Computing with Perceptions for the Linguistic Description of Complex

Phenomena through the Analysis of Time Series Data

A. Ramos-Soto, A. Bugar

ın and S. Barro

Research Center on Information Technologies (CiTIUS), University of Santiago de Compostela,

ua de Jenaro de la Fuente Dom

ınguez, s/n, 15782, Santiago de Compostela, Spain

1 RESEARCH PROBLEM

Nowadays, new technologies allow acquiring and

archiving vast volumes of data about time-evolving

phenomena in many crucial areas such as economy,

science, and industrial processes. Examples in econ-

omy include the evolution of every kind of econom-

ical indicators at local or global levels, like stock

funds, electricity/gas/water consumption, price of ba-

sic products, etc. In science, the amount of infor-

mation collected by researchers is overwhelming and

ever growing, including astronomical observations by

radio telescopes, space probes, etc. and data collected

from experiments in diverse scientiﬁc ﬁelds.

In order to be useful, this data must be exploited

and explained in an understandable way, reporting

facts, advice or commands to be performed that use

the available background knowledge about the phe-

nomenon under study. These objectives can only be

achieved by using natural language, especially if the

ﬁnal information is to be provided by a non-expert.

This is clear for example in the case of ﬁnancial news-

papers and scientiﬁc publications, in which data is not

simply made accessible or summarized as graphics

and tables, but arguments and conclusions need to be

explained using natural language.

However, there is a lack of tools and means for

processing and interpreting all this data using com-

puters. Any organization of data as provided by a

computer, either in a numerical, categorical, and/or

graphical form, is just a tool that can be employed by

human experts to produce an explanation in natural

language.

Understandable linguistic descriptions of phe-

nomena are provided by human experts, while com-

puters just provide ﬂexibility in storing and accessing

data. In fact, it is becoming easier and easier to col-

lect data, but providing a human being with expertise

on a certain domain remains difﬁcult and expensive.

This situation clearly poses a problem, since the ra-

tio data/human experts is growing dramatically as a

consequence. In summary, there is a clear need for

computational systems able to produce automatically

linguistic descriptions of data about phenomena.

More speciﬁcally, the task of generating easily un-

derstandable information for people using human lan-

guage has been addressed by two ﬁelds which, inde-

pendently until now, have researched the processes

this task involves: the natural language generation

(Reiter and Dale, 2000) and the linguistic descriptions

of data (Zadeh, 1996).

The natural language generation ﬁeld focuses its

efforts on automatically obtaining texts, with the pur-

pose of them being as much as possible indistinguish-

able from the ones created by humans. The linguistic

descriptions of data ﬁeld, which originates in the soft

computing domain, provides summaries or descrip-

tions from data sets using linguistic concepts which

deal with the imprecision and ambiguity of language

through the use of fuzzy sets.

In this context, we propose in this Ph.D. to re-

search on the linguistic descriptions of data ﬁeld, cov-

ering a group of soft computing-based concepts and

techniques, such as linguistic variables, fuzzy opera-

tors and quantiﬁcation methods. For instance, using

this kind of solutions, we can obtain quantiﬁed sen-

tences such as “most of the students are good” or “A

few days with high humidity the temperature is low”.

In fact, most of the approaches for building lin-

guistic descriptions described in the literature make

use of the concept of “quantiﬁed sentence”. In this

sense, the linguistic description approaches make use

of two different types of quantiﬁed sentences: type

I (“Q of X are A”), as in “several dogs are brown”,

and type II (“Q of D are A”), as in “a few young re-

searchers have published relevant papers”, where X is

a ﬁnite crisp set, Q is a linguistic quantiﬁer and A, D

are fuzzy properties deﬁned over X (Delgado et al.,

2014).

Despite its formal nature and its orientation to-

wards providing meaningful information from data,

as of today the linguistic descriptions of data ﬁeld

has to face several problems as a novel research do-

main. First, the sole use of quantiﬁed sentences is

Ramos-Soto A., Bugarín A. and Barro S..

Computing with Perceptions for the Linguistic Description of Complex Phenomena through the Analysis of Time Series Data.

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

usually not enough to address real life linguistic de-

scription problems. In this sense several approaches

have studied the extension of quantiﬁed sentences to

deal with time series data or to provide reasoned de-

scriptions (see next Section), but their full potential is

still unexplored. Secondly, linguistic descriptions do

not directly provide textual descriptions ready for hu-

man consumption, but rather a set of several linguis-

tic properties which can be assigned to a certain con-

cept or entity. Consequently, in order to use linguis-

tic descriptions in real contexts we do not only need

expressiveness, but also a way of translating this ex-

pressiveness into tangible solutions for actual human

users. Another issue related to the previous problem

is that linguistic descriptions approaches have been

proposed from both a formal and practical point of

view, but applications or systems that make use of

these techniques in real life environments are almost

nonexistent.

In this context, we are developing this Ph.D. in

order to research the use of linguistic descriptions ad-

dressing several of the issues that this ﬁeld currently

poses. In fact, we believe that by both providing

real solutions employing linguistic descriptions and

extending the current theoretical base to consider a

higher expressiveness, linguistic descriptions can be-

come a very useful tool as a more human-friendly al-

ternative to other widely used descriptive techniques,

such as statistics or data mining approaches.

2 OUTLINE OF OBJECTIVES

The objectives of this Ph.D. are both practical and the-

oretical. On one hand we intend to explore and ad-

dress linguistic description needs in real-life contexts.

On the other hand, based on the experience gathered

in the practical use cases, we intend to achieve a gen-

eral linguistic description model which can be used to

easily develop linguistic description solutions for any

application domain. More speciﬁcally, we will:

• Study each application domain and create a cor-

pus of commonly used linguistic expressions.

• Extend the computational theory of perceptions

(Zadeh, 2000) to include the representation needs

previously identiﬁed, so that new linguistic re-

sources can be contemplated, such as hypothesis,

causality, speculation, conclusions, as well as new

kind of quantiﬁers and operators at a syntactic-

semantic level, which allow to establish spatio-

temporal relationships among the occurrence of

events.

• Research quality assessment criteria for linguistic

descriptions. These allow to determine the quality

degree of candidate linguistic descriptions in an

objective way.

• Create search procedures which select the most

relevant descriptions according to the application

domain expert’s preferences. These procedures

may be heuristic (where, in an ad hoc way, the

expert knowledge is directly included as part of

the search procedure) or meta-heuristic (where an

optimization function including the expert knowl-

edge guides the search process), depending on the

problem characteristics. In the latter case, genetic

programming seems an appropriate strategy, since

it is oriented to learn grammar instances, which

can be used to structure and represent the target

linguistic descriptions.

• Research validation methodologies to verify the

robustness of the linguistic description generation

procedures and the subjective quality of the lin-

guistic descriptions. Although the objective crite-

ria used in the search procedures allow to obtain

good quality descriptions, these have a limited va-

lidity in this context, due to the high number of

possible answers and the subjectivity in their as-

sessment (situation complexity, divergent expert

criteria, etc.).

• Ultimately develop a generic linguistic descrip-

tion model which considers and formalizes all

the previous described aspects. This model will

be implemented as a software library which may

serve as the kernel engine of a linguistic descrip-

tion generator authoring tool.

3 STATE OF THE ART

The linguistic descriptions (or summaries) of data

(LDD) aim to obtain informative, brief and precise

descriptions from numeric datasets and cover a group

of soft computing-based concepts and techniques,

such as linguistic variables, fuzzy operators and quan-

tiﬁcation methods. It is a novel ﬁeld, whose so-

lutions provide information in the form of linguis-

tic terms. Speciﬁcally, although preliminar ideas ap-

peared early in the 1980s, it started to develop in the

second half of the 1990s, when the advances in the

ﬁeld of fuzzy sets (namely computing with words and

computing with perceptions) provided new potential

applications, mostly oriented to data mining. Due to

its short career and its formal background, many ap-

proaches in this ﬁeld are on the theoretical side, al-

though in some cases practical examples and real life

based problems have been addressed.

ICAART2015-DoctoralConsortium

3.1 Theoretical Work

(Yager, 1982) and (Yager et al., 1990) deﬁne the basic

linguistic summaries and serve as the starting point

in the characterization of linguistic descriptions, in-

troducing the concept of quantiﬁed sentences as data

summaries. Years later, (Zadeh, 1996), (Zadeh, 2000)

and (Zadeh, 2001) introduce the concept of com-

puting with words (CWW) and computing with per-

ceptions (CWP) and highlight the potential of the

fuzzy logic to provide a methodology to this concept,

including examples which show how this approach

could be structured. More recently, (Kacprzyk and

Zadrozny, 2005), (Kacprzyk and Zadrozny, 2010) and

(Kacprzyk, 2010) introduce some ideas about a poten-

tial relationship between computing with words and

the natural language ﬁeld, but do not explore them in

depth.

From the ideas and concepts proposed in the pre-

vious contributions, the construction of a linguistic

description framework which can be applied to any

kind of description problem in any domain is perhaps

one of the biggest challenges in this ﬁeld, but it is

still far from being achieved. In this sense, the Gran-

ular Linguistic Model of a Phenomenon (Menendez-

Gonzalez and Trivino, 2011), which has been used

as a solution for several practical cases in diverse

domains (Menendez-Gonzalez and Trivino, 2011),

(Alvarez-Alvarez and Trivino, 2013), (Eciolaza and

Trivino, 2011), is the nearest approach there is to an

all-in-one framework. It is based on a set of inter-

connected nodes named perception mappings (PM),

which receive a set of computational perceptions (CP)

as input. Each PM applies an aggregation function to

the input CP (for example minimum, maximum, aver-

age or even fuzzy rules) and generates a new CP as a

result which can be reused as input to other PM. Fig-

ure 1 shows an example of a GLMP model which de-

termines and describes types of climate from temper-

ature and humidity input data. A are linguistic label

partitions and W vector of fuzzy fulﬁllment degrees

associated to each label. R

...R

is a set of fuzzy rules.

1PM

temperature

1PM

humidity

2PM_climate

26 º

86 %

A = (cold,normal,hot)

W = (0,0.3,0.7)

A = (low,normal, high)

W = (0,0.2,0.8)

R1: If temperature is

hot and humidity is

high then climate is

tropical

....

Rn: If temperature is

hot and humidity is

low then climate is

desertic

Currently we have a

tropical weather (truth

degree: 0.7)

Figure 1: Example of a GLMP model which determines the

type of climate from temperature and humidity data inputs.

Other recent contributions explore the use of dif-

ferent quantiﬁers and develop evaluation criteria for

quantiﬁed sentences. For example, (Diaz-Hermida

and Bugarin, 2011) with several theoretical aspects

such as the use of semi-fuzzy quantiﬁers to model

quantiﬁed sentences and the description of some

generic methods for pattern detection. Furthermore,

ıaz-Hermida et al., 2011), (Castillo-Ortega et al.,

2012), (Wilbik and Keller, 2012) and (Menendez and

Trivino, 2012) deﬁne several and mostly coincident

evaluation criteria, such as the data coverage percent-

age or the sentence fuzzy fulﬁllment degree. Oth-

ers were inspired by the conversational maxims in

the ﬁeld of human communication (Gamut, 1991), in-

cluding the relevance or the length of the description.

In fact, when referring to criteria, it can be stated that

there is a solid consensus about which characteristics

of a linguistic description can be useful in the task of

evaluating and ranking candidate descriptions in an

objective way.

3.2 Use Cases and Practical

Contributions

In (Castillo-Ortega et al., 2011a), the concept of lin-

guistic summary applied to temporal data series is

given, which must fulﬁll the brevity, precision and

data coverage criteria. A few algorithms to obtain

linguistic summaries are presented. The given exam-

ple is made on data about the patient inﬂow in med-

ical centers, from which summaries such as “Most

of the days with cold weather” or “patient inﬂow is

low or very low” are obtained. This use case was

also explored in (Castillo-Ortega et al., 2011b) using

a genetic algorithm approach instead of the standard

heuristic algorithms used to generate the linguistic de-

scriptions.

In (Kobayashi and Okumura, 2009), a speciﬁc

application oriented to the economic domain is de-

scribed. Nikkei data time series are used, from which

several pattern proﬁles which take into account the

curvature and trend of the data series are detected to

produce summaries about the evolution of the market

in a given date. These descriptions were compared

with news reports about that evolution. The obtained

descriptions are composed of simple sentences such

as “At the end of the session the prices decreased”.

(Kacprzyk and Wilbik, 2009) orients the use of

linguistic descriptions to temporal series comparison,

with the objective of helping human decision taking

in an effective way, in this case related to economic

investments. The kind of sentences obtained include

variation patterns, such as “Among all y, most are con-

stant”, “Among all medium y, most are constant” or

“Among all moderate y, most are medium and con-

ComputingwithPerceptionsfortheLinguisticDescriptionofComplexPhenomenathroughtheAnalysisofTimeSeries

Data

stant”.

In (van der Heide and Trivino, 2009), the prob-

lem of generating linguistic descriptions for domestic

electric consumption is addressed. This work high-

lights the potential that linguistic descriptions have

for an electricity company in order to provide cus-

tomers with customized information above mere nu-

merical data. In this case, intuitive descriptions are

given, such as “About two thirds of the days the con-

sumption in the mornings is lower than the consump-

tion in the afternoons”, “Most of the days the con-

sumption in the mornings is lower than the consump-

tion in the evenings” or “About two thirds of the days

the consumption in the middays is lower than the con-

sumption in the evenings”.

Based on the GLMP model, (Eciolaza and Triv-

ino, 2011) automatically produces linguistic descrip-

tions of driving activity from vehicle simulator data.

(Alvarez-Alvarez and Trivino, 2013) also employs

GLMP together with fuzzy ﬁnite state machines, to

create a basic linguistic model of the human gait and

to generate a human friendly linguistic description of

this phenomenon focused on the assessment of the

gait quality, including rules which allow to provide

explanations to the descriptions as in “28 days after

the knee lesion, the gait quality is very low because

the gait symmetry is low and the gait homogeneity is

low”. (Menendez-Gonzalez and Trivino, 2011) uses

this framework to create linguistic descriptions from

OLAP cubes in the energy consumption domain, such

as “Your behavior is inefﬁcient, due to the high con-

sumption, the quite old devices and the low consump-

tion at the low charge period”.

4 METHODOLOGY

The methodology we are employing to develop this

Ph.D. is based on a cyclic hybrid bottom-up and top-

down approach. This approach consists in studying

practical application domains in order to gain knowl-

edge and experience on how to make the best use of

linguistic descriptions. From these practical cases, we

intend to identify and abstract (bottom-up) a generic

model which can gather the techniques employed in

these solutions, but which can be applied in other do-

mains as well (top-down).

The use of this model in other domains might not

provide enough potential to address the whole prob-

lem. In this case, the new linguistic description re-

quirements and concepts would be incorporated into

the model in order to extend its capabilities. Figure 2

shows a structural diagram of this methodology.

Since practical problems are the pillars of this

linguistic description

generic model

solutions for

speciﬁc problems

solutions for

speciﬁc problems

lead to

provides

extend

Figure 2: Schema of the Ph.D. development methodology.

methodology, we should also further extend this

methodology to deal on how to approach realistic lin-

guistic description needs. Thus, for each problem we

must:

1. Study a corpus, if available, of the descriptions

elaborated by the domain experts. This allows to

identify the structure of the target descriptions and

which sort of techniques must be used to create

them. If a set of examples is not available, then

the experts must deﬁne the target descriptions.

2. Design the linguistic description generation ap-

proach according to the requirements deﬁned in

the previous stage. Determine which operators

and techniques must be used in order to extract

the relevant information from the source data.

3. Develop prototypes, ideally including a basic nat-

ural language generation system which translates

the linguistic descriptions into texts which can be

reviewed by the experts.

4. Perform a full validation process involving the ex-

perts once the results seem satisfactory in order to

ensure that the solution can be used in a real con-

text.

These tasks are based on the natural language gener-

ation research ﬁeld, which has addressed many real

text generation necessities. Consequently, this way of

creating linguistic description solutions seems closer

to software engineering methodologies than to a re-

search process. However, the only way of proving

the usefulness of linguistic description techniques is

to apply them in practical cases, and this usually in-

volves the creation of software systems which must

deal with natural language generation, among several

other aspects.

ICAART2015-DoctoralConsortium

5 EXPECTED OUTCOME

We expect to address several linguistic description

problems in practical contexts, providing solutions to

real life description needs. These approaches will

help in the creation of the generic linguistic descrip-

tion model described in the previous section.

More speciﬁcally, the experience gathered in the

creation of practical linguistic description solutions

will allow us to provide a model which is based on

real contexts. Consequently, we would achieve a for-

mal model whose usefulness is proven by the practical

problems it is based on.

This model would provide solutions for new lin-

guistic description applications, which, in exchange,

could also provide new ideas and concepts for the ex-

tension of the original model.

6 STAGE OF THE RESEARCH

During the ﬁrst stage of the Ph.D. we explored the ap-

plicability and performance of linguistic descriptions

in meteorology, an applied domain which presents

several interesting and relevant use cases.

Our ﬁrst research approach addressed a common

task among meteorologists, which consists in the cre-

ation of weather reports about the climate behavior.

We studied monthly reports about temperatures in

Galicia (NW Spain) published by the Galician Mete-

orology Agency (MeteoGalicia) and elaborated a cor-

pus which served as a base to deﬁne the structure and

content of the target linguistic descriptions. As a re-

sult, we deﬁned the concept of fuzzy nuance and pro-

posed an heuristic procedure which obtained similar

reports to the ones issued by the experts (as in Fig. 3).

Our results showed that the automatically generated

reports were consistent with the original ones.

Shortly after, we researched the application of

quantiﬁed sentences to create descriptions about the

cloud coverage state forecasting. Our approach aggre-

gated geographical time series data from numeric pre-

diction models to provide a global forecast for Gali-

cia. These automatically generated forecasts were

compared to the cloud coverage forecasts issued by

the meteorologists, showing that, although meteorol-

ogists use subjective experience and several informa-

tion sources to elaborate forecasts, the approach we

developed can be applied to provide linguistic de-

scriptions of geographical forecasts. An explanation

of the two previous use cases can be found in (Ramos-

Soto et al., 2013a), (Ramos-Soto et al., 2012a) and

(Ramos-Soto et al., 2012b).

Due to our involvement in the meteorology ﬁeld

Meteorologist: “Temperatures were high for

October, due to the high temperatures which

were registered during the ﬁrst fortnight”.

LDD approach: “The temperature was high

for October, with very high temperatures

during the ﬁrst fortnight and very cold

temperatures during the fourth week”.

Figure 3: Example of a temperature report made by a mete-

orologist and an automatically generated report for the same

data source from (Ramos-Soto et al., 2013a).

and the experience gathered in our previous works we

created GALiWeather, an application which automat-

ically generates short-term forecast texts in Spanish

and Galician for every Galician municipality using a

linguistic description approach combined with a ba-

sic natural language generation method (see Fig. 4

for a textual forecast example). During this research

and development we have been supported by mete-

orologists from MeteoGalicia, who helped us in the

validation process of the application. A preliminary

study of the application can be found in (Ramos-Soto

et al., 2013b), while (Ramos-Soto et al., 2014a) and

(Ramos-Soto et al., 2014c) provide an exhaustive and

detailed technical explanation on the ﬁnal version of

GALiWeather. As of today, GALiWeather has been

producing daily textual forecasts in MeteoGalicia’s

test servers for more than half a year. A public re-

lease of the automatically generated texts is expected

soon.

There will be clear skies at the beginning and towards the middle of the

term, although at the end they will be very cloudy. We expect precipitations

on Thursday morning. The temperatures will be normal for the minimums

and high for the maximums for this period of the year, with minimums in

notable increase and maximums without changes.

9th December,

Monday

9th December,

Monday

9th December,

Monday

10th December,

Tues day

10th December,

Tues day

10th December,

Tues day

11th December,

Wednesday

11th December,

Wednesday

11th December,

Wednesday

12th December,

Thursday

12th December,

Thursday

12th December,

Thursday

Morn.

Aft.

Night

Morn.

Aft.

Night

Morn.

Aft.

Night

Morn.

Aft.

Night

Min: 1º Max: 14º

Min: 5º Max: 16º

Min: 7º Max: 16º

Min: 11º Max: 15º

Figure 4: Example of a meteorological short-term forecast

obtained by GALiWeather.

While GALiWeather is not as sophisticated as

other forecast generators from a natural language gen-

eration point of view (such as those described in

(Goldberg et al., 1994), (Coch, 1998) or (Sripada

et al., 2003)), it focuses on the use of linguistic de-

scriptions in a real context, which provide meaningful

information for a wide public.

Currently, we ﬁnd ourselves in the middle of the

Ph.D. development. We are considering new appli-

ComputingwithPerceptionsfortheLinguisticDescriptionofComplexPhenomenathroughtheAnalysisofTimeSeries

Data

cation domains in which linguistic descriptions may

prove useful, but are mainly focused on research-

ing a preliminary linguistic description generic model

based on our previous experience. We have also re-

cently explored the current state of the art in both nat-

ural language generation systems and linguistic de-

scriptions of data in order to ascertain the role that

generic LDD approaches (and thus our model) could

play integrated into NLG systems (Ramos-Soto et al.,

2014b).

We intend to provide a model which can be di-

rectly used in practical cases but can also be formally

described, thus maintaining both the theoretical and

practical aspects of the Ph.D. objectives. Our aim is

that this model can create linguistic descriptions from

heterogeneous data-sets, although at ﬁrst we will fo-

cus on time series data. We expect to further extend

this model to support data with spatial components

and also to include new concepts and capabilities as

new practical problems arise.

Furthermore, we will also explore the genera-

tion of linguistic descriptions using meta-heuristic ap-

proaches. This is a task which has been scarcely ex-

plored (Castillo-Ortega et al., 2011b) and which may

prove useful in the sense of providing a general algo-

rithm for creating linguistic descriptions. This could

also be one of the possible extensions to our model.

REFERENCES

Alvarez-Alvarez, A. and Trivino, G. (2013). Linguistic de-

scription of the human gait quality. Engineering Ap-

plications of Artiﬁcial Intelligence, 26(1):13 – 23.

Castillo-Ortega, R., Mar

ın, N., and S

anchez, D. (2011a).

A fuzzy approach to the linguistic summarization of

time series. Multiple-Valued Logic and Soft Comput-

ing, pages 157–182.

Castillo-Ortega, R., Mar

ın, N., S

anchez, D., and Tetta-

manzi, A. (2012). Quality assessment in linguistic

summaries of data. In Greco, S., Bouchon-Meunier,

B., Coletti, G., Fedrizzi, M., Matarazzo, B., and

Yager, R., editors, Advances in Computational Intel-

ligence, volume 298 of Communications in Computer

and Information Science, pages 285–294. Springer

Berlin Heidelberg.

Castillo-Ortega, R., Mar

ın, N., S

anchez, D., and Tetta-

manzi, A. G. (2011b). Linguistic summarization of

time series data using genetic algorithms. In Advances

in Intelligent Systems Research, volume 1 - 1, pages

416 – 423.

Coch, J. (1998). Multimeteo: multilingual production of

weather forecasts. ELRA Newsletter, 3(2).

Delgado, M., Ruiz, M. D., S

anchez, D., and Vila, M. A.

(2014). Fuzzy quantiﬁcation: a state of the art. Fuzzy

Sets and Systems, 242(0):1 – 30. Theme: Quantiﬁers

and Logic.

Diaz-Hermida, F. and Bugarin, A. (2011). Semi-fuzzy

quantiﬁers as a tool for building linguistic summaries

of data patterns. In IEEE Symposium on Foundations

of Computational Intelligence, pages 45–52.

ıaz-Hermida, F., Ramos-Soto, A., and Bugar

ın, A. (2011).

On the role of fuzzy quantiﬁed statements in linguistic

summarization. In Proceedings of 11th International

Conference on. Intelligent Systems Design and Appli-

cations (ISDA), pages 166–171.

Eciolaza, L. and Trivino, G. (2011). Linguistic reporting of

driver behavior: Summary and event description. In

Intelligent Systems Design and Applications (ISDA),

2011 11th International Conference on, pages 148–

153.

Gamut, L. T. F. (1991). Logic, Language and Meaning.

University of Chicago Press.

Goldberg, E., Driedger, N., and Kittredge, R. (1994). Using

natural-language processing to produce weather fore-

casts. IEEE Expert, 9(2):45–53.

Kacprzyk, J. (2010). Computing with words is an im-

plementable paradigm: Fuzzy queries, linguistic data

summaries, and natural-language generation. IEEE

Trans. Fuzzy Systems, pages 451–472.

Kacprzyk, J. and Wilbik, A. (2009). Using fuzzy linguis-

tic summaries for the comparison of time series: an

application to the analysis of investment fund quo-

tations. In Proceedings IFSA/EUSFLAT Conf. 2009,

pages 1321–1326.

Kacprzyk, J. and Zadrozny, S. (2005). Linguistic database

summaries and their protoforms: towards natural lan-

guage based knowledge discovery tools. Inf. Sci. Inf.

Comput. Sci., 173(4):281–304.

Kacprzyk, J. and Zadrozny, S. (2010). Linguistic data

summarization: A high scalability through the use

of natural language? Scalable Fuzzy Algorithms for

Data Management and Analysis: Methods and De-

sign, pages 214–237.

Kobayashi, I. and Okumura, N. (2009). Verbalizing time-

series data: With an example of stock price trends. In

Proceedings IFSA/EUSFLAT Conf., pages 234–239.

Menendez, C. and Trivino, G. (2012). Selection of the best

suitable sentences in linguistic descriptions of data. In

Greco, S., Bouchon-Meunier, B., Coletti, G., Fedrizzi,

M., Matarazzo, B., and Yager, R., editors, Advances in

Computational Intelligence, volume 298 of Communi-

cations in Computer and Information Science, pages

295–304. Springer Berlin Heidelberg.

Menendez-Gonzalez, C. and Trivino, G. (2011). Olap

navigation in the granular linguistic model of a phe-

nomenon. In Computational Intelligence and Data

Mining (CIDM), 2011 IEEE Symposium on, pages

260–267.

Ramos-Soto, A., Bugar

ın, A., and Barro, S. (2014a). Gen-

eraci

on autom

atica de predicciones a corto plazo:

Metodolog

ıa y validaci

on. In Actas XVII Congreso

Espa

nol sobre Tecnolog

ıas y L

ogica Fuzzy (ESTYLF),

pages 405–410.

Ramos-Soto, A., Bugarin, A., and Barro, S. (2014b). On the

role of linguistic descriptions of data in the building of

natural language generation systems. Submitted.

ICAART2015-DoctoralConsortium

Ramos-Soto, A., Bugarin, A., Barro, S., and Diaz-Hermida,

F. (2013a). Automatic linguistic descriptions of me-

teorological data a soft computing approach for con-

verting open data to open information. In Information

Systems and Technologies (CISTI), 2013 8th Iberian

Conference on, pages 1–6.

Ramos-Soto, A., Bugarin, A., Barro, S., and Taboada, J.

(2013b). Automatic generation of textual short-term

weather forecasts on real prediction data. In Larsen,

H., Martin-Bautista, M., Vila, M., Andreasen, T., and

Christiansen, H., editors, Flexible Query Answering

Systems, volume 8132 of Lecture Notes in Computer

Science, pages 269–280. Springer Berlin Heidelberg.

Ramos-Soto, A., Bugarin, A., Barro, S., and Taboada, J.

(2014c). Linguistic descriptions for automatic gener-

ation of textual short-term weather forecasts on real

prediction data. Fuzzy Systems, IEEE Transactions

on, Early Access.

Ramos-Soto, A., D

ıaz-Hermida, F., Barro, S., and Bugar

ın,

A. (2012a). Validation of a linguistic summarization

approach for time series meteorological data. In 5th

ERCIM International Conference, page 133.

Ramos-Soto, A., D

ıaz-Hermida, F., and Bugar

ın, A.

(2012b). Construcci

on de res

umenes ling

ısticos in-

formativos sobre series de datos meteorol

ogicos: in-

formes clim

aticos de temperatura. In Actas XVI Con-

greso Espa

nol sobre Tecnolog

ıas y L

ogica Fuzzy (ES-

TYLF), pages 644–649.

Reiter, E. and Dale, R. (2000). Building Natural Language

Generation Systems. Cambridge University Press.

Sripada, S., Reiter, E., and Davy, I. (2003). Sumtime-

mousam: Conﬁgurable marine weather forecast gen-

erator. Expert Update, 6(3):4–10.

van der Heide, A. and Trivino, G. (2009). Automatic gener-

ated linguistic summaries of energy consumption data.

In Proceedings of 9th ISDA Conference, pages 553–

559.

Wilbik, A. and Keller, J. M. (2012). A distance metric for

a space of linguistic summaries. Fuzzy Sets and Sys-

tems, 208:79 – 94.

Yager, R. R. (1982). A new approach to the summarization

of data. Information Sciences, 28(1):69 – 86.

Yager, R. R., Ford, K. M., and Ca

nas, A. J. (1990). An

approach to the linguistic summarization of data. In

Bouchon-Meunier, B., Yager, R. R., and Zadeh, L. A.,

editors, Uncertainty in Knowledge Bases, 3rd Inter-

national Conference on Information Processing and

Management of Uncertainty in Knowledge-Based Sys-

tems, IPMU 90, Paris, France, July 2-6, 1990, Pro-

ceedings, volume 521 of Lecture Notes in Computer

Science, pages 456–468. Springer.

Zadeh, L. A. (1996). Fuzzy logic = computing with words.

Fuzzy Systems, IEEE Transactions on, 4(2):103–111.

Zadeh, L. A. (2000). From computing with numbers to

computing with words : From manipulation of mea-

surements to manipulation of perceptions. In Intelli-

gent Systems and Soft Computing: Prospects, Tools

and Applications, pages 3–40. Springer-Verlag.

Zadeh, L. A. (2001). A new direction in ai — toward a com-

putational theory of perceptions. In Reusch, B., editor,

Computational Intelligence. Theory and Applications,

volume 2206 of Lecture Notes in Computer Science,

pages 628–628. Springer Berlin Heidelberg.

ComputingwithPerceptionsfortheLinguisticDescriptionofComplexPhenomenathroughtheAnalysisofTimeSeries

Data