Description and Quality Assessment of Poetry Translation: Application of a Linguistic Model

This article presents a linguistic model, based on systemic functional linguistics (SFL), for describing and comparing poetry translations. The proposed model takes both the form and meaning of poetry into consideration and involves linguistic analyses at the levels of graphology, phonology, lexicogrammar and context. To illustrate the applicability of the model, we offer an analysis of Rabindranath Tagore’s Stray Birds in English and its three Chinese translations, point out the choices made by Tagore and the translators at different levels, and discuss the translation shifts in the target texts. On the basis of a contextual analysis, we relate the target texts with the Chinese norms of translation and comment on the quality of the translations. Our intention is to prove that linguistic theories offer a powerful tool for analysing poetry translation and offer new possibilities in translation studies from the perspective of SFL .


Introduction
The engagement of systemic functional linguistics (SFL) with translation studies started with Halliday's (1956Halliday's ( , 1962 early investigation of machine translation and Catford's (1965) application of Halliday's (1961) scale and category theory to construct a linguistic model of translation. As has been verified in various studies, SFL is capable of analysing and interpreting texts of different types and their translations, such as advertisements (Steiner, 2004), scientific writings (Veroz, 2017), film subtitles (Espindola, 2016), novels (Wang, 2007), dramas Ma, 2018, 2020), and political texts (Munday, 2018). Although it has been proven that SFL is suitable for analysing poetry (see e.g. Halliday, 1988;Hasan, 1988), we have not found an SFL-based model that is specifically aimed at describing poetry translation and assessing its quality. In this study, we attempt to construct a comprehensive model that takes both the form and meaning of poetry into consideration, and investigate the choices made by the poet and the translators at various levels (or strata) of language based on SFL. Our research questions are as follows: (i) Which levels of linguistic analysis are needed in a model that compares the similarities and differences between poems and their translations? (ii) How valid is the constructed model when assessing the Chinese translations of English poems?

2
An Overview of Related Theoretical Concepts

A Brief Review of Some Models on Translation Quality Assessment
The quality of a translated work is a subject of much debate as it involves the nature of translation and presupposes a theory of translation. To measure the original text and its translation, House (2015) suggests that we take the following three aspects into consideration: (i) the relationship between the original and the translation, (ii) the relationship between how the original and the translation are perceived by the author, the translator and the readers and (iii) the effects of the translation and the views on this translation.
On the basis of the above-mentioned notions and regarding translation as the re-contextualisation of the source text in the target culture, House (2015) offers one of the first attempts at presenting a linguistic model of translation quality assessment. Her model applies SFL to measure equivalence between the source text and the target text. A comparison is first made at the textual level, involving the analysis of theme, mood and transitivity. Then, an investigation is conducted in terms of register and genre based on the contextual parameters of field, tenor and mode. Her model provides theoretical motivations and consistent explanations for the concept of overt and covert translation.
In recent decades, other attempts have been made to apply SFL to translation quality assessment. For instance, Zhu (1996) constructs his model entitled 'Structure of Meaning' by integrating SFL with speech act theory, involving three levels of analysis: (i) linguistic composition, with a focus on lexicogrammar, phonology and graphology, (ii) interactional dynamics that highlights the exchange of connotative meaning and illocutionary forces and (iii) aesthetic impact that manipulates information through textual means.
We also find several frameworks that focus specifically on poetry translation. For example, Pallavi and Mojibur's (2018) model draws on the pragmatic theories of speech acts and conversational maxims and combines the previous frameworks proposed by House (1997) and Dastjerdi et al. (2008). Their model points out the pragmatic features in poetry translation and adopts Nida's (1964) dynamic equivalence to determine the equivalent translation of pragmatic force. Dahlgren (2005) proposes a framework that incorporates analysis at the levels of phonology, syntax, semantics and pragmatics. Modelled on pragmatic theories, Dahlgren (2005) examines the difficulties in poetry translation and the trade-offs made by translators, pointing out the translators' failures in preserving prosodic elements, lexical mistranslations and syntactic oddity. She further proves the applicability of linguistic theories to the analysis of poetry translation.

2.2
Some Key Terms Revisited Equivalence, one of the core concepts in translation studies, legitimises the notion of translation and explicitly points out the double-bind relationship of a translated (target) text, which is constrained both by the source text and the target culture. Furthermore, equivalence is a relative concept and is determined by various factors, including the source and target languages, the source and target cultures, the source text that reflects the linguistic and stylistic source language norms, the expectation norms of the target text, the translator's strategies of translation and the tradition of translation in the target culture. Thus, to investigate equivalence, we need a holistic and comprehensive linguistic theory that is not only suitable for text analysis, but is also capable of analysing context. SFL is a holistic theory of language, which allows us to measure translation equivalence and shift along its various dimensions (Matthiessen, 2001;Halliday, 2009). To theorise language as a resource for making meaning, Halliday (1978) has opted to organise his theory according to the 'architecture' of language, which conceptualises language in terms of various relations defined along semiotic dimensions. These dimensions include the hierarchy of stratification, the cline of instantiation, the spectrum of metafunction, the hierarchy of rank and the hierarchy of axis. Matthiessen (2001) has named these dimensions as the environments of translation, within which we can point out, analyse and describe the choices made by the author of the source text and the translator of the target text (see also Matthiessen, 2014).
The hierarchy of stratification organises language in context into a hierarchy of strata (or levels). Along the hierarchy of stratification, the strata include context and language. Within language, there are the strata of the content plane (semantics and lexicogrammar) and the expression plane (phonology and phonetics in spoken language or graphology and graphetics in written language). These strata are related to each other by means of realisation.
The cline of instantiation links the instance (the text) with the potential (the system of language). On the one hand, the analysed texts are located at the instance pole of the cline, where it is possible to observe language unfolding as texts in their contexts of situation. On the other hand, translators move up and down the scale along the cline, making generalisations about the language based on the instances they have observed, and looking for instances that meet the requirements of the source text to be translated. Between the two poles, there are intermediate patterns characterised either as text types or as subpotentials (i.e. registers).
Metafunction is organised as a spectrum of different modes of meaning. The ideational meaning provides resources for construing our experience of the world both around us and inside us as meaning. The two modes of construing experience include the experiential mode that models experience configurationally and the logical mode that models experience serially. The interpersonal metafunction provides resources for interacting with people and introducing our judgments, desires and perspectives on the situation. The textual metafunction provides resources for organising ideational and interpersonal meanings as a flow of information in the context. Rank scale is a hierarchy of units based on composition. In English, the rank scale within lexicogrammar is clause-group/phrase-word-morpheme. The relationship between the units in the rank scale is realisation: one rank is composed of and realised by units of the rank immediately below it.
The hierarchy of axis distinguishes between the paradigmatic (systemic) organisation and the syntagmatic (structural) organisation, with the paradigmatic axis being realised by the syntagmatic axis. The two axes define the space in which the text unfolds. The paradigmatic axis defines the 'translation potential' , as it involves relations with the choices that are not present, but are lurking behind the text.
Another advantage of choosing SFL in this study is that the systemic descriptions of a growing range of languages, including English and Chinese, have been produced in recent years. These descriptions of language are designed to be comprehensive and can be applied to different areas, such as discourse analysis and translation.

2.3
Translation as Recreation of Meaning through Choice Various contributions have applied SFL to translation (see Steiner, 2019;Wang and Ma, 2021). Halliday (1956) investigates the computational modelling of translation based on an early version of SFL. Also, Catford (1965) applies this early framework and studies translation equivalence and shift along the dimensions of stratification and rank.
By adopting the semiotic interpretation of language, Matthiessen (2014Matthiessen ( , 2021 conceptualises translation as recreation of meaning in context through choice. The choices are made both in terms of interpreting the source text and recreating the target text, and are selected from among the options within the systems (i.e. meaning potentials) of the source language and target language. By considering the metafunctional organisation of language, translators are faced with choices located in the experiential, logical, interpersonal and textual systems.
In terms of logical meaning, translators choose how to interpret logicosemantic relations used in forming "coherent" source texts, and they choose among the options in the target language to reconstrue them in the translation they are producing.
In terms of experiential meaning, translators choose how to interpret events as configurations of elements (processes, participants and circumstances) and larger "chunks" of experience made up of events such as episodes and procedures, and they choose among the options in the target language to reconstrue the experiential meanings in the translation they are producing.
In terms of interpersonal meaning, translators choose how to interpret propositions, proposals and the assessments associated with them in the exchange of meaning embodied in the source text, and they choose among the options in the target language to re-enact the interpersonal meanings in the translation they are producing.
In terms of textual meaning, translators choose how to interpret messages and the sequences of messages that create the flow of information in the source text, and they choose among the options in the target language to re-present the textual meanings in the translation they are producing.
A number of studies in the literature have considered these four modes of meaning, such as those from the textual perspective (e.g. Ma, 2018, 2020), the interpersonal perspective (e.g. Munday, 2018), the experiential perspective (e.g. Mason, 2012) and the logical perspective (e.g. Li and Wu, 2017;Ma, 2018, 2020). In this study, we consider all the four modes of meaning.

A Linguistic Model for the Description and Quality Assessment of Poetry Translation
When opting for a linguistic theory for our model, we have selected SFL instead of combining an eclectic mix of different linguistic theories or approaches. Our model analyses the meaning and form of poetry, which are both crucial in literary creation and poetry translation (Bassnett, 1980). Furthermore, we incorporate context in the model to explain the similarities and differences between the source text and the translations in our analysis. Drawing on the hierarchy of stratification in SFL, we include four levels of analysis in the model, namely graphology, phonology, lexicogrammar and context (see Figure 1). In terms of graphology, we examine the layout of the poems. In terms of phonology, we investigate the rhyme scheme. Lexicogrammatically, the four modes of meaning are taken into consideration, with the main systems of the clause being selected, including theme in the textual analysis, mood and modality in the interpersonal analysis, transitivity in the experiential analysis as well as taxis and logico-semantic type in the logical analysis. Taken together, these comprehensive systemic analyses will offer a revealing account of how the source text and target text are organised to effectively function in their contexts of situation and contexts of culture. After the analysis at the above-mentioned strata, we examine context in terms of three contextual parameters, i.e. field, tenor and mode. The contextual investigation not only provides information on the cultural backgrounds of the ST and the TT, but also helps to identify the norms that are followed when translating poetry from English to Chinese.
The model enables the realisation of two possible goals of text analysis suggested by Halliday (2001: 13). The first goal is "to explain why the text means what it does: why it is understood the way it is -by the analyst, or by anyone else." The second goal, which includes the first goal but is more difficult to attain, is "to explain why the text is valued as it is … by anyone who may be evaluating it."

4
Working with the Model: Using Examples Selected from Rabindranath Tagore's Stray Birds and Its Three Chinese Translations

Data
To test the applicability of the model, we have selected the first three poems from Rabindranath Tagore's Stray Birds (source text, ST) and its three Chinese translations (target text, TT) as the data in this study. We have chosen Stray Birds as our data not only because the poems are highly valued in literary circles, but also because of the huge contrasts reflected in the TTs, including (i) the rhymed and unrhymed phonological choices, (ii) the form of modern and ancient poetry and (iii) the use of modern and ancient Chinese in poetry.
We have selected only a small number of poems in this study for the following reasons. Firstly, a large number of systems are involved in the analysis; to ensure a comprehensive manual analysis, the data involved have to be selective. Secondly, in keeping with House's (2001) attempt to approach translation criticism by drawing on linguistic analysis and social judgement, the analysis in this paper is largely illustrative. Thirdly, this paper serves to introduce an analytical framework for assessing poetry translation from the perspective of SFL, while the exploration of a larger data set can be found in another study conducted by the authors .
The ST is a collection of 325 short poems that are characterised by Indian culture and philosophy. Due to Tagore's unique way of using language, the way in which his innermost thoughts are embedded in the poems and the interrelationship between the poems, the ST has been regarded as difficult or even untranslatable by translators (Islam, 1995).
Target text 1 (TT1) was translated by Zheng Zhenduo in 1922. As a famous Chinese writer and poet, Zheng was one of the first people to introduce Tagore's poems to Chinese readers. Zheng's translation offers an attempt at pursuing ideological emancipation and is regarded as a classic.
Published in 1931, TT2 was translated by Yao Hua. He selected 257 poems from the total of 325, translating them into a form of classical Chinese five-character poem by using Zheng's translation (TT1), as Yao did not speak English. TT2 is regarded as a simple, sensible and aesthetic translation in the Chinese context. Translated by Feng Tang, a famous Chinese writer and poet, TT3 was published in mainland China in 2015. However, due to the translator's highly individualised form of translation, TT3 has been criticised for being disrespectful to Tagore and was eventually removed from the shelves by the publishing house. Feng (2019) has categorised the criticism that his translation received into three types, including (i) distortion of Tagore's original meaning, (ii) contamination of Tagore's purity and (iii) hype using Tagore's name.
Of these three translations, TT1 and TT3 are translated into modern Mandarin Chinese. TT2 is in classical Chinese in the form of classical poetry. In the analysis, linguistic evidence is provided for the literary criticism or comments, instead of subjectively commenting on the loyalty or faithfulness of the TTs to the ST. To attain these objectives, two steps will be taken: (i) studying the nature of the ST and translations of the three TTs by analysing them linguistically and attempting to reveal the trade-offs in the TTs and (ii) raising the issue of whether or not the TTs break the 'norms' and/or are good or bad translations (cf. Toury, 1995).

4.2
Graphological Analysis Graphologically, the layout of the ST follows a style of prose according to which the whole poem is arranged into one paragraph, with no line breaks being used. In TT1, the layout is similar to that of the ST, and the translated poem is in a prose style without line breaks. In terms of the use of punctuation marks, we note that in the ST and TT1, punctuation marks, notably commas and full stops, are commonly used following the English conventions. Additional commas are observed in Poems 1 and 3 in TT1, informing readers where to pause while reading the poem, such as '夏天的飞鸟, 飞到我窗前唱歌' (Flying birds of summer, fly to my window to sing).
In TT2, there is a fixed graphological pattern in all the poems, which are composed of four to eight lines, with each line containing five Chinese characters. Punctuation marks are used after every five characters at the end of each line, with commas being applied in lines with odd numbers and full stops in lines with even numbers.
In TT3, a different pattern is adopted as the translator arranges the poem in lines and applies no punctuation marks such as commas or full stops. Furthermore, the translator sometimes divides the poems into two stanzas to contrast them.

4.3
Phonological Analysis The phonological analysis focuses on the choice of rhyme in the poems. In general, the ST and TT1 do not rhyme; whereas rhyming is present in TT2 and TT3.
The ST is free in form and is more prose-like. According to Tagore (2016), rhyme does not mean everything in a poem; it is the emotion in a poem that moves its readers. Following the dimension of stratification in SFL, we find that Tagore (2016) emphasises the choices made in the content plane of language rather than in the expression plane. Zheng (2004), the translator of TT1, believes that a translation should be very close to the original, and therefore he adopts a literal form of translation. Following his method, he transplants the whole artistic arrangement from English to Chinese. Thus, his translation strictly follows the form of the ST, with no attempt being made to rhyme.
In TT2, rhyme is found in classical Chinese style poems. The rhyme schemes are not fixed and vary in the three poems from 'abab' to 'abcb' and 'abcdabcd' . The rhyming words are found at the end of lines, including lines with both odd and even numbers.
By admitting that rhyme is the most powerful weapon in a poem, the translator of TT3 openly states that he has tried his best to rhyme (Feng, 2015) and, as a result, most poems in TT3 rhyme. In the three poems, rhyming words are located at the end of lines, following the rhyme schemes of 'abcaa' , 'aa' and 'aba' .

4.4
Lexicogrammatical Analysis To conduct a lexicogrammatical analysis, we first chunked the clause complexes (i.e. sentences) in the ST and TTs into clauses, because the main lexicogrammatical systems, such as mood, theme and transitivity,1 all operate at the clause rank -the highest rank scale in lexicogrammar.
An interpersonal analysis focuses on choices made in the systems of mood and modality. mood, as the primary interpersonal system of the clause, deals with the organisation of clause as a move in an exchange and is the realisation of the semantic system of speech function in lexicogrammar. The analysis of mood suggests that six declaratives, one imperative and two bound clauses are found in the ST; whereas an increased number of clauses are found in the TTs, particularly in TT2, including ten declaratives and one imperative in TT1, twenty declaratives and one interrogative in TT2 as well as thirteen declaratives and two imperatives in TT3. As illustrated in Example 1, the poem in the ST consists of one imperative, which is equivalently recreated in TT1. In TT2, the imperative mood in the ST is changed to five declaratives and one interrogative, leading to several mood shifts. In TT3, the translator chooses to add one imperative mood, which is translated from a prepositional phrase in the ST -'在 我的 文字 里' (CV: at my word in modality is one of the main systems modelled on the interpersonal metafunction. It includes expressions of indeterminacy between positive and negative, constructs the semantic region of uncertainty and helps to express the speaker's judgement or request of the judgement. There are four main types of modality: probability (the likelihood of a situation to be true), usuality (how frequently a situation is true), obligation (how confident the interactant is to carry out the command) and readiness (how willing the interactant is to fulfil the offer). No instance of modality is found in the ST and TT1, while one instance is seen in both TT2 and TT3. In Example 2, '能' (can), which reflects the ability of the song in the poem, is added in TT2. In Example 3, '一直' (always) is added by the translator of TT3 to indicate the frequency with which the yellow leaves stay in front of the window.
Example 2 (Adapted from Tagore, 1931: 1) ST: It becomes small as one song (modality: -) TT2: 短 歌 能 入神。(modality: readiness: ability) short song can enthrall Example 3 (Adapted from Tagore, 2015: 1) ST: And yellow leaves of autumn, <<which have no songs>>, flutter (modality: -) TT3: 秋天 的 黄 叶 一直 在 窗 前 (modality: usuality) autumn SUB yellow leaves always CV: in window front A textual analysis examines the choices made in the system of theme, which reveals the point of departure of the message and serves to locate and orient the clause. Through the choice of Theme, one part of the clause becomes more prominent and the poet and translators can guide the readers in processing the message. There are three types of Theme, viz. textual, interpersonal and topical. We find four textual Themes in the ST, which are realised by three instances of the conjunction 'and' as well as a continuative 'o' (see Example 1). In the TTs, however, all the textual Themes are regarded as being redundant and are omitted due to the typological differences between the two languages, as it is a general feature for Chinese to use fewer conjunctions (see also Wang and Ma, 2020). As illustrated in Example 4, no translator has rendered the conjunction 'and' in the ST, leading to omissions of the textual Theme in the TTs.
Regarding the topical Theme in the ST, the addition of a mood type has led to the addition of various topical Themes, most of which are realised by the participants. In TT1 and TT3, a small number of topical Themes (i.e. two and six, respectively) are added by the translators. In TT2, thirteen topical Themes are added, including one realised by a circumstance -'临 流' (beside stream), which has no equivalent in the ST (see Example 1).
An experiential analysis examines the choices made in the system of process type, which construes human experiences into six types, viz. material, mental, relational, behavioural, verbal and existential. In the ST, we find six material and two relational processes as well as one verbal process. TT1 is mostly equivalent to the ST, except for the addition of one relational process and one verbal process due to the additions of clauses. However, in TT2, more changes are made, and the number of material and relational processes are increased to twelve and six, respectively (see Example 5). As shown in Example 5, various process types are added in TT2, based on the translator's recreation of the text.
In TT3, the frequencies of material and relational processes have increased from six and two in the ST to seven and six, respectively. Example 1 has already illustrated how an additional material process is used in TT3. We also find an additional behavioural process '笑' (laugh) that describes the behaviour of the stray birds in TT3, which has no equivalent in the ST.
The logical metafunction construes our experiences serially as chains of phenomena related by logico-semantic relationships. Through a logical analysis, we can highlight the organisation of the clauses in the systems of taxis and logico-semantic type. Of these, taxis describes the relationship between grammatical units according to their interdependency and logico-semantic type differentiates between the different types of relationships. Due to the increase in the number of clauses, we find additional logico-semantic types in the TTs, particularly in TT2 and TT3. As shown in Example 6 (see Table 1), five enhancing relations (×) are added in TT3. In this way, by adding clauses that are not found in the ST, the translator of TT3 creates a contrast between the two stanzas, with one focusing on '夏日 的 飞 鸟' (summer SUB flying bird) and the other on '秋天 的 黄 叶' (autumn SUB yellow leaf). The other translators, however, have not made such an explicit contrast. In Figure 2, we visualise the organisation of the text structure and highlight the contrast in TT3 (see also Chik and Taboada, 2020; in press for discussions on the applications of logical relations and Rhetorical Structure Theory).

4.5
Contextual Analysis According to the systemic functional theory of language, context is a higherorder semiotic system than the linguistic system, covering the spectrum of field, tenor and mode (Matthiessen et al., 2010). It is also noted that context is modelled along the cline of instantiation from the potential pole (i.e. context of culture) to the instance pole (i.e. context of situation) via the subpotential/ instance type (i.e. institution/situation type) (see also Section 2.2). In this section, we describe Stray Birds and its three translations from the perspective of the contextual parameters of field, tenor and mode. These are highly generalised concepts that are used to describe how context determines meaning. Halliday and Hasan (1976: 22) have explained the three semiotic systems of field, tenor and mode in the following way (see also Halliday & Hasan, 1985): field is the total event, in which the text is functioning, together with the purposive activity of the speaker or writer; it thus includes the subject-matter as one element in it. tenor refers to the type of role interaction, the set of relevant social relations, permanent and temporary, among the participants involved. mode is the function of the text in the event, including therefore both the channel taken by the language -spoken or written, extempore or prepared -and its genre, or rhetorical mode, such as narrative, didactic, persuasive, phatic communion and so on.
In Table 2, we tabulate our contextual analysis of the ST and TTs. The analysis of field focuses on the field of activity, i.e. what is going on in the text. Following Matthiessen's (2015) typology of field of activity, Poems 1 and 3 in the ST are categorised as reporting fields because they describe what is happening in nature, while Poem 2 is of enabling field because it commands the troupe of vagrants to take certain actions. In TT1 and TT3, the fields are equally recreated; in TT2, the field in Poem 2 is changed from enabling to reporting.

Description and Quality Assessment of Poetry Translation
Contrastive PragmaticS 3 (2022) 89-111 Table 2 Contextual analysis of the ST and TTs

Analysis of TT3
Field Field of activity In terms of tenor, Tagore and the three translators all adopt the institutional role of a poet. However, the three translators have each used different forms of translation. The translator of TT1, Zheng (2004), applies a literal (word-forword) form of translation. As his method is congruent with the Chinese norms of poetry translation, in which readers expect a 'faithful' translation of poetry, TT1 has thus been regarded as a Chinese classic. By taking initial norms into consideration, we find that Zheng is subject to the source norms and has produced an adequate translation, which is largely equivalent to the ST both in content and form.
Although the translator of TT2, Yao Hua, could not read Tagore's original English text, he manages to translate by adapting Zheng's translation, and he creatively maps his translation to the classical form of Chinese poetry. To carry out this adaptation, the translator makes various changes to TT2, leading to a large number of translation shifts in lexicogrammar. Despite the changes, readers who are fond of classical poetry might think that the translator has infused beauty into the translation. In contrast to TT1, TT2 is subject to target norms, according to which adequacy has to give way to acceptability in the target culture.
The translator of TT3, Feng Tang (2015), is a contemporary Chinese poet and writer, who openly states that he has the freedom to balance 'fidelity, fluency, and elegance' , which have been regarded as the three principles of translation and the parameters for discussions on translation in China since the 1900s. Compared to the translators of TT1 and TT2, who maintain a distant relationship with their readers, Feng attempts to shorten the distance between himself and his readers by using amorous lexical choices, such as '解开 裤裆' (unlock crotch) and '舌吻' (French kiss) in Poem 3, '做 爱' (make love) in Poem 13 and '肉欲' (lust) in Poem 43. We believe that Chinese readers oppose Feng's idiosyncratic choice of words with obscene associations, which are unlikely to be found in Tagore's original. Therefore, most of the objections to Feng's translation are associated with his lexical choices, which change the tenor of TT3, rather than with his grammatical choices.
The modes of the ST and TTs are similar in that the poems are all written to be read by readers. As the translators of TT2 and TT3 have made deliberate choices of rhyme and layout pattern, TT2 and TT3 are more suitable for reading aloud.

Conclusion
This article has sought to explore the possibility of building a model, based on SFL, for describing and assessing the quality of poetry translation. Grounded in the literature on poetry translation, the proposed model took both the form and meaning of poetry into consideration and involved linguistic analyses at the levels (strata in SFL) of graphology, phonology, lexicogrammar and context. To illustrate how the model worked, we offered an analysis of Tagore's Stray Birds in English and its three Chinese translations, pointing out the choices made by Tagore and the translators at different levels. In addition, on the basis of a contextual analysis, we explained, on the one hand, why TT3 is not welcomed by some Chinese readers, but, on the other hand, why TT1 and TT2 are enjoyed by readers despite the translation shifts in a number of respects.
In terms of the effectiveness of the constructed model, we found that the graphological and phonological analyses took features such as the rhyme scheme, punctuation marks and line breaks into consideration. Furthermore, such choices in graphology and phonology influenced the translators' lexicogrammatical choices. For instance, in order to rhyme, the translator of TT3 added some clauses and changed certain process types; in order to map the English prose style of poetry to the classical Chinese style of poetry, various translation shifts in lexicogrammar were found in TT2 (see also Ma and Wang, 2021).
Having proved that linguistic theories offer a powerful tool for analysing poetry translation, we argue that our model can be suitably applied to the description, comparison and evaluation of translated poems and their source texts. With a large sample of data, it will be possible to apply our model to quantify the translation shifts found in translation and examine translation in terms of probability (Wang and Ma, 2020;Ma and Wang, 2021;cf. Toury, 2004). Furthermore, our findings serve as evidence of the applicability of SFL to poetry translation and offer new possibilities in translation studies from the perspective of SFL.