BMS Modal Auxiliaries in Text Analysis

7/29/2019 BMS Modal Auxiliaries in Text Analysis

http://slidepdf.com/reader/full/bms-modal-auxiliaries-in-text-analysis 1/19

http://bms.sagepub.com/

Sociologique

Bulletin de Méthodologie

http://bms.sagepub.com/content/70/1/5The online version of this article can be found at:

DOI: 10.1177/075910630107000103

2001 70: 5Bulletin de Méthodologie Sociologique

Roel PoppingModal Auxiliaries in Text Analysis

Published by:

http://www.sagepublications.com

On behalf of:

Association Internationale de Methodologie Sociologique

found at:can beBulletin de Méthodologie Sociologique Additional services and information for

http://bms.sagepub.com/cgi/alertsEmail Alerts:

http://bms.sagepub.com/subscriptionsSubscriptions:

http://www.sagepub.com/journalsReprints.navReprints:

http://www.sagepub.com/journalsPermissions.navPermissions:

http://bms.sagepub.com/content/70/1/5.refs.htmlCitations:

by Marcelo González on September 17, 2010bms.sagepub.comDownloaded from



http://bms.sagepub.com/content/70/1/5



http://www.sagepublications.com/

http://bms.sagepub.com/cgi/alerts



http://bms.sagepub.com/subscriptions


http://www.sagepub.com/journalsReprints.nav


http://www.sagepub.com/journalsPermissions.nav

http://bms.sagepub.com/content/70/1/5.refs.html








http://www.sagepub.com/journalsPermissions.nav




http://www.sagepublications.com/





5

MODAL AUXILIARIES IN TEXT ANALYSIS*

by

Roel Popping(Department of Sociology, Groningen University,

NL 9712 TG 15; [email protected])

Résumé. Les auxiliaires modaux dans l’analyse textuelle. Les verbes auxiliaires modaux tels

que devoir, vouloir... fournissent des informations concernant les intentions des sujets sémantiquesdans les morceaux de phrase où ils apparaissent. Par exemple, en déclarant qu’une personne doit

agir d’une certaine manière, on montre la différence entre l’action et la possibilité que la personne

agisse autrement. Cette information peut être utilisée dans la recherche. L’analyse textuellesémantique permet le codage des auxiliaires modaux. Cet article examine comment faire pour

garder en présence les auxiliaires modaux lors une analyse textuelle par réseaux. Ce type d’analysetextuelle nous permet de traiter des argumentations assez complexes, mais le résultat montre quele chercheur ne doit pas utiliser l’analyse par réseaux dans ces cas et plutôt employer l’analysesémantique. Analyse textuelle sémantique, Analyse textuelle par réseaux, Verbes auxiliaires

modaux.

Abstract. Modal auxiliary verbs (e.g., ought, want, etc.) convey information about the intentions

of the semantic subjects within the clauses in the text in which they appear. For example, in

asserting that a person ought to act in a certain way, one contrasts the action with the person’s

potential intentionto act

otherwise.This

qualitycan be used

in research. Semantictext

analysisallows coding modal auxiliaries. This paper investigates how to make modal auxiliaries remain

visible when network text analysis is used. This type of text analysis allows one to deal with rather

complex argumentation. The answer shows that in this situation network text analysis should not

be used, the investigator should stay with semantic text analysis. Semantic Text Analysis,Network Text Analysis, Modal Auxiliary Verbs.

INTRODUCTION,

~

- &dquo;

Classical (or thematic) text analysis only counts the occurrence or

co-occurrence of words, phrases or themes as they are found in

blocks of text. More recent approaches like semantic and networktext analysis concentrate on the relation between themes. This givesadditional information. First of all, it is certain that a relation exists

between the themes. In the thematic approach, this was onlyassumed, since themes occurred in the same block of text. Secondly,








6

the relations are named. The structure of such a relation is

Subject(Verb(Object (S(V(O).Often the relation is also

weightedor

valenced (&dquo;He loves her&dquo; is stronger than &dquo;He likes her&dquo;, and &dquo;Hehates her&dquo; is the opposite). Now the relation becomes

Subject(Valence(Verb(Object (S(V(V(O).

Modal auxiliaries express a specific kind of relation. They denotesome intentionality. They may be part of some political or ideologicaldiscourse. Modal auxiliary verbs cannot be simply valenced, one can

not say in advance that &dquo;must&dquo; is always stronger than &dquo;should&dquo;.This valencing and the use of certain modal auxiliaries depend on

the context in which they are used; for example, in a culture or a

political system. Modal auxiliaries can be recognised when semantictext analysis is performed. The question is whether they can also be

captured in network text analysis. This type of analysis is attractive

as it concentrates on complete argumentation instead of individualsentences or clauses.

As we are investigating popular developments in the transition fromformer Communist-lead Central-Eastern European countries to

more democratic states (Popping and Roberts, 1998), we will

concentrate on the political system. In this context, not all modalauxiliaries are relevant.

REPRESENTATION OF TEXTS

’

.,

~..

In text analysis today, three methods are used to represent the texts

that are analysed. For a long time, researchers have been consistent

intheir

portrayalof

text analysisas

involving the quantification ofqualitative data (e.g., words, gestures, art forms, etc.) for the

purpose of permitting statistical inference (Krippendorff, 1980;Popping, 2000). There are only a few differences between theseresearchers with respect to the (especially manifest versus latent)nature of their subject matter. Computer programs following the

approach of these researchers provide users with &dquo;counts&dquo; of wordsor phrases within blocks of text (e.g., newspaper articles,transcribed speeches, etc.). Such word counts are then aggregatedaccording to a dictionary of theoretically expedient meaning

categories (or themes), theme occurrences can be used to makestrong inferences regarding thematic differences across varioussocial and temporal contexts (e.g., Weber, 1990; Popping, 2000).Hereafter, I will use the term &dquo;concept&dquo; to denote a single idea

represented by a single word or a phrase. It is the basic unit for the

meaning content of a piece of text. The term &dquo;theme&dquo; is usually usedfor broader classes ofconcepts.








7

Since the late 1980s, a number of social scientists developed their

own methods not only for identifying concepts, but also for encodingthe relations among concepts in texts that thematic text analysisignores (Carley, 1986 & 1988; Van Cuilenburg, Kleinnijenhuis and

De Ridder, 1986 & 1988; Franzosi, 1989; Roberts, 1989). Theseresearchers’ methods for encoding texts are strikingly similar. In

each case, S(V(V(O meaning relations among words are encoded as

they appear in clauses of the texts under analysis. A clause is a partof a sentence with its own inflected verb and associated subject and

object. A sentence might consist of several clauses. Instead of

sentence, the more general term &dquo;discourse&dquo; is also used. A

discourse isa collection of

statementsin a text in

whichrelations

are made between subject and object.

Inflected verbs can be recognised as all words that change form

when the person and/or tense of the clause are changed. Anexample: The word &dquo;go&dquo; in &dquo;I go&dquo;, and the word &dquo;goes&dquo; in &dquo;He goes&dquo;are inflected verbs because they change form when the subjectchanges from first to third person.

Unlike thematic text analysis, such &dquo;clause-based text analysis&dquo;affords inference about how texts’ sources use words in their speechor writing. Where the social scientists’ methods differ is in the

research purposes to which the relationally encoded texts are

applied. &dquo;-..’ ; : - ..’~ ~ ... , ... -... , . _ . - ,..

’’ &dquo;

.~...:&dquo;~ &dquo;

..... , -...

SEMANTIC APPROACH .. , , I - &dquo;-

The first of the newer

approachesis semantic text

analysis.This

method yields data and inference about the intended (as opposed to

the manifest but more superficial grammatical) relations among

words in various socio-temporal contexts (Roberts, 1997a & 1997b).Whereas thematic text analysis concentrates on word or phrasecounts, this type of analysis looks at theme relations. Each clause is

coded according to the meaning that it was intended to convey;

namely, as a description or as a judgement of a process or of a state-

of-affairs. Each of the resultant types of intention (description of

process, description of state-of-affairs, judgement of process,

judgement of state-of-affairs) has associated with it a distinct, butunambiguous semantic grammar. Once a clause is encoded

according to one of these four semantic grammars, the available

computer program reconstructs that clause by &dquo;translating&dquo; it

according both to the meaning categories into which its words fall,and its words’ intended interrelations. For example,

..

__, ... , ._ ,. , .,...,.

( I’ ’ ’ ’,

I’

°. J,°_ , ’








8

SOMEBODY is obligated to SUMMARIZE a badness/harm (that THEPAST_GOVERNMENT FACILITATED) and SOMEBODY is obligated to

SUMMARIZE a LESSON about a badness/harm.

is a translation of the following sentence, also to be used later in

this text, from an article in the newspaper, N6pszabadsag, of April14, 1992:

Somebody should summarise for each new government the mistakes

made by the previous government, and the lessons of these

mistakes.

Such translations allow the coder to evaluate the face validity of the

encoding. Semantic text analysis yields inferences abouttoccurrences of specific statements among socio-temporal contexts of

theoretical interest (e.g., democratic imagery in Eastern Europeaneditorials prior vs. following the recent collapse of their Communist

governments). The objective is not to capture logical arguments or

cognitive maps, but to estimate the probability that specific classes

of statements occur. The technique is useful for comparingstrategies of communication in different socio-historical settings,and for measuring shifts in public opinion (or public perceptions)when texts have grass roots sources.

The above illustration shows that modal auxiliaries are captured inthe coding process. The computer program, PLCA, codes them as

separate variables (Roberts, 1989). The clause is the record unit.

This example contains two of such units. The first one contains themodal auxiliary verb &dquo;ought&dquo; (or to be obligated to). This computerprogram also allows the user to encode, among other things:

the syntactic form of the clause,.

the semantic subject of the verb,the modifier of the subject,the verb,

-

’

.

the tense, .

’

the semantic object of the verb,. &dquo;..

the modifier of the object, and

the valence.’

’

’

Each clause’s syntactic form reflects the author’s intended meaningas a description or an evaluation in reference to a process or a state

of affairs. The texts to be analysed are obtained by applying a

sampling procedure.~ ~

~_~ ~

~ ~

When a clause is used, its sender and intended audience are often

known. In written editorials, the journalist is usually the sender,and the people reading the newspaper or listening to the radioconstitute the audience. This sender-audience pair remains








9

constant within any particular editorial. In situations where, for

example,the

primeminister of one

countrytells the

presidentof

another country what to do, it is important to know both sender and

audience. In computer programs for semantic and network text

analysis, the sender and audience can be coded.

NETWORK APPROACH ..- .

The Network analysis of Evaluative Texts (NET) approach provides

data on latent propositions that can be logically derived from texts’manifest content (Van Cuilenburg, Kleinnijenhuis and De Ridder,1986 & 1988). The approach permits inference about how such tacit

propositions are related to the social contexts within which texts are

authored. Map Extraction, Comparison, and Analysis (MECA)provide data and inference regarding similarities and differences in

the ways that groups of individuals relate (i.e., cognitively maps) to

various aspects of their worlds (Carley, 1986 & 1988). In these

analyses, the network positions are relevant. The NET approachlends itself more readily than MECA to analysing the modal

auxiliaries within texts.

The network evaluation approach has its roots in the work on

evaluative assertion analysis by Osgood et al. (1956). The

fundamental premise of this methodology is that every language hasthree kinds of words:- &dquo;Common meaning terms&dquo; These are words that have a

common evaluative meaning among &dquo;reasonable-sophisticated users

of the language&dquo;. For example, the common meaning of words such

as &dquo;peace&dquo; are always positive; whereas that of words like &dquo;enemy&dquo;are always negative in connotation.- &dquo;Attitude objects&dquo; These have no fixed evaluative meaning. For

example, a word like &dquo;car&dquo; is likely to be evaluated differently bydifferent people.- &dquo;Verbal connectors&dquo; These are words that indicate the

association (&dquo;it is...&dquo;) or dissociation (&dquo;it is not...&dquo;) of attitude objectswith common meaning terms or with other attitude objects.

By investigating how attitude objects are associated or dissociated,one can investigate how these attitude objects are valued in a text.

In brief, the method requires that texts be parsed into clauses, that

instances of the three above-mentioned word-types be located withinthese clauses, and that the clauses be recombined in a way that

reveals structure in the text. The NET text analysis method goesbeyond this approach to reveal the logically implied structure of

texts as well.








10

Two important steps are required when one encodes networks from

texts. First,one

must specify the concepts whichare to

be relatedor

linked within networks. These concepts may originate in one’s

theory or in the texts themselves (i.e., with concepts that are

empirically recurrent). Encoding then involves the classification oftexts’ words/phrases as occurrences of these concepts. Yet this

encoding process is likely to generate invalid data if contextual

information is not taken into consideration. Thus, it is crucial that

native speakers determine when specific words/phrases are

instances of one concept, and, given their idiomatic usage, when

they are instances of another.

The second step in network encoding involves the assignment of

links between pairs of concepts. After removing idiomatic ambiguityduring the classification of concept occurrences, most interrelations

among words/phrases can be unambiguously identified according to

the grammar of the language in which the texts were written. Yet

even identical non-idiomatic, but grammatically-correct statements

can have different meanings, depending upon the meaning intended

by the statement’s source. For example, the statement &dquo;Joe was

abandoned&dquo; could mean either that Joe was

aloneor that others

departed. All relational text analysis methods must ensure that such

illocutionary ambiguity is systematically disambiguated (Roberts &

Popping, 1996).

Within any single network, no concept appears more than once. As a

direct consequence, graphical displays of networks become complexwhen specific pairs of concepts are allowed multiple types of links.

For this reason, network analysis software developers have tendedto gravitate toward six general types of links (Popping & Roberts,

1997):*

Similarity - records that one concept is identical with another. Therelation is symmetric (abbreviations: ALIke, SIMilarity, EQUal,EQuiValent);*

Causality - denotes a cause-effect relation. The relation is

asymmetric and transitive. In all methods using networks based on

text, the causal relation is read as &dquo;might&dquo; cause (CAUses, is Caused

BY);* Relation - indicates an ASSociation, an ORDering, an EVAluation,or a REAlization. In the first case the relation is

symmetric,in the

other three it is asymmetric;* Classification - indicates transitive (is A Kind Of, Has As Kind),asymmetric (Is Property Oi), or symmetric (INConsistent with or

contradicts, DIStinct) classification;* Structure - indicates a structuring. The relation is transitive (isPARt of, Has As Part);* Affective - establishes a judgement of the subject about the object(AFFective, WILL). ,

,

.








11

Several software

applicationsalso allow

recordingthe

degreeof

similarity, causal determination, etc. of links, usually on a ( to (11scale. For example, &dquo;He hates her&dquo; would be valenced with (1, the

&dquo;He loves her&dquo; with + 1. The &dquo;He likes her&dquo; is more than a neutral

statement, so it might be valenced with +0.5. In the positivedirection, there are usually only two valences: +0.5 and + 1. All three

sentences are of the affective type. Evaluative links (&dquo;this is good&dquo;)can often be encoded on a minus-neutral-positive scale. Moreover,statements may themselves be conditionally encoded. For example,they may be encoded as &dquo;if statement 1 then statement 2&dquo; or as

&dquo;statement 1 or statement 2 is true&dquo;. The sentence: &dquo;If theCommunists do not win the elections, democracy has a chance&dquo; is

encoded as:

Clause: democracy / [condition] has (+) / chance ....

Condition: Communists / do not win (+) / elections.

The NET approach also uses two special-purpose concepts. These

concepts are related to the distinction between &dquo;attitude objects&dquo;(words with no fixed evaluative meaning) and &dquo;common meaningterms&dquo; (words that have a common evaluative meaning among

reasonably-sophisticated users of the language). Usually, a

statement can also be encoded as a positive (is good) or negative (isbad) evaluation of a concept by relating it to the abstract concept,&dquo;Ideal&dquo;. For example, the statement &dquo;the man is friendly&dquo; is

reformulated into &dquo;the man has a good relationship with the Ideal (ofthe statement’s source)&dquo;. By connecting a concept to the concept&dquo;Real&dquo;, a researcher can encode a statement as an affirmation that a

concept’s referent exists (is) or does not exist (is not). The statement

&dquo;unrest is rampant&dquo; is abbreviated as &dquo;Reality shows a high level ofunrest&dquo;. This implies that a concept can also be encoded in an

abstract manner.

A network represents an argumentation presented, for example, in

an editorial in a newspaper. Such a network can be reduced to a

simpler one, as is shown in Figure 1. Say the network contains three

concepts, namely A, B, and C. A positive valued relation exists from A to C, a negative valued relation from A to B, and two negativevalued relations from B to C. The four arrows are shown in the left-

most arrow network. In the link network, the two arrows from B to C

are reduced to one. In the chain network, chain A-B-C is reduced to

A-C. Finally, the two relations between A and C are combined in thebundle network. For details, the reader is referred to De Ridder

(1994) or Popping (2000). Carr6 (1979) provides the mathematics on

which these aggregations are based. In network text analysis, a

complete reasoning (for example, as presented in a paragraph in an








12

editorial) might be investigated. This is different from semantic text

analysiswhere

onlytendencies are

investigated.

Figure 1 - Network aggregations . ,.

MODAL AUXILIARY VERBS &dquo; .

Generally speaking, modal auxiliary verbs are verbs that are usedwith (usually, the infinitive form of) another verb to express

possibility, necessity, probability, certainty, etc. They are not used to

talk about things that definitely exist or events that definitelyhappened. The meanings of the verbs are grouped together in

several ways. In a sense, two main groups are distinguished:degrees of certainty (certainty, probability, possibility, choice) and

obligation / freedom to act (ability, necessity, permission, politerequest). This list is not exhaustive.

In text analysis focusing on democratic imagery, a firm restriction is

posed on these auxiliaries. First, they have to express intentionality

(e.g., they may describe how the future society should look), and notactuality (as in &dquo;Tomorrow the parliament will vote for...&dquo;). This

intentionality is confined by ego’s needs or alter’s morality, or

intentionality enabled by ego’s potentiality or alter’s obligations. An

example is found in the statement: &dquo;We must get more freedom to

express our opinions&dquo;. Second, the auxiliaries are used to express

implausibility or plausibility that ego’s intentions will be realised. An

example of plausibility is: &dquo;I can do it&dquo;. Therefore, modal auxiliaryverbs must indicate both (im-) plausibility and intentionality. For

example, the sentence &dquo;The washing machine can clean clothes&dquo;

indicates plausibility but not intentionality, whereas &dquo;He can cleanclothes&dquo; indicates both (as long as the sentence is not intended to

deanthropomorphise &dquo;he&dquo; as machine-like). Although, from a

grammatical standpoint, the verb &dquo;to be able&dquo; is a modal auxiliaryverb, for the purposes of encoding texts, verbs that merely indicate

degrees of certainty, probability, possibility, impossibility, etc., are

not the type of modal-intentionality indicators that are sought here.

Illustrations of sought-after modals are listed below:








13

-

Ego’sneeds Must (not in case of non-persons; e.g., &dquo;leaves must

fall&dquo;), refuses, would (not in case of mere possible futurity), etc. &dquo;The

politician would lose the vote if she failed to lobby for her case.&dquo;

- Alter’s morality May (in the sense of &dquo;permission&dquo; not

&dquo;possibility&dquo;), attempts, might (not in case of mere possible futurity),etc. &dquo;They might change the plan.&dquo;

- Ego’s potentiality Can (not in case of non-persons; e.g., washingmachines), wants, could (not in case of mere possible futurity), etc.

&dquo;I can do it.&dquo;

- Alter’s obligation Ought, hopes, should (not in case of mere

possible futurity), etc. &dquo;Next year I expect that the administration

should reduce taxes.&dquo;

Such modal auxiliary verbs can only be coded validly when the

context in which they are used is taken into account. This impliesthat the re p resentational view of coding is to be followed. Here,coding is performed by a human coder who codes a message from

the perspective of its sender. This is distinct from instrumental

coding, where &dquo;automatic&dquo; coding is performed from the perspectiveof the investigator (Shapiro, 1997). Another characteristic of the

representational way of coding is that it can overcome problems due

to ambiguity in language.

Many sentences or phrases do not explicitly contain a modal

auxiliary verb, but implicitly they do. In such cases, a specifictransformation to a modal form must be applied. Some examples

follow:

X &dquo;only makes sense if’ Y. --> IfY, then X ought to happen.X &dquo;is afraid that&dquo; Y. --> X does not want Y.

’What&dquo; X &dquo;needs is&dquo; Y. --> X ought to have Y.

X’s &dquo;plans to do&dquo; Y &dquo;are unrealistic&dquo;. --> X cannot do Y.

In the coding process, such transformations are made almost

automatically, otherwise the coding itself almost becomes

impossible. The hypothesis is that the type of modal auxiliary verb

used by a country’s citizens is related to the country’s type of

political system. In particular, modal auxiliary use differs amongauthoritarian, capitalistic, or social democratic states. This

hypothesis will be examined elsewhere.’

. &dquo;

’

,

,,,.








14

MODAL AUXILIARY VERBS IN NETWORKS

As indicated before, modal auxiliary verbs must indicate both (im-)plausibility and intentionality. This implies that, in terms of theNET approach, they refer to some ideal situation. In the network,this becomes visible in the relation between a specific concept and

the concept, &dquo;Ideal&dquo;. In case a sender or a receiver is known,additional information is added. In this case, one knows that these

specific parties share a specific Ideal. The valence will usually be

extreme, either -1 or + 1. For example: &dquo;The leader of the party stated

thatwe

should get more freedom&dquo; might be interpreted as &dquo;...weought to get more freedom&dquo; or as &dquo;...we shall likely get more freedomin the future&dquo;. Only the first interpretation involves the use of a

modal because it indicates both plausibility and intentionality. With

respect to coding for a network analysis, the sentence is read as a

universal statement spoken by the leader of the party: &dquo;Freedom is

an ideal.&dquo; As an S-V-V-O statement, it would be represented as

follows:

(Source) leader party: &dquo;(Subject) freedom / (verb) is (relation: EVA +1)/ (object) an ideal.&dquo;

Most sentences do not contain such universal statements. For

example, &dquo;The administration should reduce taxes&dquo; is rendered as

&dquo;The administration, if the administration reduces taxes, is actingwell&dquo; or as the following S-V-V-O statement:

&dquo;Administration / if (administration / reduces / taxes) (ACT -1) /Ideal.&dquo;

&dquo;

Note that the relation between administration and taxes isconsidered as a negative one. Say the chance that the condition willbe fulfilled is estimated as 40%; this is finally coded in CETA as: ,

&dquo;Administration (0.4) / (EVA + 1 ) / Ideal.&dquo; /.

In general, the modal auxiliary verbs are found in the following typesof links: similarity, cause, and association. But with the project on

imagined democracy, restrictions are imposed. These require that

similarity and cause be not used. The modal auxiliary verbs, as

described, are found in statements in which one of the concepts isthe Ideal. The type of relation here is an evaluation, which isconsidered as a special kind of association. This relation might bevalenced either positive or negative. In the case where the sentenceis read as &dquo;The administration reduces taxes&dquo;, different results willbe found:








15

&dquo;Administration / reduces (ACT -1 ) / Taxes.&dquo; .:

Here, the condition is lost completely. Therefore, this alternative is

not allowed. In network analysis not based on the NET approach,sentences are read as in this example.

Links ’

’

&horbar; ’- ~’ &dquo; ;&dquo;.i . . ~_: ~ ’ . ’ > .. &dquo;

- ~ ~ ...

..:.> > ...... , ... : :.’: ., ... ,.. :

When aggregation starts, one of the relation types involved is the

evaluation. In the aggregation process, the evaluation is consideredas an association. In case the relations are linked, the resulting typeis the type of the other relation (i.e., association and other typealways results in other type) (De Ridder, 1994: 98; Popping, 2000:

112). The following example is a paragraph from an editorial

entitled, &dquo;Brand-Old Lessons&dquo;, in the Hungarian newspaper,Nepszabadsag, of April 14, 1992:

&dquo;Somebody should summarise for each new government the

mistakes madeby

theprevious government,

and the lessons of these

mistakes; during a change of political systems [tr.: somebody should

summarise] the previous system’s offences in style or manner (and I

will not even mention crimes), including the one-sided personnelpolicies when the previous system was started, the foolish priority of

politics, and the negative consequences of this [tr.: politics-dominated society] that lasted for long years, decades.&dquo;

&dquo;

The relation:&dquo;

.

,. .~. :.. ’.~) _ :~~:~::j

&dquo;somebody / if (....) (ACT -1) / Ideal&dquo;-

’ &dquo; ’ =’-,&dquo; .’ .’.&dquo;&dquo;;

is found at least three times. When these relations are linked,nothing changes. The linked relation is:

&dquo;somebody (0.4) / ASS -1 / Ideal.&dquo;, - ;.’ ,’,’ &dquo;..

&dquo;

~ - -~ - &dquo;’

assuming a chance of 40% that these things will be summarised. In

linking, EVA + EVA, or more general ASS + ASS, is used whichresults in ASS. Note that here

only specificforms of association

relations have been joined so other types are in fact excluded. If

sentences are also coded that do not contain modal auxiliary verbsas treated here, it is not excluded that other types of relations will befound between &dquo;somebody&dquo; and &dquo;Ideal&dquo;. In the aggregated network, it

is now possible that the final relation between the two concepts is of

another type than association. However, no concrete examples were

found in the sampled editorials.








16

Linking, and also chaining and bundling, has far more

consequences.It influences the valence. It is also useful to

saysomething about the quality of the resulting relations. For this typeof information, the reader is referred to De Ridder (1994) and

Popping (2000). The investigator has to decide here how modal

auxiliary verbs related to the original relations will be aggregated.This is possible if the number of modal verbs is large. Otherwise,information about the modal verb itself will usually be lost. CETA

provides no convention for deciding what the result of &dquo;could&dquo; plus&dquo;should&dquo; will be.

Chains

~~ ~

..

With regard to chains, the resulting type is always an association

(i.e., association with other type results in association) (De Ridder,1994: 97; Popping, 2000: 112). Above, we had a statement in which

freedom is evaluated as an Ideal. The &dquo;saying what you want to say&dquo;might be coded as a part of freedom. Now we have two statements:

freedom / EVA + 1 / Idealsay what you want / PAR + 1 / freedom.

The PAR (is part of) relation is a special case of the similarityrelation. The resulting chain is:

.

say what you want / ASS + 1 / Ideal.

This comes from PAR x EVA, or SIM x ASS, which results in ASS.

The concept under investigation, however, is nearly always found inthe subject part of the statement. The object part contains the Ideal,just as in the situation of linking. In the editorials I investigated, thelatter situation always occurred. The fact remains that the objecttype is the Ideal. Based on the same argumentation as before, themodal auxiliary verbs need to be counted, based on the originalnetworks.

A problem that is relevant with regard to chaining is that of

transitivity. The reasoning that is followed should be correct. De

Ridder (1994: 64) presents the following example:

Boredom leads to vandalism.~ .. , .

.

Vandalism is punishable.

Multiplying the sentences would yield: &dquo;Boredom is punishable.&dquo; Theissue is not whether boredom is punishable, but whether the

negative common meaning &dquo;punishable&dquo; is transmitted by an








17

argument to &dquo;boredom&dquo;. I will not go further into this discussion

here.

Bundles .

Bundles result from linking and chaining. Due to chaining, the final

relations are in general of type association. In the bundles, the final

object type is again the Ideal. The modal auxiliary verbs occur in

combination with this type. It was stressed previously that the

counting of the modal auxiliary verbs should be based on theoriginal networks. These verbs are usually related to the object,Ideal. Therefore, it is possible to relate the counts to this concept.

The main question was whether network text analysis might be

preferred rather than semantic text analysis in the analysis of

democratic imagery. In semantic analysis, one codes the semantic

subject and object of the verb, plus the verb and its modal auxiliary.In network analysis, the occurrence of the various modal auxiliaryverbs is related to a concept. The (aggregated) network contains

other information than the clause as coded in semantic analysis.Modal auxiliary verbs are more explicitly depicted in the coded

clauses than in the coded networks. Therefore, when the modal

auxiliary is a dependent variable, it seems that the data should be

collected by applying semantic text analysis and not by applyingnetwork text analysis.

EXAMPLE .

,!

The above paragraph from the editorial, &dquo;Brand-Old Lessons&dquo;, in

Népszabadsag, of April 14, 1992, which is a very complex one, is

analysed by applying network text analysis (CETA) and semantic

text analysis (PLCA). The differences in results are discussed below.

We start with the network analysis by coding as follows: &dquo;Somebodyshould summarise for each new government the mistakes made bythe previous government, and the lessons of these mistakes.&dquo;

Somebody / (EVA; + 1.0 ~) / Ideal’

’

Somebody / (EVA; + 1.0 -) / Ideal .

Reality (= Previous_Government) / made (REA -1.0) /Mistakes

...,;..

Note the sentence consists of two parts, one about the mistakes and

one about the lessons. The relations as presented here only hold in








18

case the condition is fulfilled. Once this means that the mistakes are

summarised, andonce that the

lessonsare

summarised. Conditionsin CETA are indicated by the tilde (-).

&dquo;During a change of political systems somebody should summarise

the previous system’s offences in style or manner (and I will not even

mention crimes), including the one-sided personnel policies whenthe previous system was started, the foolish priority of politics, and

the negative consequences of this politics dominated society that

lasted for long years, decades.&dquo;

All non-essential prepositional and other phrases are dropped.These include, &dquo;During a change of political systems&dquo;, &dquo;in style or

manner&dquo;, and &dquo;for long years, decades&dquo;. The investigator also has to

locate idiomatic phrases and render them according to their

intended meaning. For example, &dquo;and I will not even mention crimes&dquo;is not literally what it means. It probably is an ironic statement thatmeans precisely its opposite; namely, &dquo;I am mentioning crimes.&dquo; The

sentence, &dquo;consequences of this politics dominated society that

lasted&dquo;, is not clear. It could mean &dquo;lasting consequences of prioritiesfor a

politics-dominated-society&dquo;, &dquo;consequencesof

priorities fora

lasting politics-dominated-society&dquo;, or &dquo;consequences of this politicsdominated a lasting society&dquo;. For the coding process, this will beread as &dquo;consequences of these priorities that lasted&dquo;. This leaves us

with the following coding:

Somebody / (EVA; + 1.0 -)/ Ideal

Somebody / (EVA; + 1.0 -)/ Ideal,

Somebody / (EVA; + 1.0 ~) / IdealSomebody / (EVA; + 1.0 -)/ Ideal ..

Reality (= I) / mention (REA + 1.0) / Crimes

This sentence is coded as consisting of five parts, four of which are

conditional. The conditions refer to the system offences, the

personnel policy, the priority of politics, and the negativeconsequences.

In total, there are seven relations now. Assuming each time a

conditional chance of 1, the condition will take place, the number of

conditional relations is reduced to one afterlinking.

This linked

relation is of the type ASS (association). From these results, we

know that six modal auxiliaries are used. We do not know whetherthese auxiliaries are of the type we are looking for, and we do not

know which subject and object are involved. If the modal auxiliarieswere not taken into account, one would find six relations where

somebody is the subject and mistakes, lessons, offences, policies,priority and consequences are the objects. All these relations are of

type ASS and are valenced -1.








19

Analysing the text by using semantic text analysis, here PLCA,results in the

following codingfor the first sentence:

SOMEBODY is obligated to SUMMARIZE a badness/harm (.50) (thatTHE PAST_GOVERNMENT FACILITATED (1.00)) and SOMEBODY is

obligated to SUMMARIZE a LESSON about a badness/harm (.50).

This closely follows the original sentence’s syntax. The sentence has

two inflected verbs, namely &dquo;should&dquo; and &dquo;(implicitly: were) made&dquo;. If

we eliminate the prepositional phrase &dquo;for each new government&dquo;,this leaves: Somebody should summarise the mistakes (clause

relative to mistakes: that the previous government made) and(implicitly: somebody should summarise) the lessons of these

mistakes. As the parts &dquo;somebody should summarise&dquo; are mentioned

twice, they are weighted 0.50, which is mentioned between brackets.

When dealing with the network approach, some steps in the codingprocess were already mentioned. Here, the following is added. In the

semantic analysis, the first step is to identify each of the text’s

inflected verbs. In the second sentence, these are &dquo;should&dquo;, &dquo;will&dquo;,

&dquo;was&dquo;, and &dquo;lasted&dquo;.In

fact,there is

anotherinflected verb that is not

explicitly stated. The conjunctive clause &dquo;when the previous systemwas started&dquo; is not semantically linked to &dquo;including the one-sided

personnel policies&dquo;. In other words, the time when &dquo;the offences

included policies&dquo; was not at the time that the past system started.

A more explicit statement would be something like &dquo;including the

one-sided personnel policies (that occurred) when the previoussystem was started.&dquo; The final step is an attempt to preserve the

relations between the clauses as much as possible. The second

sentence becomes after coding:__

.. _ ,’., ,.__ ,,

SOMEBODY is obligated to SUMMARIZE THE

PAST_GOVERNMENT’s OFFENCE (.33) (that is a bad

PERSONNEL_POLICY (that OCCURRED (EXCLUDE) when [there wasTHE PAST_GOVERNMENT’s BEGINNING (.33)1)) and there is THE

CRIMES (.33).SOMEBODY is obligated to SUMMARIZE THE

PAST GOVERNMENT’s OFFENCE (.33) (that is a bad

POLITICS_AS_PRIORITY (.33)) and there is THE CRIMES (.33).

SOMEBODY is obligated to SUMMARIZE THEPAST_GOVERNMENT’s OFFENCE (.33) (that is a CONSEQUENCEfor THE DOMINATED_SOCIETY (1.00) (that was a DURATION (1.00)))and there is THE CRIMES (0.33).

The sentence is coded as three sentences, each part weighted 0.33,but, in the analysis, the sentence is still treated as one sentence. In

total, one clause is excluded from analyses. Now 15 clauses are left,but, when weights are taken into account, this total is 9 clauses. In








20

this latter situation, we find two clauses having depth zero; i.e.,there are two main clauses. Both main clauses contain a modal

auxiliary verb, &dquo;ought&dquo;. In the first of these main clauses, the subject&dquo;somebody&dquo; is linked to the object &dquo;bad&dquo;; in the second main clause,the same subject is found, but this time linked to the object, &dquo;the

offence&dquo;. These relations allow us to examine which combinations of

subjects and objects are found in combination with a specific modal

auxiliary. It is up to the investigator to determine whether or not a

modal auxiliary belongs to one of the classes mentioned earlier;these are ego’s needs, alter’s morality, ego’s potentiality, and alter’s

obligation.

The network text analysis does not determine these relations. Here,no weighting is allowed, only one type of relation results

(evaluation), and the number of clauses is different. It might be that,in the network analysis, the first evaluative statement should not

have been added. This is a decision to be made by the investigator. It

has no consequences for the final conclusion. In case modal

auxiliaries are to be investigated, the investigator should use

semantic text analysis.

CONCLUSIONS

We investigated whether network text analysis should be used to in

the situation where modal auxiliary verbs contain a lot of relevant

information with regard to the topic under study. It turns out that it

is hard to capture the meaning of modal auxiliary verbs using theCETA approach to network text analysis. Therefore, the other type of

text

analysis,the

semantic one,is to

be usedin

this type ofinvestigation.

The problem of selecting sentences containing the desired modal

auxiliary verbs incorporated in the sample has not been treated.

This selection process must be agreed upon. Starting with network

analysis, the investigator might use (random or stratified) sampleparagraphs and code all sentences (to catch the argumentation), or

only code the sentences that contain the desired verbs. In case

semantic text analysis is used, it seems sufficient to code the

sentences containing the desired verbs. What is left is that all thishas to be tested in empirical research.








21

NOTE &dquo; ~ &horbar; ’. ’-’....

* An earlier version of this text was presented at the Fifth

International Conference on Logic and Methodology, Cologne, 3-6

October 2000.

REFERENCES

Carley, K. (1986). "An approach for relating social structure to

cognitive structure". Journal of Mathematical Sociology, 12, 137-189.

Carley, K. (1988). "Formalizing the expert’s knowledge". SociologicalMethods & Research, 17, 165-232.

Carré, B. ( 1979). Graphs and Networks. Oxford: Clarendon Press.

De Ridder, J.A. (1994). Van tekst naar informatie. Ontwikkeling en

toetsing van een inhoudsanalyse-instrument [From text to

information. The

developmentand

testingof an instrument for

content-analysis]. Unpublished Ph.D. thesis, University of

Amsterdam.

Franzosi, R. (1989). "From words to numbers: A generalized and

linguistics-based coding procedure for collecting textual data". In:

C.C. Clogg (ed.), Sociological Methodology 1989. Oxford: Basil

Blackwell: 263-297.

Krippendorff, K. (1980). Content Analysis: An Introduction to Its

Methodology. Beverly Hills: Sage.

Osgood, C.E., Saporta, S. and Nunnally, J.C. (1956). "Evaluativeassertion analysis". Litera, 3, 47-102.

Popping, R. (2000). Computer-Assisted Text Analysis. London: Sage.

Popping, R. and Roberts, C.W. (1997). "Network approaches in text

analysis". In: R. Klar and O. Opitz (eds), Classification and

Knowledge Organization. Heidelberg: Springer: 381-389.

Popping, R. and Roberts, C.W. (1998). "Democracy as reflected in

East-Central European newspapers". In: H. Van der Wusten (ed.),Proceedings of the Conference Transformation Processes in Eastern

Europe, Part IV. Politics and the Environment. The Hague: ESR/NWO:75-86.








22

Roberts, C. W. (1989). "Other than counting words: A linguisticapproach to content analysis". Social Forces, 68, 147-177.

Roberts, C.W. (1997a). "Semantic text analysis: The problem of

linguistic ambiguity". In: C.W. Roberts (ed.), Text Analysis for the

Social Sciences: Methods for Drawing Statistical Inferences from Texts

and Transcripts. Mahwah, NJ: Lawrence Erlbaum Associates: 55-77.

Roberts, C.W. (1997b). "A generic semantic grammar for quantitativetext analysis: Applications to East and West Berlin radio news

content from 1979". In: A.E. Raftery (ed.), Sociological Methodology

1997. Oxford: Basil Blackwell: 89-129.

Roberts, C.W. and Popping, R. (1996). "Themes, syntax, and other

necessary steps in the network analysis of texts". Social Science

Information, 35, 657-665.

Shapiro, G. (1997). "The future of coders: Human judgments in a

world of sophisticated software". In C.W. Roberts (ed.), Text Analysisfor the Social Sciences: Methods for Drawing Statistical Inferencesfrom Texts and Transcripts. Mahwah, NJ: Lawrence Erlbaum: 225-238.

Van Cuilenburg, J.J., Kleinnijenhuis, J. and De Ridder, J.A. (1986)."A theory of evaluative discourse: Towards a graph theory of

journalistic texts". European Journal of Communication, 1, 65-96.

Van Cuilenburg, J.J., Kleinnijenhuis, J. and De Ridder, J.A. (1988)."Artificial intelligence and content analysis". Quality and Quantity,22, 65-97.

Weber, R.P. (1990). Basic Content Analysis. Beverly Hills: Sage.

BMS Modal Auxiliaries in Text Analysis

Documents