1 The Monitor Model and its Misconceptions A Clarification Michael Carl

2025-04-30 0 0 232.57KB 20 页 10玖币

侵权投诉

The Monitor Model and its Misconceptions: A Clarification

Michael Carl,

Kent State University

Abstract

Horizontal (automatic) and vertical (control) processes have been observed and reported for a long time in

translation production. Schaeffer and Carl’s Monitor Model integrates these two processes into one

framework, assuming that priming mechanisms underlie horizontal/automatic processes, while

vertical/monitoring processes implement consciously accessible control mechanisms. The Monitor Model

has been criticized in various ways and several misconceptions have accumulated over the past years. In

this chapter, I update the Monitor Model with additional evidence and argue that it is compatible with an

enactivist approach to cognition. I address several misconceptions related to the Monitor Model.

Keywords: Translation Process Research, Monitor Model, Computationalism, Enactivism

1. Introduction

Over the past 40 years, Translation Process Research (TPR) has developed numerous models to explain

“what goes on the mind of translators”. One influential theory in this respect is the computer theory of

mind (CTM). The CTM views the mind as a system of computations, a program in the brain, that

manipulates symbols according to a set of rules. The program that runs in the brain could equally be

implemented and executed on a machine, and thus simulate human intelligence and cognition, outside the

brain. The symbols that are believed to be processed in the brain (or another computational device) are

physical states that are manipulated with truth-preserving rules, regardless of their semantic content.

Pereplyotchik (2016, 171) points out that in “classical” cognitive architectures, such as ACT-R, behavior

is thought to emerge from production rules that fire in response to dynamically changing contents of

buffers

. While the buffer states are thought to be mental representations (symbols) of some outside state

of affairs, the procedural knowledge that is encoded in production rules is not representational.

Probabilities, frequencies, rhythm of action, etc., emerge from firing rules but are non-representational in

nature.

Mental representations are, according to this view, the basis on which we produce (i.e., compute)

inferences, e.g., infer meaning, plan and execute actions, including translations. Accordingly, “cognitive

processes can be described and explained as [a] manipulation of formal symbols in a language of thought”

(Martín de León 2017, 109). Translation, under a CMT, is thus a type of mechanical theorem-proving,

where “[t]ranslation was understood as a rule‐guided transformation of symbols from one code into

symbols of another code” (Martín de León 2017, 109). In support of this functional, computationalist

view, Bechtel and Abrahamsen (2002, 191) maintain that “the time consumed by a particular cognitive

process is a matter of implementation and does not inform us as to the nature of the architecture [of the

mind] itself.” The temporal structure of these operations is of no interest for the CTM.

a limited set of ‘variables’ that are believed to exist in a cognitive module.

Criticism of the CTM took shape in the 1980s. The computationalism controversy revolved mainly

around the notion of whether the mind produces and manipulates mental representations of an outside

world. Perhaps not by coincidence, starting from the mid-1980s, Translation Process Research (TPR) set

out to investigate "What happens in the heads of translators" (Krings 1986, see also Königs 1987) and to

assess “by what observable and presumed mental processes do translators arrive at their translations?”

(Jakobsen 2017, 21). A large body of data and research findings has been produced in the past three

decades that documents, among other things, the role of expertise, and translation directionality,

ergonomic, linguistic, and emotional factors, as well as the usage of (external) resources - such as

computer assisted translation and machine translation (MT) - on the translation process. TPR focuses

thereby on process related issues, such as translation duration, translation effort, and the distribution of

attention. In contrast to CTM, the temporal structure of cognitive operations is of central importance in

TPR.

Theories and models that reject the CTM are non-computationalist or postcognitivist; some are also non-

representational. Sometimes these theories are labeled 4EA cognition (embodied, embedded, enactive,

extended, and affective). Postcognitivists have proposed a large number of alternative theories (for an

overview see e.g., Risku and Rogl 2020) which suggest that the body and the environment have a

facilitating or even a constitutive role in cognition. According to some postcognitivists, cognition does

not only take place in the brain but may be distributed in the body and/or the environment, knowledge is

embodied and situated, emotions and rational thought cannot be separated in human experience, cognition

and action can be direct without intermediate mental representation of an external world, which led some

researchers to assume that the brain is a resonant organ rather than a representational one (Ryan and

Gallagher 2020). In this view, the temporal structure associated with cognition and action is crucial for

the understanding of the involved processes.

Hutto and Myin (2017), for instance, endorse a radical enactivist version of postcognitivism. They make a

distinction between basic cognition and higher-order, content-involving cognition. Basic cognition, they

say, is intentional but does not involve representational content

. Higher-order mental representations are

assumed to have “special properties – e.g., truth, reference, implication – that make it [the

representations] logically distinct from, and not reducible to mere covariance relations” (Hutto and Myin

2013, 67). Basic cognition is non-representational, as it does not specify representational content.

However, the notion of content is “elastic enough” to include “accuracy, veridicality or some other kind

of satisfaction condition” (ibid.).

Within an enactivist framework the question thus shifts from what counts as a representation of “real

entities” to whether or to what extent do translators engage in basic or higher-order (e.g., propositional)

content to produce translations. Rolla and Huffermann (2021) extend Hutto and Myin’s framework with a

notion of shared know-how. Linguistic agency, they say, requires sophisticated forms of shared know-

how on which the normativity of human cognitive capacities rests. Rightness and truth are in this view

relative to paradigmatic sets of actions, rather than to external entities and their mental representations.

Rolla and Huffermann suggest a hierarchy of three different levels of content in which higher-order

For Chemero (2010, 77) ‘non-representational’ can be any state in a processing chain “produced by one part of the

system, for the use by some other part of the system.” This notion of basic representation does not assume any

particular higher-order content or satisfaction condition. But see Rolla and Huffermann (2021).

content implies the existence of lower-level content. All levels of content and all kinds of cognitive

performance involve know-how with different, but stable and reproducible success conditions.

Not everyone agrees with the notion of (non-) representationalism. Carey (2009), for instance, finds the

notion of non-representationalism “puzzling.” According to her, there are different types of

sensory/perceptual, core and conceptual representation, which have different characteristics, independent

from their “veridical” status. Perceptual representations, she says, “are iconic or analog, whereas at least

some conceptual representations are stated over discrete, arbitrary symbols” (ibid. 8). “Representations

are graded in robustness or strength, are constructed in real time, and are subject to multiple interacting

influences during the processes of construction” (ibid. 48). Representations in “core” cognition “need not

be (and often are not) veridical and therefore need not be knowledge” (ibid. 10). What Hutto and Myin

take to be a defining criterion for representation-hood (i.e., truth conditions) is, for her, the veracity of

representations. Despite those differences, researchers seem to agree, however, that there are different

levels of representation and processing. And for the translation process, specifically, different processing

levels have been conceptualized in the Monitor Model (Schaeffer and Carl 2013; 2015) in terms of

automatic vs. monitoring processes.

In section 2, I trace the development of the Monitor Model within a historical context. I show how the

Monitor Model was conceived as an approach to integrate automatism and monitoring processes in

translation. Section 3 develops the concept of automatism in translation in more detail and shows how it

can be considered an instance of goal-oriented basic cognition. Priming, I suggest, is an automatism that

is important in translation production and a form of basic cognition that does not involve mental

representation and specifications of truth conditions. In section 4, I address some misconceptions

regarding TPR and the Monitor Model. I argue that TPR and the Monitor Model are not primarily focused

on developing a language of thought or on studying representational or conceptual content, such as truth,

reference, or implication of translations. Rather, TPR and the Monitor Model share many features with a

non-representational, radical enactive framework, stipulating that much of translational activity can be

explained in terms of basic content.

2. TPR and the Monitor Model

In the editorial introduction to Krings (2001) the translation process is described as follows:

When human beings translate, they construct meanings from the sentences they read,

then take this meaning and express it in another language, taking into account all of the

nuances of the source and target cultures, the textual world of the text in both cultures,

and their knowledge of the languages involved and the differences between them

(Koby, in the editor’s introduction to Krings 2001, 6-7).

Despite the fact that this statement implies the existence of representations of culture, language, and

textual meaning, the main focus of Krings’ work was not on representational content, but rather on the

dynamics of post-editing effort. Using Think-Aloud (TA) protocols, his work aimed at overcoming the

“complete lack of empirical, controlled observations concerning all issues associated with the post-editing

process” (ibid. 65). He introduced the famous distinction between temporal, technical, and cognitive

effort and provided numerous examples that explain his conception of translational effort, but he does not

address issues of reference or truth conditions in translation.

The same holds for Lörscher (1991) who used TA protocols to investigate problem solving and

translation strategies of professionals and university students. According to Lörscher, there are two kinds

of translation: automatic, nonstrategic translation, and controlled strategic translation which involves

problem solving. TA is indicative of strategic translation; little TA can be observed in phases of automatic

language processing. Lesser TA for professional translators shows, therefore, that they have reached a

higher degree of automatization. However, as a translation process researcher, Lörscher was less

interested in developing a (higher-order) notion of representation, or even assessing the quality of

translation. The development of a representational CTM was not primarily at stake in earlier TPR.

In the 1990s, keystroke logging and eye tracking technologies were introduced to investigate the

translation process at a much finer-grained temporal resolution (below the level of a second). Based on

the assessment of such logs it has been suggested that at least two concurrent processes take part in the

translation process. Englund-Dimitrova (2005, 26) notes that “there are segments which are translated

apparently automatically, without any problems, and other segments where the translation is slow, full of

many variants and deliberations, which necessitates a problem-solving approach and the application of

strategies.” Based on similar observations, Tirkkonen-Condit (2005) distinguishes a default translation

procedure (or automaton) and its monitoring mechanism. The default translation automaton

“operates on

a lexical as well as syntactic level” (ibid. 405) and produces translation hypotheses while the “monitor

supervises text production processes, and triggers disintegration of the translation activity into chunks of

sequential reading and writing behavior” (Carl and Dragsted 2012, 127). Indeed, much of TPR focused

on conceptualizing, measuring, and evaluating various aspects of lexical and syntactic (translation)

ambiguity and its impact on effort and quality in post-editing and from-scratch translation.

Schaeffer and Carl (2013; 2015) introduced a recursive model of the translation process in which priming

mechanisms are identified to activate shared lexical and syntactic representations

. Automated translation

production is driven by priming processes in which “features of both source and target language items

[are activated] which share one single cognitive representation” (2015, 21). However, shared, non-

selective activation (de Groot 1997) implies distributed representations and tensor product parsing

processes which do not share the structure of the sentences they parse. These non-isomorphic cognitive

structures do not account for assumed systematicities that classical computationalism suggests. Priming –

as modeled through the translation automaton – may be considered a form of basic cognition that

establishes the bulk of translation relations. Vertical monitoring processes, in contrast, work in a

monolingual mode, controlling and evaluating the output of the translation automaton: “Vertical

processes access the output from the automatic default procedure recursively in both the source and the

target language and monitor consistency as the context during translation production increases” (ibid. 38).

Schaeffer and Carl maintain that “[o]nly after the output has been received and evaluated by some kind of

monitor can these processes become consciously controlled” (ibid. 26).

The distinction between more or less automatized processing is not addressed in “classical”

computationalism. Cummins (1989, 20) speculates that “perhaps consciousness isn’t essential to the mind

in the way that cognition is.” Cognition, in his view, is the processing of basic mental symbols in an

internal language of thought (LOT) with a context-free grammar. But Fodor (1983) himself is skeptical

Also labeled literal translation automaton or literal translation procedure

Primed representations are basic representations: they carry basic content; they do not specify propositional content

of truth, implication, or higher-order accuracy conditions.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

1TheMonitorModelanditsMisconceptions:AClarificationMichaelCarl,KentStateUniversityAbstractHorizontal(automatic)andvertical(control)processeshavebeenobservedandreportedforalongtimeintranslationproduction.SchaefferandCarl’sMonitorModelintegratesthesetwoprocessesintooneframework,assumingthatprimingmech...

展开>> 收起<<

1 The Monitor Model and its Misconceptions A Clarification Michael Carl.pdf

共20页,预览4页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

1 The Monitor Model and its Misconceptions A Clarification Michael Carl

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: