-group>
        <kwd>eol&gt;Conceptual Spaces</kwd>
        <kwd>Temperature</kwd>
        <kwd>Adjectives</kwd>
        <kwd>Comparatives</kwd>
        <kwd>Metaphor</kwd>
        <kwd>Natural Language</kwd>
      </kwd-group>
    </article-meta>
  </front>
  <body>
    <sec id="sec-1">
      <title>1. Introduction</title>
      <p>
        In the conceptual spaces approach to the representation of meaning, concepts and their instances occupy
spaces composed of quality dimensions. The proximity of objects along those dimensions equates to
semantic similarity. The approach allows for graded and ambiguous class membership. Gärdenfors [
        <xref ref-type="bibr" rid="ref1 ref2">1, 2</xref>
        ]
describes how conceptual spaces can be used to describe a wide range of concepts involved in perception
and in natural language semantics and provides example spaces such as for colour, shape, and kinship
relations.
      </p>
      <p>An important yet seemingly simple conceptual space is that for temperature. Perception of
temperature is important for survival (and comfort) so is also a frequent topic of conversation. Although
temperatures occupy just one dimension — a number line — close consideration of English words for
temperature unveils a complex arrangement of that space and a range of types of concept within it.</p>
      <p>This paper forms a guide through temperature space which highlights the issues that need to be
considered when developing a practical implementation of conceptual spaces. It begins with a summary
of relevant literature on the representation of adjective semantics along dimensions. The constituent
concepts of temperature space are then introduced, followed by a discussion of their positions, compound
forms, and behaviour in diferent contexts.</p>
    </sec>
    <sec id="sec-2">
      <title>2. Conceptual Spaces</title>
      <p>
        Conceptual Spaces order concepts along quality dimensions, with similar concepts close to one another.
Gärdenfors [
        <xref ref-type="bibr" rid="ref1">1</xref>
        ] pairs conceptual spaces with prototype theory. Each concept has a prototype which
occupies a point in the space. Objects closer to the prototype are more likely to be members of its class
and, because they are more similar to the prototype, have higher typicality. Voronoi tessellation of a
space, such that boundaries are drawn equidistant from each prototype, divides a space into a set of
discrete regions, each corresponding to a concept.
      </p>
      <sec id="sec-2-1">
        <title>2.1. Conceptual Spaces for Adjectives</title>
        <p>
          Paradis [
          <xref ref-type="bibr" rid="ref3">3</xref>
          ] identifies three classes of adjectives according to their gradability and boundedness:
Scalar adjectives These are adjectives which exist on a scale. The scale contains two antonyms
(such as tall and short); has a region between the two antonyms where neither applies; and is
open-ended. Because the adjectives exist along a scale, they have comparative and superlative
forms and can be combined with scalar modifiers ( very tall). Because the scale is open-ended,
adjectives cannot be combined with totality modifiers (* completely tall).
        </p>
        <p>Extreme adjectives These also exist on a scale with two antonyms (such as terrible and excellent).</p>
        <p>But the scale is not open-ended; instead the adjectives exist at the extremes of the scale. For this
reason, they cannot combine with scalar modifiers (* very terrible), but can combine with totality
modifiers ( completely terrible). There is disagreement over whether or not these adjectives have a
comparative form.</p>
        <p>Limit adjectives These are adjectives (such as dead) which do not exist on a scale and do not have
graded membership. They do not have a comparative form (*deader) and combine only with
totality modifiers ( completely dead, almost dead).</p>
        <p>
          A large number of basic adjectives fit within the scalar category. According to Barsalou [
          <xref ref-type="bibr" rid="ref4">4</xref>
          ], when
concepts such as these exist along a single dimension (or ideal), the typicality of instances increases as
they take a more extreme position along it. Since the dimensions that scalar adjectives belong to are
unbounded, this makes questionable the existence of prototypes for these adjectives.
        </p>
        <p>
          But, Hampton [
          <xref ref-type="bibr" rid="ref5">5</xref>
          ] and Kamp and Partee [
          <xref ref-type="bibr" rid="ref6">6</xref>
          ] point out that membership of a given category is often
ill-defined without a specific context. It would be reasonable for tall to refer to diferent heights when
describing mountains, buildings, or people. Prototypes for tall and short would therefore occupy
diferent points in a conceptual space depending on context. Nonetheless, according to Hampton, the
typicality of an instance of tall would still be unbounded: someone who is taller than the prototype
would still have a high typicality and high degree of membership. Hampton does not consider phrases
such as unusually tall, the existence of which suggests that an object can have a value so far along a
dimension that it ceases to be typical.
        </p>
      </sec>
      <sec id="sec-2-2">
        <title>2.2. Conceptual Spaces for Comparatives</title>
        <p>As described above, only scalar adjectives have comparative forms. According to Gärdenfors [2, p.136],
the meaning of comparatives depends upon the dimensionality of the space.</p>
        <p>
          Comparison of objects along a single dimension, such as height, requires consideration of their
position along the dimension, but not the position of the prototype. Dietz [
          <xref ref-type="bibr" rid="ref7">7</xref>
          ], defines comparatives as
asymmetric and transitive relations:  can be taller than , and  taller than  and so on until the
objects under discussion are far away from the prototype for tall.
        </p>
        <p>
          On the other hand, comparison of objects within a multi-dimensional space, such as colour, involves
comparing their proximity to the prototype. For  to be redder than ,  must be closer than  to the
prototype for red. It also seems meaningless to use the term redder if neither  nor  are at all red [
          <xref ref-type="bibr" rid="ref7">7</xref>
          ].
        </p>
      </sec>
    </sec>
    <sec id="sec-3">
      <title>3. Temperature Space and its Inhabitants</title>
      <p>
        Temperature space has one dimension [
        <xref ref-type="bibr" rid="ref1">1</xref>
        ]. For the most part, it contains scalar adjectives just as
height does. Indeed temperature can be mapped onto height words high and low via conceptual
metaphor [
        <xref ref-type="bibr" rid="ref8">8</xref>
        ]. But, temperature contains multiple concepts, each with diferent behaviour, whose
presence complicates pragmatic considerations of word choice.
      </p>
      <sec id="sec-3-1">
        <title>3.1. Temperature Concepts and their Arrangement</title>
        <p>Temperature contains the antonyms cold and hot. But depending on context, a greater variety of
adjectives are also available. The words cool and warm act as gentler compliments, and sometimes mild
can also be used. Along the dimension of temperature, the antonyms cold and hot sit opposite one
another, leaning towards the two extremes; the antonyms cool and warm are closer to the middle; and
mild sits in the middle.</p>
        <p>The antonym of mild, is extreme, which can occupy either end of temperature space. Alternatively,
mild and extreme can sit along a mildness/extremeness dimension which maps onto temperature in
the same way that the integers map onto the natural numbers via the absolute value function. Figure 1
shows such an arrangement, with temperature concepts ordered along a coldness/hotness dimension
that has mild at its origin. The Temperature line is folded at the origin in order to demonstrate its
mapping onto mildness/extremeness.</p>
        <p>Concepts not shown in figure 1 include limit adjectives such as boiling and freezing, which have
strict definitions when used literally, and adjectives such as lukewarm and tepid, which are restricted to
certain contexts. It makes less sense to place these adjectives along a scale.</p>
      </sec>
      <sec id="sec-3-2">
        <title>3.2. Temperature Concepts and their Behaviour</title>
        <p>Hot and cold are unbounded scalar adjectives which behave analogously to tall and short in height
space. The typicality of instances of these concepts increases as they get further along the dimension,
even as they surpass the prototype value of the concept.</p>
        <p>On the other hand, temperatures become more typical instances of mild as they get closer to its
prototype. This behaviour is more typical of concepts in a multi-dimensional space, but mild is
perhaps not strictly native to temperature space, but rather to mildness/extremeness space where
its behaviour is more similar to that of short in height space.</p>
        <p>Warm and cool seem to be in between these two types of behaviour. They are not strictly bounded,
but at distances increasingly far from their prototype, alternative concepts become more salient.</p>
        <p>
          It is not clear that hard boundaries should be drawn between each of these concepts with Voronoi
tessellation as suggested by Gärdenfors [
          <xref ref-type="bibr" rid="ref1">1</xref>
          ]: some temperatures can be described as either hot or warm
and others as either warm or mild.
        </p>
      </sec>
      <sec id="sec-3-3">
        <title>3.3. Comparatives and Modifiers</title>
        <p>Because temperature concepts for the most part have the behaviour of scalar adjectives, they have
comparative forms and can combine with scalar modifiers.</p>
        <sec id="sec-3-3-1">
          <title>3.3.1. Comparatives</title>
          <p>To determine semantically valid and pragmatically appropriate comparative concepts for the description
of a given situation, both the ordering of points in temperature space and their proximity to prototypes
is important.</p>
          <p>To describe a point x as hotter or colder, less hot, or less cold, than point y, only the relative position
of the two points along the temperature dimension is required in order to produce a logically true
predicate.</p>
          <p>But further consideration of their position in temperature is necessary in order to produce a
pragmatically sensible sentence of English. The situation is further complicated by the presence of
warm and cool, which provide a further array of comparatives to choose from.</p>
          <p>The phrases  is hotter than ,  is warmer than , and  is less cold than  all correspond with the
inequality  &gt; , but to English speakers they each have a slightly diferent meaning, which relates to
the original adjective’s location along the temperature dimension.</p>
          <p>For example, in figure 1: a is strictly speaking hotter than e, but it is better to describe it as less cold
than e; similarly c is strictly speaking hotter than a, but it is better to describe it as warmer than a</p>
          <p>Adjectives’ comparative forms (hotter, colder) are equivalent to prefixing the adjective with more. If A
is more X than B then A is further out along the unbounded dimension than B. If A is less X than B then
A is closer to the centre of the space than B. In the case of mild, A is milder/more mild than B means
A is closer to the prototype for mild. On the other hand, A is less mild than B means that A is further
away than B (in either direction) from the prototype for mild. Figure 2 illustrates these diferences in
behaviour. While warm and hot have positive polarities, and cool and cold have negative polarities, mild
has neither.</p>
        </sec>
        <sec id="sec-3-3-2">
          <title>3.3.2. Modifiers</title>
          <p>An adjective’s polarity also governs the interpretation of grammatical modifiers such as very, quite, and
extremely. When conjoined with an adjective they form a compound with a prototype that is shifted
from the location of the lone adjective (see figure 3).</p>
          <p>Words such as very and extremely, which increase the magnitude of the adjective they are attached
to, follow the same rules as the comparative more: when attached to a positive adjective such as warm,
they shift meaning further in the positive direction; when attached to a negative adjective such as cold,
they shift meaning further in the negative direction; when attached to an adjective at the space’s origin
(mild), they shift meaning closer to the origin and contract its boundaries: extremely mild has a more
limited purview than mild.</p>
          <p>Words such as quite, slightly, or a bit have the opposite efect and are therefore more akin to the
comparative less. Since they shift meaning away from the extreme of the prototype, when combined
with mild two prototypes are formed: one in the colder half of space, and one in the warmer half. Quite
mild therefore occupies a non-convex region of temperature space. This is a further sign that mild does
not really belong in temperature space, but is temporarily introduced in certain contexts.</p>
          <p>When an unbounded concept such as hot has a magnitude-increasing modifier such as extremely
applied to it, the resulting compound is also unbounded. When it has a magnitude-decreasing modifier
such as quite applied to it, the resulting compound becomes bounded. The phrase quite hot only has
limited applicability before hot or very hot become more appropriate descriptions. To say the surface
of the Sun is quite hot would be acceptable only in a non-literal, sarcastic context, or perhaps when
comparing with even hotter stars.</p>
        </sec>
      </sec>
      <sec id="sec-3-4">
        <title>3.4. Context and Space Warping</title>
        <p>
          As mentioned by Hampton [
          <xref ref-type="bibr" rid="ref5">5</xref>
          ], the precise meaning of an adjective is often unclear without a context.
The prototypes of basic temperature adjectives can therefore be quite diferent in contexts such as
the weather, dining, cookery, or cosmology. Within these contexts, there are further sub-contexts: a
warm winter is normally colder than a warm summer; a cold dish is normally warmer than a cold drink.
Meaning is also necessarily subjective, thus what counts as hot or cold will vary from person to person.
Figure 4 suggests approximate locations of temperature adjectives in broad contexts.
        </p>
        <p>The non-uniform warping of space caused by the context changes in Figure 4 cause hot and warm
to shift more than cool or cold. The limit adjectives boiling and freezing do not move at all unless
they are used for exaggeration as when describing the weather, in which case they adopt the graded
category membership of scalar adjectives. They could also shift if applied to more exotic contexts such
as chemicals other than water.</p>
        <p>
          Certain concepts are not shown in figure 4 and are not necessarily available in all contexts. For
example, mild is largely restricted to describing the weather, lukewarm to describing food, and tepid to
describing water [
          <xref ref-type="bibr" rid="ref9">9</xref>
          ]. Since lukewarm and tepid also have negative connotations (despite being in the
positive part of temperature space), they perhaps also belong to other dimensions which only become
relevant in particular contexts.
        </p>
      </sec>
      <sec id="sec-3-5">
        <title>3.5. Metaphor and Projection Across Spaces</title>
        <p>Not only can context shift the position of concepts within a space, it can also project them from one space
to another. This occurs when words from one conceptual space are used metaphorically to describe
entities from a diferent conceptual space. Temperature adjectives are commonly used metaphorically
for example in descriptions of spicy food or of people.</p>
        <p>But in metaphorical usages, only a subset of temperature adjectives are imported into their new
context: food’s degree of spiciness can be described as hot or mild, but not warm or cold; people’s
behaviour to others can be described as warm, lukewarm, or cold, but not hot or mild; and people’s
attractiveness can be described only as hot (or not).</p>
        <p>Adjectives can be used and re-used metaphorically. If someone were to describe the weather as
lukewarm, this would be an odd use of words since lukewarm is usually reserved for descriptions of
food. But the description could also be interpreted as a personification of the weather into a person who
is lukewarm, this in turn being a metaphorical description of someone with an unfriendly disposition.</p>
      </sec>
    </sec>
    <sec id="sec-4">
      <title>4. Conclusion: How to Implement Temperature Space</title>
      <p>Temperature is an everyday topic and its perception is fundamental to human existence. Ways of
describing temperature indicate a variety of types of conceptual behaviour that need to be covered by a
representation based on conceptual spaces, even in the case of a simple, unidimensional space.
Flexible prototype locations Adjectival concepts have prototype values, but these must be capable
of changing (potentially by orders of magnitude) depending on the context where they are used.</p>
      <p>Certain adjectives, such as lukewarm are also unavailable or less available in given contexts.
Flexible concept behaviour Not all adjectives within the same space have the same behaviour. In
temperature, most adjectives are scalar, but there are also limit adjectives. Limit adjectives can
adopt the behaviour of scalar or extreme adjectives if they are used metaphorically, for example
boiling weather has graded membership, whereas boiling water does not.</p>
      <p>Concept polarity As well as a prototype value, concepts should also be associated with a polarity,
or direction. This helps to determine how comparatives are formed and how modifiers afect
meaning. There are at least three types of behaviour in relation to comparatives and modifiers:
for positive adjectives (e.g. warm) more and very indicate higher temperatures; for negative
adjectives (e.g. cool) more and very indicate lower temperatures; for neutral adjectives (e.g. mild)
at the conceptual space’s origin, more and very indicate temperatures closer to the origin.
The applicability of Voronoi tessellation The existence of potentially overlapping concepts (e.g.
cold and cool or cool and mild), the introduction of compounds (e.g. quite cold, very cool), and the
metaphorical use of concepts (e.g. freezing) means that there is a plethora of ways to describe
the same temperature. It is therefore unclear if it is useful to divide a space into disjoint regions
belonging to separate concepts.</p>
      <p>Odd behaviour of non-native concepts A space can gain non-native concepts, even without
metaphorical usage, just as temperature space gains mild and extreme. These concepts or
compounds involving them can occupy non-convex regions and therefore have unusual
behaviour compared with other concepts.</p>
      <p>Exaggeration and metaphor Words can be used non-literally within a conceptual space for
exaggerated efect ( boiling weather); metaphorically in a foreign conceptual space (a warm person); and
can be projected metaphorically out of and back into their home conceptual space (lukewarm
weather).</p>
      <p>Other issues not considered in this paper include the problem of negation, and the association of
concepts across conceptual spaces, for example the varying association in diferent contexts between
hot or cold and good or bad. These important issues no doubt cause further complications.</p>
    </sec>
    <sec id="sec-5">
      <title>Acknowledgments</title>
      <p>This research has been supported by EPSRC grant EP/R513106/1. Many thanks also to the attendees of
the CAOS workshop for their questions and feedback.</p>
    </sec>
  </body>
  <back>
    <ref-list>
      <ref id="ref1">
        <mixed-citation>
          [1]
          <string-name>
            <given-names>P.</given-names>
            <surname>Gärdenfors</surname>
          </string-name>
          , Conceptual Spaces: The Geometry of Thought, The MIT Press,
          <year>2004</year>
          .
        </mixed-citation>
      </ref>
      <ref id="ref2">
        <mixed-citation>
          [2]
          <string-name>
            <given-names>P.</given-names>
            <surname>Gärdenfors</surname>
          </string-name>
          ,
          <source>The Geometry of Meaning: Semantics Based on Conceptual Spaces</source>
          , The MIT Press,
          <year>2014</year>
          .
        </mixed-citation>
      </ref>
      <ref id="ref3">
        <mixed-citation>
          [3]
          <string-name>
            <given-names>C.</given-names>
            <surname>Paradis</surname>
          </string-name>
          , Adjectives and boundedness,
          <source>Cognitive Linguistics 12</source>
          (
          <year>2013</year>
          )
          <fpage>47</fpage>
          -
          <lpage>64</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref4">
        <mixed-citation>
          [4]
          <string-name>
            <given-names>L. W.</given-names>
            <surname>Barsalou</surname>
          </string-name>
          ,
          <article-title>Ideals, central tendency, and frequency of instantiation as determinants of graded structure in categories</article-title>
          ,
          <source>Journal of Experimental Psychology: Learning, Memory, and Cognition</source>
          <volume>11</volume>
          (
          <year>1985</year>
          )
          <fpage>629</fpage>
          -
          <lpage>654</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref5">
        <mixed-citation>
          [5]
          <string-name>
            <given-names>J. A.</given-names>
            <surname>Hampton</surname>
          </string-name>
          , Typicality, graded membership, and vagueness,
          <source>Cognitive Science 31</source>
          (
          <year>2007</year>
          )
          <fpage>355</fpage>
          -
          <lpage>384</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref6">
        <mixed-citation>
          [6]
          <string-name>
            <given-names>H.</given-names>
            <surname>Kamp</surname>
          </string-name>
          ,
          <string-name>
            <given-names>B.</given-names>
            <surname>Partee</surname>
          </string-name>
          ,
          <source>Prototype theory and compositionality</source>
          ,
          <source>Cognition</source>
          <volume>57</volume>
          (
          <year>1995</year>
          )
          <fpage>129</fpage>
          -
          <lpage>191</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref7">
        <mixed-citation>
          [7]
          <string-name>
            <given-names>R.</given-names>
            <surname>Dietz</surname>
          </string-name>
          ,
          <article-title>Comparative concepts</article-title>
          ,
          <source>Synthese</source>
          (
          <year>2013</year>
          ).
        </mixed-citation>
      </ref>
      <ref id="ref8">
        <mixed-citation>
          [8]
          <string-name>
            <given-names>G.</given-names>
            <surname>Lakof</surname>
          </string-name>
          , M. Johnson, Metaphors We Live By, The University of Chicago Press,
          <year>2003</year>
          .
        </mixed-citation>
      </ref>
      <ref id="ref9">
        <mixed-citation>
          [9]
          <string-name>
            <given-names>Oxford</given-names>
            <surname>English</surname>
          </string-name>
          <string-name>
            <surname>Dictionary</surname>
          </string-name>
          , Oxford University Press,
          <year>2024</year>
          .
        </mixed-citation>
      </ref>
    </ref-list>
  </back>
</article>