Poverty ’ s Agent : The Framing of Poverty and Responsibility in Colombian Newspapers

This article examines, by way of linguistic analyses, to whom or what Colombian newspapers assign notions of responsibility, cause or blame when they address the topic of poverty. It is our assumption that specific attitudes may reveal themselves in linguistic patterns (among other things). Here, we examine the verbs that appear with the word pobreza (poverty) as syntactic and semantic argument. Specifically, we examine the agent, or highest argument of the different verbs in detail, since the agent is the initiator of the verbal action and thus has the potential of accepting responsibility or blame for the action described.


Introduction
Without making any specific claim about the humanities' responsibility in relation to social relevance, this article offers an example of a linguistic research project that has been driven precisely by the desire to address a specific social problem.
In concrete terms, we aim to examine, by way of linguistic analyses, to whom or what the Colombian newspapers assign notions of responsibility, cause or blame when they address the topic of poverty.It is our assumption that specific attitudes may reveal themselves in linguistic patterns (among other things).Here, we examine the verbs that appear in connection with the word pobreza (poverty) as syntactic and semantic argument.In particular, we examine the agent, or highest argument (explained below) of the various verbs in detail, since the agent is the initiator of the verbal action, and therefore has the potential of accepting responsibility or blame for the action described.
The analysis considers the topic primarily from a quantitative perspective, in that features will be classified and counted (McEnry and Wilson 1996, 76).However, the analysis is also qualitative in nature, because, according to Susan Conrad (2002): all studies include both aspects of analysis to some extent.Recognizing patterns of language use necessarily entails assessing whether a phenomenon is common or unusual -a quantitative assessment.At the same time, numbers alone give little insight about language.Even the most sophisticated quantitative analyses must be tied to functional interpretations of the language patterns.(Conrad 2002, 78) This study is a contribution to the cross-disciplinary project polame, (Poverty, Language and Media), whose objective is "to study the language used by the most important newspapers in Mexico, Brazil, Argentina and Colombia to construct and convey the notions of poverty" (UiB 2016).The project has compiled an electronic linguistic corpus comprised of newspaper articles on poverty from the four countries included in the research project.
The present study is limited to the Colombian newspapers in the polame-corpus (El Tiempo, El Espectador, El Colombiano), and to verbs that have pobreza as part of their subcategorization. 1 Thus, adverbials and other adjuncts 2 have not been taken into account.
In the subsequent sections (2.1-2.2),we review relevant theories related to thematic roles, grammatical voice, lexical aspect and the influence of agenda-setting media.

Theoretical considerations 2.1. Argument structure, voice and lexical aspect
From a linguistic standpoint, the examination of a predicate's argument structure represents a formal, efficient and measurable way of identifying and analysing the participants of the event described by the verb.This, in turn, is an indispensable step in the quest for a responsible/culpable party for the concerns addressed by the relevant predicates.
The present paper will employ Lexical-Functional Grammar's (lfg) descriptive apparatus for the identification and classification of various kinds of verbs and their arguments.We have chosen this framework because we will be reviewing the relationship between syntactic argu-1 The term subcategorization is used in Lexical-Functional Grammar (lfg) theory "to refer to the set of grammatical relations which are specified in a verb's lexical entry" (Kroeger 2008, 17).The lexical entry of a word "contains information about the meaning, pronounciation, and grammatical features of that particular word" (Kroeger 2008, 14). 2 An adjunct is "a kind of phrase [that] can be added freely to virtually any clause that describes an event" (Kroeger 2008, 7).Thus, an adjunct is not an argument of the predicate (Kroeger 2008, 14).
ments and semantic roles, and the lfg framework is designed to give in-depth accounts of precisely this relationship, among others.lfg operates with three different levels of representation for sentence structure: a(rgument) structure, which deals with predicate-argument information, such as Agent, Patient or Theme; f(unctional) structure, which deals with functional information such as subj and obj (Austin 2001, 5); and c(onstituent) structure, which is the overt surface phrasal syntactic representation (Dalrymple 2001, 45).
In the search for an agent and/or a responsible party for the verb action, it is the argument structure that is the most central level.Finocchiaro, Capasso, Cattaneo, Zuanazzi, and Miceli (2015, 223) state that "'argument structure' captures the idea that verbs denote relations between entities, and their role in the event is completely independent of the specific words used."Thus, the argument structure of a verb like combatir (combat), which appears frequently in the corpus with the word pobreza as argument, could be represented in the following manner: combatir <agent, patient> Furthermore, the predicate is associated with specific grammatical relations that are linked to the arguments that make up the argument structure.The mapping of the relationship between the semantic roles (in the a-structure) and grammatical (in the f-structure) is referred to as linking (Kroeger 2008, 17), and can be represented in the following manner: combatir <agent, patient> subj obj   (2) illustrates the fact that for combatir, its agent-role (see definition below) appears in the subject position in an unmarked sentence, and that the patient-role appears in the object-position.This means that the identification of the syntactic arguments (f-structure) of the verbs in the corpus also makes their semantic arguments (a-structure) accessable.Kroeger (2008, 9) proposes an inventory of ten thematic roles, eight of which have been identified in the present analysis:3 agent: causer or initiator of events experiencer: animate entity which perceives a stimulus or registers a particular mental or emotional process or state recipient: animate entity which receives or acquires something beneficiary: entity (usually animate) for whose benefit an action is performed instrument: inanimate entity used by an agent to perform some action theme: entity which undergoes a change of location or possession, or whose location is being specified patient: entity which is acted upon, affected, or created; or of which a state or change of state is predicated stimulus: object of perception, cognition, or emotion; entity which is seen, heard, known, remembered, loved, hated, etc. location: spatial reference point of an event accompaniment (or comitative): entity which accompanies or is associated with the performance of an action The interaction between these thematic roles and the various kinds of verbs that subcategorize for them, is accounted for in part by the thematic hierarchy, which is a hierarchy of relative prominence for the thematic roles (from left to right) (Bresnan 2001, 307).In other words, this hierarchy ranks the thematic roles with respect to their likelihood of functioning as logical subject (Austin 2001, 13).The hierarchy is as follows: Agent > beneficiary > experiencer/goal > instrument > patient/theme > locative (Bresnan 2001, 307) This hierarchy has turned out to be especially relevant for the explication of cases where thematic roles other than the agent function as subject, such as the passive and middle voices.
For the identification of agent/responsible party for the predicates that appear with pobreza, it is also necessary to classify the agent roles according to their degree of animacy.This is because an inanimate entity, even if it has the potential of denoting cause, cannot be assigned blame or responsibility.Hence, the agents of the different predicates have been classified, among other things, according to Silverstein's animacy hierarchy, as presented by Deane (1987, 67): 1 st person pronoun > 2 nd person pronoun > 3 rd person pronoun > 3 rd person demonstrative > Proper name > Kin-Term > Human and animate NP > Concrete object > Container > Location > Perceivable > Abstract For our analysis, we have added a further distinction: plural NPs are lower on the animacy scale than singular NPs, because number has been identified as a feature that influences the animacy of an NP (Bianchi 2006(Bianchi , 2025)), and also because the ascription of blame to collective entities is questionable (Pizarro 2014, 234).
Furthermore, since animacy correlates with the aforementioned concept of prominence (shown in the thematic hierarchy), in that agents typically are high in animacy (Fauconnier 2011, 534), the analysis will examine the frequency with which inanimate agents appear with the different predicates.This is relevant because inanimates are not expected, crosslinguistically, to occur as agents (Pizarro 2014, 533), and because they cannot accept blame.
In addition, it will be necessary to examine a specific kind of predicate that does not allow the identification of an agent argument at all.This is the middle voice, made up of unaccusative predicates (García-Miguel 1985, 323), and it is distinguished not only by the absence of an agent, but also by the impossibility of including one; nobody can be implicated as initiator of the process described by the predicate (García-Miguel 1985, 323).This is different from the passive voice, in which the predicate is analysed as having a suppressed agent argument (Austin 2001, 16).This means that, in the case of the middle voice, the verb has a complete argument structure in which the highest argument is some argument other than the agent, whereas for the passive voice, the predicate has an "in-complete" argument structure, because the agent argument has been suppressed.
There is a third semantic distinction (in addition to argument structure and voice), which is important for the linguistic examination of the ascription of blame, and that is lexical aspect.This is relevant not only because it intersects with argument structure and linking at the lexicalsemantic interface (Filip 2012, 721), but also because it reflects what the prospective agent is held responsible for.For example, is he held responsible for eliminating poverty, or simply for combatting it?In the first case, the verb is telic, while in the second case it is atelic.According to Filip (2012, 721), "Telic verbs express 'an action tending toward a goal', while atelic ones describe situations that 'are realized as soon as they begin'[…]" One specific prediction related to telic and atelic predicates is relevant to our study, and that is the fact that they "yield different patterns in the ordinary […] progressive: an atelic predicate V should licence the inference from the progressive to the perfect (X is Ving entails that X has Ved), while a telic predicate should not (X is Ving does not entail X has Ved)" (Marín & McNally 2011, 485).In other words, X está combatiendo la pobreza4 entails X combatió/ha combatido la pobreza,5 but X está eliminando la pobreza6 does not entail X eliminó/ha eliminado la pobreza. 7ence, the predicates found in the corpus will not only be classified according to argument structure and voice, but also according to lexical aspect.

Media influence on public perception
Since the aim of the polame project is to ascertain how the agenda-setting media portray poverty, i.e. whether they instil a notion of poverty in the mind of the public that makes it difficult to combat it, it is necessary to consider previous studies that have examined measurable effects on the public of different media portrayals.
Several scholars have found that there is indeed an influence.Wanta, Golan and Lee (2004, 364) state that "under certain circumstances, the news media do tell people what to think by providing the public with an agenda of attributes -a list of characteristics of important newsmakers."Shanto Iyengar (1990, 19) addresses the issue of poverty specifically, and claims: "How people think about poverty is shown to be dependent on how the issue is framed."Martin Gilens (1996, 528) declares that "past research has shown that the mass media can exert a powerful influence on public perceptions and attitudes […]."Chauhan and Foster (2014, 391) employ the Social Representations Theory (srt) to explain how this influence works: "The theory rejects the notion of knowledge being a facsimile of some objective event or a mere description of events in the social world.Instead, it considers knowledge to be produced through acts of communication that are guided by the interests of the people involved."This is important, because "[t]he role of public opinion in democratic societies is critical, since it can influence governmental responses to social problems" (Chauhan & Foster 2014, 401).
The aim of the present paper is not to measure the public's response to the way Colombian newspapers' frame poverty, although we operate under the assumption that what we observe here could influence the way the community perceives this pervasive social challenge.

Analysis 3.1. Method
The present analysis, like the polame project as a whole, uses a linguistic corpus as its empirical basis.The polame-corpus is an annotated, searchable compilation of newspaper articles about poverty, from Brazil, Colombia, Agentina and Mexico.It comprises 37 million words, with articles assembled between the years 2000 and 2014.The use of a confined linguistic corpus is of great value for many different kinds of linguistic investigation, because it allows the investigator to carry out exhaustive, linguistically itemized searches on excerpts of authentic language, taking into account all relevant examples, not only those that conform to some pre-conceived hypothesis.
Furthermore, it has been shown that corpus linguistic techniques have been "important for discerning the strong associations that exist between the lexicon and grammatical structures" (Chauhan & Foster 2014, 81), and voice and argument structure, which are what we are focusing on here, are concerned precisely with the association between the lexicon and grammatical structures.
One of the greatest benefits of an electronically stored linguistic corpus, however, is the fact that it is possible to conduct large-scale searches on massive chunks of text in an automated and efficient manner.For the present analysis, this means that we have been able to identify each and every instance in which a verb co-occurrs with the lexeme pobreza.
There are, nevertheless, a great many linguistic phenomena that do not lend themselves to automated retrieval.These include the identification and delimitation of a verb's argument structure and voice, as well as the lexical aspect of predicates.A substantial part of the corpus search for the present analysis, therefore, has been conducted manually.This has precluded the possibility of a comprehensive search of the whole corpus, and that is the reason why this paper is limited to the Colombian newspapers.Annelie Ädel (2014, 68) states: "it is often not doable to examine a full set of data qualitatively.An important issue, then, is how to select a subset for further qualitative analysis."The makeup of the polamecorpus allows us to select a natural subset, since the corpus is divided into geographical domains.
Hence, our search has been conducted in the following manner: Firstly, the corpus was searched for all strings that contained a verb with the lexeme pobreza as one of the five8 words to the right or to the left of it.The next step was to manually examine the search results to discard any instances in which the lexeme poverty was not an argument of the verb.All the remaining verbs were registered and counted individually.Subsequently, the most frequently used verbs were grouped together according to their subcategorization and voice.This step also involved identifying what thematic role the lexeme pobreza displayed in each case.Thereafter, the agents of these verbs were examined in further detail to ascertain specifically what kind of entities filled the agent role in each case (see section 3.2 for the specifications).
The following section presents the results of this analysis.

Results
There are 1564 cases in total of verbs that appear with pobreza as argument.These cases represent 82 different verbs.For these verbs, pobreza is either the direct object, the subject (of a passive or an active verb), the indirect object, or a prepositional object.The most frequent verbs to oc-cur with pobreza as argument, are the following (the number to the right of each verb indicates its number of occurrences): 1 These twelve verbs are the ones whose argument structures will be scrutinized in further detail here.
In order for the analysis of the verbs' argument structures to be relevant for the issue of responsibility framing, we must also classify the verbs according to what kind of action they hold the agent responsible for.For instance, the agent of a verb like tener should not be grouped together with the agent of a verb like eliminar, because they are not performing the same kind of action (or any action at all, in the case of tener).
The analysis of the corpus occurrences has yielded the following four relevant semantic categories for the twelve verbs analysed.Some of the verbs may be categorized as belonging to more than one semantic category, depending on context.All in all, these verbs constitute 1090 cases, which were analysed in detail.74% (812 cases) were instances of verbs referring to solving poverty, 1.5% (16 cases) referred to cause or culpability, 10% (105 cases) described measurement or definition, and 14% (157 cases) described an agentless rise or fall in poverty.The subsequent step of our analysis involves the examination of each verb's agent argument, or rather the argument highest on the thematic hierarchy (see section 2.1).The following modifications to the Silverstein hierarchy (Deane 1987, 67) were made based on our analysis: a) The three specifications "1st person pronoun > 2nd person pronoun > 3rd person pronoun" have been reduced to "personal pronoun sg > per-sonal pronoun pl," since all personal pronoun agents in the corpus are 1st person plural.
b) The 3 rd person demonstrative, kin-term and concrete object as agents have been eliminated, since they do not appear in the corpus.c) Singular entities have been placed higher on the scale than plural ones, since our definition of animacy is motivated by our desire to identify a responsible or culpable party, and assigning culpability to a plural entity is questionable (Pizarro 2014, 234).
d) The categories "proper name" and "human and animate NP" have been conflated here, because they have the same kind of reference in the corpus.
The results of the analysis show that: of the agents of the telic verbs that refer to solving the problem (erradicar, eliminar, superar, acabar  The highest argument of the telic verbs that describe an agentless rise or fall in poverty (only acabar, and only 1 case) is the lexeme pobreza.
Of the highest arguments of the atelic verbs that describe an agentless rise or fall in poverty (disminuir, caer, aumentar (156 cases)), 83% are the lexeme pobreza 17% are NPs with pobreza as modifier 24Before we extract the most important findings from the preceding overview, it is important to revisit the thematic hierarchy and consider how it accounts for the argument structure of some of these predicates.Some of the verbs that have received a detailed scrutiny of their highest argument, lack an agent argument altogether.This is the case for tener, caer, and some cases of acabar, disminuir and aumentar.
For all of these verbs, the highest, and only, argument, is the patient: "entity which is acted upon, affected, or created; or of which a state or change of state is predicated" (Kroeger 2008, 9).Since the patient is the only argument in these cases, it is the only one that can be assigned the subject position.All these predicates, except for tener, are cases of the middle voice.Tener is not a case of the middle voice because it does not denote a process that affects the subject's referent (García-Miguel 1985, 323, rae 2009, 3037-38), but rather a state.
In summary, the main findings of the present analysis are the following: 1) There are more than 16 times as many cases of verbs referring to solving poverty as there are of verbs referring to cause.2) Among the verbs referring to solving the problem, there are more than twice as many atelic verbs as there are telic ones.In other words, mentions of combatting and minimizing poverty are far more frequent than mentions of eliminating or terminating it altogether.
3) For the verbs that referred to cause or culpability, none of the agents are animate.4) 31% of all agents/highest arguments, for all of the verbs, are nonidentifiable.5) For verbs referring to solving poverty, the most frequent agent arguments are political, religious or ideological organizations.6) For verbs referring to cause or culpability, the most frequent agent argument is social phenomena.7) The most frequent highest argument for verbs of measurement or definition is geographical area.8) 19% of all of the verbs examined describe an agentless rise or fall in poverty.

Conclusions
The aim of the present article has been to use linguistic analysis to examine to whom or what Colombian newspapers assign notions of responsibility, cause or blame when they address the topic of poverty.Several scholars (Wanta, Golan & Lee 2004;Iyengar 1990;Gilens 1996;Chauhan & Foster 2014) have found that mass media can exert a powerful influence on public perceptions and attitudes.The present paper operates on the assumption that what we observe here could influence the way the community perceives this pervasive social challenge.
In order to ascertain how Colombian newspapers assign notions of responsibility, cause or blame, we have examined the verbs that appear with the word "pobreza" as syntactic and semantic argument.We have based the analysis on the lfg framework, and scrutinized the agent, or highest argument of the various verbs in detail, since the agent is the initiator of the verbal action, and thus has the potential to accept responsibility or blame for the action described.
For the identification of the agent/responsible party of the predicates that appear with pobreza, we classified the agent roles according to their degree of animacy, using a modified version of Silverstein's animacy hierarchy (Deane 1987, 67).An animacy ranking was deemed necessary because an inanimate entity, despite having the potential of denoting cause, cannot be assigned blame or responsibility.
The predicates' lexical aspect was also registered, both because it intersects with argument structure and linking at the lexical-semantic interface, and because it reflects what the prospective agent is held responsible for, an accomplished action, such as eliminate, or a process, such as combat.
The verbs were also subdivided into the following four semantic categories: a) Verbs that refer to solving poverty b) Verbs that refer to cause/culpability for poverty c) Verbs that describe measurement/definition of poverty d) Verbs that (lexically) describe an agentless rise or fall in poverty The most important findings were: 1) There are more than 16 times as many cases of verbs referring to solving poverty as there are verbs referring to cause.2) Among the verbs referring to solving the problem, there are more than twice as many atelic verbs as there are telic ones.In other words, mentions of combatting and minimizing poverty are far more frequent than mentions of eliminating or terminating it altogether.
3) For the verbs that refer to cause or culpability, none of the agents are animate.4) 31% of all agents/highest arguments, for all the verbs, are non-identifiable.
5) For verbs referring to solving poverty, the most frequent agent arguments are political, religious or ideological organizations.6) For verbs referring to cause or culpability, the most frequent agent argument is social phenomena.7) The most frequent highest argument for verbs of measurement or definition is geographical area.8) 19% of all of the verbs examined described an agentless rise or fall in poverty.
It is worth noting that none of the verbs have arguments that refer to the general public's attitude towards poverty, i.e. their perception of it, which is the focal point for the polame-project.In other words, society's perception of the poor as victims or culprits, and the effects this view may have on the fight to overcome poverty, is not reflected in the verbal argument structure of the articles in the newspapers examined.
The present study could, and should, be used as a basis for a comparative scrutiny of the same phenomena for the other countries represented in the polame-project.It may also benefit from a supplementary study that takes into account a larger segment of the linguistic contexts in which the verbs considered occur.

Figure
Figure (2) illustrates the fact that for combatir, its agent-role (see definition below) appears in the subject position in an unmarked sentence, and that the patient-role appears in the object-position.This means that the identification of the syntactic arguments (f-structure) of the verbs in the corpus also makes their semantic arguments (a-structure) accessable.

Table 1 .
The semantic categories of the verbs