From Vision to Discourse: The Grammaticalization of the Perception Verb Thấy in Vietnamese (13–20th C.)

Abstract

This paper offers the first long-range account of the grammaticalization of the Vietnamese perception verb thấy ‘see’ from the 13th to the mid-20th century. Using a balanced diachronic corpus of ten representative texts (1345 tokens), we combine frequency profiling with constructional analysis to trace thấy’s shift from a literal visual predicate to a high-frequency resource for epistemic stance, evidentiality, evaluation, and discourse management. The results reveal a robust progression aligned with the sensory hierarchy and canonical event-schema pathways: early literal uses and multimodal bundling (13–14th c.) provide bridging contexts; the 15th century introduces raising (thấy + VP/Adj) and clausal complementation (thấy (rằng/là) + CP); the 16–17th centuries expand resultative perception complexes (e.g., xem/chiêm bao/nghe + thấy) and reportative frames; the 18th century brings evaluative and speaker-anchored uses (chúng tôi thấy); the 19–20th centuries stabilize discourse-pivot (thấy… thì…), epistemic (thấy cần phải…), and exclamative/affective (thấy ghét) readings. We argue that Vietnamese clause-linking options and optional complementizers facilitate constructionalization via loose complementation and subjectification, while retaining perceptual residues that motivate evidential and interactional meanings. The study contributes: (i) a comprehensive diachrony of thấy; (ii) diagnostics separating perceptual, experiential, propositional, and discourse layers; and (iii) a case study bearing on the relationship between grammaticalization and constructional change in an isolating language.

1. Introduction

Diachronic developments of perception verbs offer a privileged window on mechanisms of semantic change, especially the shift from concrete sensory meanings to markers of knowledge, stance, and discourse. Vietnamese thấy ‘see’, originally denoting visual perception, undergoes precisely such a trajectory over seven–eight centuries, expanding into evidential, epistemic, evaluative, and interactional functions in ways that are both typologically expected and specific for a Southeast Asian isolating language. Our study contributes the first long-range, corpus-based account of thấy across premodern and modern Vietnamese, situating it within cross-linguistic pathways known from grammaticalization research and lexical typology of perception (Viberg, 1984; Evans & Wilkins, 2000; Viberg, 2001; Norcliffe & Majid, 2024). We adopt a usage-based, constructional perspective (Hopper, 1987; Traugott, 2014; Traugott & Trousdale, 2013), tracking how recurrent form–meaning pairings with thấy proliferate and stabilize, yielding emergent constructions (e.g., thấy + N/NP, thấy + Adj/VP, thấy + CP) that encode evidential stance, evaluation, mirativity, and discourse management. The Vietnamese evidence allows us to test how general cognitive pressures (invited inferencing, subjectification; Traugott & Dasher, 2002; Traugott, 1989) interact with morphosyntactic characteristics of languages in Mainland Southeast Asia (analytic clause structure, loose subordination) in driving change.

2. Theoretical Background

This study situates Vietnamese thấy ‘see’ in three intersecting lines of inquiry: (i) the sensory hierarchy and its role in semantic extension, (ii) grammaticalization pathways modeled as conceptual event schemas and realized via bridging contexts, and (iii) constructionalization, i.e., how new form–meaning pairings emerge and stabilize in usage. Together, these perspectives offer a coherent framework for explaining how a concrete perception verb can expand into epistemic, evidential, and discourse domains, and they yield operational diagnostics that we apply to the diachronic Vietnamese record.

Cross-linguistically, sensory verbs follow a robust implicational scale for semantic extension—vision > hearing > touch > taste/smell—reflecting informational richness, experiential distance, and cognitive privilege of the higher senses (Viberg, 1984, 2001; Heine et al., 1991).1 Because vision sits at the top, ‘see’ verbs are cross-linguistically the most frequent sources for meanings of knowledge, belief, and inference, then for evidential and discourse functions (Sweetser, 1990; Evans & Wilkins, 2000). This predicts that Vietnamese thấy should be an early and productive locus of semantic expansion relative to other perceptual lexemes (nghe ‘hear’, ngửi ‘smell’, nếm ‘taste’, sờ/chạm ‘touch’). We operationalize the hierarchy by tracking proportional growth of non-visual readings across centuries and by testing whether lower-sense collocations (e.g., thấy mùi ‘perceive smell’) appear later and remain narrower—an outcome the hierarchy leads us to expect.

In grammaticalization, recurrent pathways are not arbitrary; they instantiate conceptual re-analyses encoded as event schemas (Heine et al., 1991; Heine, 1993; Heine & Kuteva, 2002, 2007). For perception verbs, a widely attested cline is:

(1)

PERCEIVE

→

KNOW/THINK

→

SAY/CLAIM

→

DISCOURSE

Change proceeds through bridging contexts that support both the source and the emergent target interpretation (Heine, 2002; Traugott & Dasher, 2002). In Vietnamese, such bridges include: (a) propositional complementation (thấy (rằng/là) P), inviting ‘realize/come to know’; (b) reportative frames in narrative where thấy scopes over quoted or inferred content; and (c) evaluative collocations (e.g., dường thấy ‘it seems’), which license epistemic/mirative readings while retaining traces of perceptual semantics. The persistence of perceptual residue in later meanings is a core prediction of grammaticalization theory (Hopper & Traugott, 2003): thấy-based epistemic/evidential uses should continue to index a vantage point grounded in experience or observation, even as literal seeing recedes.

As items move from objective perception to epistemic judgment and interactional stance, they tend to become increasingly speaker-anchored (Traugott, 1989; Traugott & Heine, 1991). We therefore expect thấy to develop functions that encode information source and commitment—i.e., evidentiality in Aikhenvald’s (2004) sense—alongside mirativity (surprise/new information; see also DeLancey, 1997; Mélac, 2023). Diagnostics include person-orientation (tôi thấy ‘I see’…), compatibility with stance adverbs (hình như, dường như ‘it seems...’), and uses where thấy evaluates a proposition or an interlocutor’s move (e.g., discourse-medial thấy… thì…). These speaker-oriented functions should cluster in specific constructions and spread over time.

We treat the evolution of thấy as constructionalization (Traugott & Trousdale, 2013; Traugott, 2014): the emergence of new nodes in a network (e.g., [thấy]V[perception] → [thấy]V[experiential/psych] → [thấy]V[evidential/stance] → [thấy]DiscPM), accompanied by host class expansion, decategorialization, and semantic bleaching (Lehmann, 2015; Hopper & Traugott, 2003). Functional typology predicts systematic relations between the hierarchy of clause types and complementation with perception verbs (Dik & Hengeveld, 1991; Noonan, 2007; Palmer, 2001; Nuyts, 2001)2. We therefore segment Vietnamese uses into four layers that map onto increasing propositionality:

(i)

Direct perception (thấy + NP),

(ii)

Factive/experiential (thấy + VP/Adj ‘experience/feel’),

(iii)

Propositional (thấy (rằng/là) + CP ‘realize/think that’), and

(iv)

Discourse-pragmatic (thấy as stance/discourse operator).

We exploit these layers as diachronic diagnostics to anchor the directionality of change.

Several language-internal properties of Vietnamese plausibly accelerate thấy’s development. First, an analytic profile with abundant serial verb and resultative complexes (e.g., xem/nhìn/coi/chiêm bao + thấy) creates recurrent perception + outcome templates that scaffold epistemic readings. Second, the stance-rich discourse of religious prose, epistolary genres, and later realist fiction creates pragmatic niches for evidential/mirative and discourse-organizing uses. Third, historical bilingualism with Literary Sinitic offers potential calquing of complement patterns (e.g., introduction of propositional content) while still allowing endogenous Vietnamese trajectories to unfold; our account keeps contact effects as hypotheses and relies on distributional evidence for internal reanalysis.

The framework yields concrete, testable expectations we bring to the corpus: (a) non-visual and speaker-oriented uses should rise monotonically over time; (b) propositional complements should expand in frequency and host range; (c) fixed frames (e.g., thấy (rằng/là) ‘see that…’; thấy… thì…’see… then…’.) should crystallize and generalize; and (d) collocations reflecting the sensory hierarchy (e.g., nghe thấy ‘lit: hear see’, thấy mùi ‘lit: see smell’) should emerge later and remain more restricted. The ensuing sections apply these diagnostics to seven centuries of Vietnamese data to evaluate whether thấy follows the cross-linguistic pathway PERCEIVE → KNOW/THINK → SAY/CLAIM → DISCOURSE, and how Vietnamese-specific constructions mediate that trajectory.

We therefore address three core research questions:

(i)

Semantic change: How has thấy’s meaning shifted from concrete visual perception to epistemic, evidential, evaluative, and discourse-structuring uses over seven centuries?

(ii)

Syntactic reanalysis: What new constructions and patterns emerge as thấy grammaticalizes, and how do these reflect its evolving functions?

(iii)

Cross-linguistic alignment: To what extent do Vietnamese developments mirror the perception-to-grammatical markers pathways documented in other languages?

The rest of the paper is organized as follows. Section 3 details the diachronic corpus (10 texts, 13–20th c.), the annotation scheme, and operational diagnostics for four functional layers: direct perception, experiential/factive, propositional, and discourse-pragmatic. Section 3 also provides an aggregate frequency profile of thấy across periods and genres. Section 4 is the empirical core, presenting century-by-century analyses with glossed examples: (i) early literal and multimodal bundling (13–14th c.); (ii) emergent raising and clausal complementation (15th c.); (iii) resultative perception complexes and reportative frames (16–17th c.); (iv) evaluative and speaker-anchored uses (18th c.); (v) narrative pivoting and first-person stance (19th c.); and (vi) stabilization of epistemic, affective, and discourse-marker functions (20th c.). Section 5 discusses implications for grammaticalization vs. Diachronic Construction Grammar, showing how Vietnamese data support a PERCEIVE → KNOW/THINK → SAY/DISCOURSE trajectory while retaining source-meaning residues. Section 6 concludes with a summary of contributions and avenues for future testing and comparative work.

3. Methodology and Data

3.1. Corpus Design and Text Coverage

To document the long-range grammaticalization of thấy ‘see’, we built a diachronic corpus that spans seven centuries, multiple scripts (Nôm, Literary Sinitic, early and modern Quốc ngữ)3, and a wide spectrum of genres from Buddhist didactic prose to modern realist fiction. Each component text was selected because it (i) is philologically reliable, (ii) represents a register in which perception and stance expressions are likely, and (iii) anchors a step in the transition from bilingual Sinitic–Vietnamese discourse to fully vernacular prose. Below we give the Vietnamese title with Chinese characters (where relevant), an English gloss, period and script, followed by a brief rationale for inclusion.

i.

Phật thuyết đại báo phụ mẫu ân trọng kinh (佛說大報父母恩重經)—The Sutra on the Buddha’s Teaching of the Profound Gratitude Owed to Parents (13th c.; bilingual Nôm/Literary Sinitic).

Chosen as one of the earliest vernacularized Buddhist texts with interlinear or paraphrastic Nôm, it offers a baseline for literal visual uses of thấy embedded in Sinitic exegetical frames.

ii.

Thiền tông Khóa hư ngữ lục (禪宗課虛語錄)—Recorded Sayings of the Zen School: The Book of Emptiness (14th c.; bilingual Nôm/Literary Sinitic).

A canonical Zen compilation with abundant didactic exempla and commentarial prose, ideal for tracing early inferential and introspective readings associated with meditative “seeing/realizing.”

iii.

Quốc âm thi tập (國音詩集)—Collection of Poems in the National Language (15th c.; Nôm).

As the first large Nôm poetry collection by Nguyễn Trãi, it represents high literary vernacular style; poetic syntax and imagery allow us to observe metaphorical extensions of visual perception.

iv.

Tân biên Truyền kỳ mạn lục tăng bổ giải âm tập chú (新編傳奇漫錄増補解音集註)—Newly Edited Tales of the Marvelous, with Added Explanations and Phonetic Annotations (16–17th c.; bilingual Nôm/Literary Sinitic).

This prose narrative with scholastic glossing bridges learned and popular registers; frequent narrator stance and reported perception provide fertile contexts for evidential drift.

v.

Phép giảng tám ngày (1651)—The Catechism in Eight Days (early Quốc ngữ with Latin paratext).

An early missionary catechism where thấy occurs in doctrinal exposition and exempla; its controlled prose and metalinguistic glosses expose emerging non-literal (experiential, epistemic) uses in standardized romanized Vietnamese.

vi.

Thiên Nam ngữ lục (天南語錄)—The Versified Chronicles of the South (late 17th c.; Nôm).

A long narrative poem blending history and legend; dialogic narration and evaluative commentary make it a key witness to clause-level thấy in reportive and mirative readings.

vii.

34 thư từ Công giáo thế kỷ 18—Eighteenth-century Catholic Correspondences (18th c.; early Quốc ngữ).

Private and administrative letters in emerging romanized prose capture colloquial stance, information source marking, and formulaic discourse frames where thấy is frequent.

viii.

Sách sổ sang chép các việc—Records of Various Matters (Philippe Bỉnh, 1822/1968; early Quốc ngữ).

Autobiographical and documentary prose with travel and ethnographic descriptions; rich in first-person observation → knowledge transitions crucial for evidential mapping.

ix.

Thầy Lazaro Phiền—Lazaro Phiền (Nguyễn, 1887; modern Quốc ngữ).

Among the earliest modern novels; prose narration in everyday style enables close analysis of psychological “feel/find” uses and clause-combining patterns with thấy.

x.

Sống mòn—Worn-Out Life (Nam Cao, 1944/1956; modern Quốc ngữ).

High-frequency, colloquial modern fiction; dense first-person reflection and dialogue supply the clearest evidence of fully grammaticalized epistemic, affective, and discourse-organizational thấy.

The ten texts used in this study were accessed through a combination of (i) published critical editions and scholarly transcriptions (for Nôm and bilingual Sinitic–Nôm texts) and (ii) printed editions of Quốc ngữ materials (missionary texts, letters, and modern prose). Since these materials differ in script and editorial conventions, there is currently no single open online corpus that provides all these texts in a unified, searchable format. To support replication, we cite examples with stable page/folio identifiers from the editions used. Together, these ten sources provide a temporally continuous, genre-diverse record of Vietnamese from early Nôm/Sinitic bilinguality to modern Quốc ngữ prose, allowing us to observe how thấy moves from concrete visual perception to an array of evidential, epistemic, mirative, evaluative, and structuring functions across styles and centuries.

3.2. Extraction and Annotation

All occurrences of thấy were extracted from each text. Each token was annotated for (a) date and text; (b) register/genre; (c) syntactic environment (intransitive; thấy + NP; thấy + VP; thấy + CP/[rằng/là]; impersonal/expletive uses); and (d) semantic–pragmatic value, using a usage-based coding scheme grounded in crosslinguistic diagnostics for perception-verb change (Viberg, 1984; Evans & Wilkins, 2000; Viberg, 2001; Norcliffe & Majid, 2024). The functional inventory distinguishes: literal visual perception; epistemic/experiential (‘come to know’, ‘find, think’), evidential (reportative/observational source marking), mirative (unexpected discovery), evaluative/intensifying (speaker-stance uses), and discourse-structuring.

In addition to token-level labels, we recorded constructional information needed to trace reanalysis: (i) complement choice (NP vs. VP vs. CP, and complementizer type rằng/là/Ø), (ii) subjecthood and control (e.g., raising-like and impersonal) (iii) co-occurrence with adverbials (e.g., degree and polarity items that cue stance/mirativity), and (iv) clause linkage (thấy… thì…). This enables us to model constructionalization: the emergence and stabilization of new form–meaning pairings in which thấy serves increasingly abstract epistemic and discourse functions.

3.3. Instances of thấy by Text and Initial Frequency Profile

Across the ten texts, we extracted 1345 tokens of thấy. In raw counts, the largest contributions come from Sống mòn (327 tokens; 24.31% of all tokens), Thiên Nam ngữ lục (270; 20.07%), Truyền kỳ mạn lục (187; 13.90%), and Phép giảng tám ngày (179; 13.31%), while the earliest Buddhist materials contribute comparatively fewer tokens.

Table 1. Corpus of texts analyzed for thấy.

No.	Text	English Translation	Historical Period	Language/Script	No of thấy	No of Syllables
1	Phật thuyết đại báo phụ mẫu ân trọng kinh (佛說大報父母恩重經)	The Sutra on the Buddha’s Teaching of the Profound Gratitude Owed to Parents	13th century	bilingual (Literary Sinitic/Vietnamese in Nôm)	15	5423
2	Thiền tông Khóa hư ngữ lục (禪宗課虛語錄)	Recorded Sayings of the Zen School: The Book of Emptiness	14th century	bilingual (Literary Sinitic/Vietnamese in Nôm)	33	12,576
3	Quốc âm thi tập (國音詩集)—Nguyễn Trãi	Collection of Poems in the National Language	15th century	monolingual (Vietnamese in Nôm)	42	12,852
4	Tân biên Truyền kỳ mạn lục tăng bổ giải âm tập chú (新編傳奇漫錄増補解音集註)	Newly Edited Tales of the Marvelous, with Added Explanations and Phonetic Annotations	16–17th century	bilingual (Literary Sinitic/Vietnamese in Nôm)	187	44,678
5	Phép giảng tám ngày (1651)	The Catechism in Eight Days	17th century	bilingual (Latin/Vietnamese in Early Romanized Quốc ngữ)	179	56,812
6	Thiên Nam ngữ lục (天南語錄)	The Versified Chronicles of the South	late 17th century	monolingual (Vietnamese in Nôm)	270	28,205
7	34 Thư từ công giáo thế kỷ 18 (34 18th-century-Catholic correspondences)	Letters and Correspondence of Catholic Clergy	18th century	monolingual (Vietnamese in early Romanized Quốc ngữ)	75	52,820
8	Sách sổ sang chép các việc—Philippe Bỉnh	Records of Various Matters	1822	monolingual (Vietnamese in early Romanized Quốc ngữ)	160	62,940
9	Thầy Lazaro Phiền—Nguyễn Trọng Quản	Lazaro Phiền	1887	monolingual (Vietnamese in modern Romanized Quốc ngữ)	57	9596
10	Sống mòn—Nam Cao	Worn-Out Life	1944	monolingual (Vietnamese in modern Romanized Quốc ngữ)	327	83,825
SUM					1345	369,727

reports, for transparency, both the number of thấy tokens and an approximate size denominator for each source.4

Three macroscopic tendencies are visible. (i) The earliest Nôm–Hán Buddhist materials (13–15th c.) show relatively low token density, consistent with doctrinal/verse styles that favor Sinitic calques and fewer explicit perception predicates. (ii) From the late 16th through the 17th century there is a pronounced rise (Truyền kỳ: 187; Phép giảng: 179; Thiên Nam ngữ lục: 270), precisely in narrative, catechetical, and verse-chronicle genres that encourage both literal perception and incipient epistemic/evidential readings (e.g., thấy rằng/thấy là). (iii) The 20th-century realist novel (Sống mòn: 327, the largest single share) maintains high frequency but redeploys thấy in psychological, evaluative, and discourse-organizing functions, in line with the well-known drift from perception → inference/knowledge → stance/discourse (Viberg, 1984, 2001; Hopper & Traugott, 2003; Heine & Kuteva, 2002).

To help readers contextualize the raw token counts of thấy, we now provide approximate text-length denominators and rough normalized rates. Because Vietnamese is largely isolating, we use orthographic syllable/word tokens for Quốc ngữ texts and vernacular Nôm character counts for Nôm materials; for bilingual sources we count only the Vietnamese (Nôm) lines and exclude the Literary Sinitic (Hán) lines. The resulting denominators are therefore best understood as comparable within the same script type and as an approximation across the whole corpus, rather than as perfectly uniform “word counts” in a strict corpus-linguistic sense. With this caveat, the normalized density of thấy per 1000 tokens is: Phật thuyết 15/5423 ≈ 2.77; Khóa hư lục 33/12,576 ≈ 2.62; Quốc âm thi tập 42/12,852 ≈ 3.27; Truyền kỳ mạn lục 187/44,678 ≈ 4.19; Phép giảng tám ngày 179/56,812 ≈ 3.15; Thiên Nam ngữ lục 270/28,205 ≈ 9.57; 18th-century Catholic letters 76/52,820 ≈ 1.42; Sách sổ sang chép các việc 160/62,940 ≈ 2.54; Thầy Lazaro Phiền 57/9596 ≈ 5.94; and Sống mòn 327/83,825 ≈ 3.90. Across the full corpus (369,727 tokens), the overall density is ≈ 3.64 tokens of thấy per 1000. Although these figures should not be over-interpreted statistically (given differences in edition conventions, script, and genre), they corroborate the qualitative picture: the early Buddhist materials show lower densities, the narrative/didactic boom of the 16–17th centuries exhibits higher densities, and modern prose sustains a high density with a markedly broader functional profile of thấy (psychological, evaluative, and discourse-structuring uses).

Finally, the steep growth from the 16–17th centuries onward co-occurs with constructional broadening: rise of propositional complementation (thấy rằng/là P ‘see that P’), impersonal/raising-like uses with inanimate subjects (knowledge/evaluative readings), and stance-laden collocations (dường thấy, ắt thấy). By the 19–20th centuries, these resources feed speaker-anchored epistemic and discourse functions—frequent first-person frames (tôi thấy…) and discourse pivots (thấy… thì…). This pattern is consistent with typological expectations in two concrete ways. First, the earliest extensions build outward from vision, then recruit hearing, while lower senses remain marginal: we observe early multi-sensory bundling thấy-nghe ‘see and hear’, later productive nghe-thấy ‘hear and perceive/’hear and see’, and only much later—and sparsely—collocations such as thấy mùi ‘sense/notice a smell’, which accords with the hierarchy vision > hearing > touch > taste/smell (Viberg, 1984, 2001). Second, the direction of semantic change is anchored in identifiable bridging constructions: thấy begins as a concrete perception verb (thấy + NP, ‘see X’), then expands to propositional scope (thấy rằng/là + CP, ‘see/realize/think that P’), develops reportative and metatextual uses (thấy chữ rằng…, ‘see/read the wording that...’), and finally stabilizes as stance and discourse machinery (thấy… thì…, ‘when/if (one) sees…, then…’; exclamatives like thấy ghét! ‘(how) hateful!/so annoying!’; evaluative–modal frames like thấy cần phải…, ‘find/feel it necessary to…’). Taken together, these steps instantiate the cline PERCEIVE → KNOW/THINK → SAY/CLAIM/DISCOURSE proposed in event-schema accounts of grammaticalization (Heine et al., 1991; Traugott & Dasher, 2002), while also illustrating how Vietnamese realizes this trajectory primarily through constructional diversification—serial/resultative complexes and optional complementation—consistent with an isolating morphosyntax.

4. Diachrony of Thấy from 13th to 20th Century

This section presents a chronologically ordered profile of the Vietnamese perception verb thấy ‘see’ across ten periodized sources (13–20th c.).

4.1. Thấy in 13th Century

Phật thuyết đại báo phụ mẫu ân trọng kinh (佛說大報父母恩重經 ‘Sūtra on the Profound Gratitude Owed to Parents’) is one of the earliest Vietnamese works to interleave vernacular Nôm with Literary Sinitic. Scholars disagree on the precise dating of Phật thuyết đại báo phụ mẫu ân trọng kinh: while Hoàng (1999) and Shimizu (2002, 2020) propose a fifteenth-century origin, others, such as Nguyễn Q. H. (2001) and Trần (2011), suggest an earlier composition, possibly dating back to the twelfth century. For consistency in our diachronic analysis, however, the text is here treated as representing the earliest available stratum, corresponding to the thirteenth century. This position allows us to test our core theoretical questions at the baseline: What are the earliest constructional frames of thấy? Do any ‘bridging contexts’ (Heine et al., 1991; Heine, 2002; Heine & Kuteva, 2002, 2007; Kuteva et al., 2019) already foreshadow movement along the perception → cognition/stance cline? And do we see the sensory hierarchy (VISION > HEARING > TOUCH > TASTE/SMELL; Viberg, 1984, 2001) organizing multi-modal composites?

The text yields 15 tokens of thấy. The distribution is highly conservative: 14/15 tokens occur in a plain transitive S–V–O frame (thấy + N/NP), denoting concrete visual perception. Only 1/15 shows a serial, bi-modal pattern (thấy nghe mắng ‘see-hear’). No tokens take clausal complements (no thấy rằng/là), no epistemic/evidential readings, and no discourse-organizing uses. This profile typifies the first stage of the grammaticalization pathway: a core visual verb tightly bound to object NP complements.

Two properties matter for downstream change. First, thấy occasionally combines with highly abstract, doctrinal referents (e.g., bụt ‘Buddha’).

(2)

30a4

能

造

一

卷、

得

見

一

佛

Literary Sinitic

be.able make one scroll obtain see one

Buddha

nhược

hay

làm

nên

một

quyển kinh,

thì

Vietnamese

if

can

make become one CLF5 scripture TOP

được

thấy

một

bụt

get

see

one

Buddha

‘If one can complete a scripture, then one gets to see a Buddha.’

Syntactically thấy is transitive (thấy + NP); semantically, the highly abstract object (bụt) supports an incipient reality-assessment (‘come to realize/attain vision of X’) without abandoning the literal frame—precisely the kind of bridge that later licenses PERCEIVE → KNOW/JUDGE extensions (Heine, 1993; Traugott & Trousdale, 2013).

Second, the single serial token thấy nghe mắng is a calque-like echo of the Sinitic pair 見聞 ‘see–hear’, which aligns with the sensory hierarchy: the higher-rank visual sense precedes hearing in a compacted ‘evidence bundle.’

(3)

6a2

若

有

見

聞

者，

悉

發

菩 提

心

Literary Sinitic

if

have

see

hear

person

all

arouse bodhi mind

nhược có

người

nào

hoặc

thấy

nghe

mắng

Vietnamese

if

have person

any

or

see

hear

hear

cho

hết

đều

mở

lòng

bồ-đề

for

all

all

open mind bodhi

‘If there are any persons who see-hear (this), let all of them open the bodhi-mind.’

Functionally, this is still perceptual, but it already bundles evidence from two senses, a discourse configuration that later facilitates reportative/evidential readings (cf. Dik & Hengeveld, 1991 on complement hierarchies; Viberg, 1984, 2001). Such multi-modal packaging is an early discourse configuration from which reportative or experiential readings can later emerge (PERCEIVE → KNOW/THINK → SAY/CLAIM/DISCOURSE).

The earliest attested Vietnamese thấy is strictly visual and argument structurally simple (transitive with NP objects), yet it already offers two seeds for grammaticalization: (i) abstract objects inviting “realize/come to know” construals (bridge to epistemic/subjective readings; Traugott, 2014), and (ii) bi-modal serials (thấy nghe mắng) that mirror Sinitic 見聞 and respect the vision first hierarchy, foreshadowing the later drift from PERCEIVE to KNOW/THINK and ultimately to DISCOURSE/EVIDENTIAL functions (Heine & Kuteva, 2002, 2007; Kuteva et al., 2019; Traugott & Heine, 1991).6 At this stage, however, no clausal complements and no stance/evidential uses are attested—establishing a clean baseline for the expansions that surface from the 14th century onward.

4.2. Thấy in the 14th Century

The Khóa hư lục 禪宗課虛語錄 (‘Recorded Sayings of the Zen School’), a bilingual Literary Sinitic/Nôm compilation (Trần, 2012), shows both a sharp rise in frequency and a broader functional range of thấy.

We identified 33 tokens, distributed as follows: literal visual thấy + NP in plain transitive frames (24/33), adjectival complements signaling experiential/subjective readings (2/33), and postverbal/resultative V + thấy sequences (e.g., coi/xem/nghe + thấy) (7/33).

Qualitatively, two innovations are crucial for later change: (i) thấy begins to take adjectival predicates (‘feel/experience X’), and (ii) thấy occurs postverbally after another perception verb, where it behaves like a resultative/perception attainment marker. Both are classic bridging contexts on the path PERCEIVE → KNOW/THINK → SAY/CLAIM/DISCOURSE (Heine et al., 1991; Heine, 1993; Heine & Kuteva, 2002, 2007; Kuteva et al., 2019), and they align with the sensory hierarchy (VISION > HEARING > TOUCH > TASTE/SMELL; Viberg, 1984, 2001): vision remains dominant (coi/xem + thấy), while nghe thấy expands into hearing.

(4)

Thấy + adjective (experiential ‘feel/experience’)

bệnh tật

trầm trệ

ghê

tháng

mà

chưa

thấy

hèn,

illness

stagnate

many month PRT    not.yet

see

lessen

gối

giường

nằm

lâu

trải

ngày

mà

chưa

thấy

đỡ.

pillow bed

lie

long

span day

PRT

not.yet

see

ease

‘The illness has dragged on for many months yet I do not feel it lessen; lying in bed day after day, I have not yet felt any relief.’

Here, thấy selects adjectival predicates (hèn ‘less/abate’ [arch.], đỡ ‘better/eased’). The subject is an experiencer; thấy no longer denotes literal seeing but experiential assessment (‘feel/experience’). This is the first clear move toward subjectification (Traugott, 1989, 2014) and toward the epistemic/stance domain in our corpus.

(5)	Resultative V + thấy (vision)
	nghiệp tà dâm	ấy	lòng	đam	chưng	bề	thanh	sắc,
	karma lustful	DEM	heart	crave	because	toward	beauty	form
	con	mắt	nom thấy,	mắc	phải	thức	tốt,
	CLF	eye	look see	fall	get	thing beautiful
	chẳng	đoái	đến	thửa	lòng	ngay thực
	NEG	care	to	that	heart	honest
	That sinful karma of lust—because the heart craves beauty and form, the eyes look and see, becoming ensnared by pleasing things—pays no heed to the true and up right heart.’

Here nom thấy ‘look and see’ expresses direct visual perception within a moral–didactic frame. The co-occurrence of nom ‘look’ and thấy ‘see’ strengthens the sensory grounding (vision at the top of the sensory hierarchy, Viberg, 1984, 2001), but the clause also begins to profile the internal consequence of perception—how seeing leads to temptation and attachment.

(6)

Resultative V + thấy (hearing)

mắt

ngươi Ly Lâu

mà

xem

cái

sắc

tốt

chỉn

khôn

eye

PRN Ly Lau

PRT

look

CLF

form good thus not

sáng

mà

nghe

thấy tiếng

hát

hay

cũng

chẳng

biệt.

bright PRT hear

see voice

sing

good also

NEG

distinguish

‘Ly Lau’s eyes, when they look at fine forms cannot truly discern; likewise, even when they hear and (actually) perceive beautiful singing, they still cannot tell (it apart).’

The collocation nghe thấy (‘hear see’ ≈ ‘hear and (come to) perceive’) extends the resultative pattern from vision to hearing, matching the sensory hierarchy: vision provides the template (nom thấy), then hearing adopts the same construction (nghe thấy). This multisensory packaging strengthens the interpretation of thấy as a perception attainment marker, paving the way for evidential readings (‘it is attested/perceptible that…’) in later periods.

The 14th century evidence captures a transitional stage. Frequencies remain dominated by literal seeing, but the first incursions into (a) experiential/subjective predication (thấy + ADJ) and (b) resultative perception complexes (V + thấy) are now in place. Both constructions situate thấy at the semantics–syntax interface as either stance predicate or light/resultative element.

4.3. Thấy in the 15th Century

In Quốc âm thi tập (Nguyễn Trãi), we identify 42 tokens of thấy. This collection consolidates the profile already visible in the 14th century (literal visual readings; V + thấy resultatives; thấy + ADJ ‘experiential/subjective’ predications), but it also introduces two constructional innovations that are central to our theoretical agenda: (i) raising-like thấy + VP with non-agentive/inanimate subjects, and (ii) clausal complementation thấy rằng ‘see/realize that’, which provides a clear bridge from PERCEIVE → KNOW/THINK. A further recurrent pattern, ‘thấy… thì’, begins to code conditional/temporal linking, pushing thấy toward discourse function.

(7)	Raising-like thấy + VP with inanimate subject
	雪	霜	体	㐌	鄧	饒	𣈜
	tuyết	sương thấy		đã	đặng	nhiều	ngày
	snowfrost		see	PFV	get	many day
	‘Snow and frost appear to have lasted many days.’
	(219. 松 Tùng ‘The Pine Tree’ 2)

Here, the surface subject tuyết sương ‘snow, frost’ is not a perceiver. The reading is epistemic—’it is seen/it appears’—which we analyze as raising: thấy scopes over an embedded state (đã đặng nhiều ngày ‘have persisted many days’). This is exactly the kind of bridging context predicted by grammaticalization models: a perceptual verb in a non-canonical subject environment shifts toward evaluation/appearance (‘seem, be found’), i.e., PERCEIVE → KNOW/THINK.

(8)	Clausal complement with rằng
	埃	𧡊	浪	唭		羅		世	態
	ai	thấy	rằng	cười		là		thế	thái
	who see		COMP laugh			COP		worldly.mores
	‘Whoever realizes that laughter is (mere) worldly fashion.’
	(124. 即事 Tức sự ‘Concerning the present matter’ 2)

The complementizer rằng overtly marks a CP complement to thấy. This construction encodes propositional knowledge or judgment, not direct vision, and maps neatly onto complement hierarchies in which ‘see that P’ readily reanalyses as ‘know/think that P’. Within our corpus, thấy rằng is the earliest robust sign that thấy has begun to function as a mental-state/reportative predicate.

(9)

Conditional/discourse linking: thấy… thì…

体

利

𪰛

朱

沛

thấy

lợi

thì

làm

cho

phải

nghĩa

see

profit

TOP

do

give

right righteousness

‘When/If one sees profit, then act in accordance with righteousness.’

(173. 寶鏡警戒 Bảo kính cảnh giới ‘Precious Mirror for Admonition’ 46)

In (9) thấy profiles an event of detection, followed by thì ‘then’, yielding a conditional/temporal schema (‘on seeing/if one sees’). The verb thus contributes procedural information—how a subsequent clause is to be interpreted—nudging thấy toward a discourse-structuring cue.

The 15th-century evidence therefore strengthens the diachronic trajectory argued in this paper. Literal perception remains robust, but non-literal readings gain grammar via (i) raising contexts with inanimate subjects, (ii) clausal complements with rằng, and (iii) (thấy… thì) linking.

4.4. Thấy in the 16th Century

For the sixteenth century, we analyzed 187 tokens of thấy in Tân biên Truyền kỳ mạn lục tăng bổ giải âm tập chú. Relative to the fifteenth century, the system both consolidates earlier patterns and broadens along two theoretically important dimensions. First, the constructional envelope of thấy expands: besides the robust transitive S–V–O ‘literal seeing’, we find (i) resultative perception complexes (V + thấy: chiêm-bao thấy ‘lit: dream see’, mộng thấy ‘lit: dream see’, nghe thấy ‘lit: listen see’, coi/xem thấy ‘lit: look see’), (ii) raising-like strings with inanimate subjects and (iii) clausal complements (thấy rằng…). Second, the sensory scope extends beyond vision, as predicted by modality hierarchies (Viberg, 1984, 2001), with first attestations of olfactory readings (smell) and emergent reportative/evidential uses.

(10)

Resultative perception complex (dream)

Q1.80a

thửa

đêm

ấy

chiêm bao thấy

hai

người

gái

ấy

that

night

DEM dream

see

two

person

woman

DEM

lại

tạ

rằng.

come

thank say.COMP

‘That night (he) dreamt and saw those two women come to thank (him)…’

Here chiêm bao thấy forms a tight [V_dream + thấy] complex meaning ‘see in a dream’. This remains a visual perception event, but crucially it is internally sourced rather than grounded in the shared, external perceptual field. In narrative discourse, such dream-vision commonly serves as the trigger for reporting or interpreting subsequent propositional content (e.g., what is later said, done, or inferred), and thus provides a bridging context in which thấy begins to support PERCEIVE → KNOW/REPORT extensions.7

(11)	Cross-modal extension (smell)
	Q2.56a
	chẳng	phải	mùi	thế-gian	thửa	hay	phảng-phất	thấy
	NEG	right	smell	world	that	often	faintly	see
	‘It is not a worldly scent; it can only be faintly perceived (lit. ‘faintly seen’).’

The collocation with mùi ‘smell, scent’ signals a modal drift from vision to olfaction. Vietnamese deploys thấy as a cover verb of perception, allowing ‘faint perception’ readings across modalities—an expected step on the sensory hierarchy (vision > hearing > touch > taste/smell). This context constitutes a classic bleaching environment because the original visual semantics of thấy is pragmatically and semantically incompatible with the complement (mùi), forcing a weakened, schematic interpretation (‘perceive/notice’) rather than literal ‘see’. The adverbial phảng phất ‘faintly’ further promotes generalization by foregrounding low-salience perception rather than a visually grounded event. This kind of modality-mismatch and schematic reinterpretation is a well-attested route to semantic bleaching in grammaticalization accounts (Heine, 1993; Hopper & Traugott, 2003).

(12)	Impersonal/reportative matrix use
	Q1.51a
	ngươi Thiên Tích rằng: “ngày xưa		thấy  thần-nhân				cùng bảo rằng
	PRN Thiên Tích say day old		see	deity			also tell COMP
	‘Thiên Tích said: “It is said that long ago a divine being told:…”‘

Clause-initial thấy is best parsed as impersonal reportative, i.e., ‘it is seen/said that…’. The reading is not a concrete visual event but a source-of-information marker, aligning Vietnamese with the PERCEIVE → SAY/CLAIM/DISCOURSE leg of Heine’s event-schema cline.

The sixteenth century evidence shows thấy (i) stabilizing in resultative perception complexes (mộng/chiêm-bao/nghe/xem/coi + thấy ‘lit: dream/dream/listen/watch/look + see’), (ii) traversing the sensory hierarchy into olfaction, and (iii) acquiring impersonal/reportative scope in matrix position.

4.5. Thấy in the 17th Century

In the 17th century we can observe a step-change in both the frequency and the functional range of thấy. Two large, genre-distinct sources anchor this period: the Catholic catechism Phép giảng tám ngày (mid-17th c.; early Quốc ngữ, 179 tokens) and the verse chronicle Thiên Nam ngữ lục (late 17th c.; Nôm, 270 tokens). The catechism’s didactic and reportive style favors uses where thấy evaluates truth and aligns the discourse with scriptural authority; the verse chronicle exploits thấy to stage inward sensations, suppositions, and predictive stances.

4.5.1. Phép Giảng Tám Ngày: Evaluative–Evidential Expansion

In Phép giảng tám ngày, a recurrent collocation is thấy tỏ tường ‘see clearly; it is evident that’, where thấy combines with an adverbial of clarity to present a proposition as demonstrably true. This strengthens an evidential/evaluative function while remaining morphosyntactically verbal.

(13)	dẫu	trong sách		Thích Ca			đã	thấy	tỏ tường,	khi
	though	in	scripture	Shakyamuni			PFV	see	clear,	when
	chưa	có	Thích Ca		đã	có		trời	đất	trước,
	not.yet	exist Shakyamuni			PFV	exist		heaven earth		before
	đã	lâu.
	PFV	ago
	‘Even though it is clearly seen in the scriptures of Shakyamuni that long before Shakyamuni there had already been Heaven and Earth.’

Here, thấy no longer reports a visual event but anchors an assessment of obviousness—a classic bridging context from perception to epistemic/evidential status.

A second pattern is thấy như ‘find/see as; turn out as’, which compares what is encountered with a prior report or expectation. The comparison links perception to reportative evidence (‘as [it had been] said’), edging the verb toward discourse management.

(14)	Bấy giờ		kẻ chăn chiên			thì	sang	cho	đến	thành
	that time		shepherd			TOP	go	up	to	city
	Bethleem,		mà	thấy		như	lời		đức	thánh	thiên thần
	Bethlehem		and	see		as	word		CLF	saint	angel
	đã	bảo	mình		vậy.
	PFV	tell	PRN		thus
	‘Then the shepherds went to Bethlehem and found it as the holy angel had told them

In both (13)–(14), visual semantics motivate the stance (‘I/one sees that…’), but the target meaning is epistemic/evidential: thấy validates a proposition, either by clarity (13) or by match with a report (14). These are key diagnostics that the verb is moving beyond literal vision: (i) modified by epistemic adverbs (tỏ tường), (ii) followed by như + NP/Clause comparing fact and report, and (iii) selecting propositional content rather than concrete objects.

4.5.2. Thiên Nam Ngữ Lục: Mirativity, Prediction, and Experiential Stance

In the late-century verse chronicle Thiên Nam ngữ lục, thấy frequently scopes over internal sensations and eventualities that are not literally visible, often with stance markers that encode seeming or inevitability.

(15)	5992	Dạ	bỗng	bừng bừng	dường	thấy	có	thai.
		belly	suddenly	flare-up	seem	see	have	pregnancy
		‘My belly suddenly flares, as if (I) sense being pregnant.’

The sequence dường thấy ‘seem/appears to be seen’ profiles mirativity—a speaker-internal, surprised realization—using the visual verb to present a felt or inferred state. It provides a ready test for subjectification: literal visibility is impossible; the clause reports inward sensation or inference.

(16)

3602

Loạn

rồi

ắt

thấy

trị

âu

đến

ngày.

chaos already

surely

see

order worry

reach

day

‘After turmoil, one will surely see order return one day.’

Here ắt thấy ‘surely will see’ is predictive–experiential: it promises a future experience to the addressee (‘you/one will see’), a rhetorical stance marker more than a report of vision. The verb functions as a prospective evidential, tying eventual verification to the speaker’s commitment.

Across both texts, thấy expands from visual events to evaluation, report-matching, mirative seeming, and predictive stance. Formally, this expansion is cued by (i) collocations with stance adverbs (tỏ tường, dường), (ii) comparative như-phrases, and (iii) deictic futurity with ắt ‘surely’. Semantically, the verb now licenses propositional complements and non-visual content, consistent with the cross-linguistic cline PERCEIVE → KNOW/THINK → SAY/CLAIM/DISCOURSE. The 17th-century evidence thus marks the take-off point where thấy begins to operate as an evidential–epistemic backbone of Vietnamese discourse while still retaining its perceptual core.

4.6. Thấy in the 18th Century

In the 18th-century Catholic correspondence corpus (75 tokens), thấy advances decisively beyond literal vision into speaker-oriented evaluation and belief, while also developing a metalinguistic/reportative niche tied to literacy practices. These letters—argumentative, strongly stance-laden, and often addressed to ecclesiastical authorities—supply rich ‘bridging contexts’ (Heine et al., 1991; Heine & Kuteva, 2002, 2007; Kuteva et al., 2019) where ‘seeing’ plausibly shades into ‘finding, judging, taking to be the case’. Formally, we observe (i) thấy + [sự/việc NP] as an evaluative predicate (‘find X [to be]…’), (ii) [chúng tôi] thấy (rằng) + CP as an explicitly anchored belief/report construction, and (iii) metatextual thấy chữ rằng… ‘see/read the wording that…’, which licenses reportative/evidential readings via ‘seeing (on the page) → knowing/reporting (in discourse)’.

(17)	Evaluative ‘find/judge’ with nominal complements
	chẳng	có	mấy	khi	thầy	thấy		được	sự		gì
	NEG	have many time			priest	see		get	thing		any
	trái	lẽ
	contrary	reason
	‘the priest seldom finds anything to be contrary to reason.’

Here thấy is no longer a purely perceptual verb: with the abstract nominal sự (gì) trái lẽ ‘(something) unreasonable’, it yields an evaluative ‘find, judge’ reading. This is a natural semantic extension from visual access to epistemic assessment (Viberg, 1984, 2001), facilitated by abstract complements (sự/việc X) and frequency in argumentative prose.

(18)

Belief/assessment with clausal complements

thấy

chẳng

có

lẽ

nào

cho

phô

thầy

ở

lại

see

NEG

have

reason

any

allow

PL

priests

stay

remain

‘(We) think/consider there is no proper ground to allow the priests to stay.’

With a proposition as its scope, thấy functions as a raising-like predicate encoding speaker belief (‘consider/find that…’). This matches the cross-linguistic pathway PERCEIVE → EPISTEMIC think/know (Heine et al., 1991; Hopper & Traugott, 2003), and reflects increasing complement-taking capacity (cf. thấy (rằng) + CP in later stages).

(19)

Metalinguistic ‘see (in writing)’ → reportative

sau

nữa

trong các

sách

quen

đọc,

hễ

đâu

đâu

after

again in

PL

book usual read whenever where where

có

thấy  chữ

rằng: kẻ

không đạo,

thì

phải

chữa

lại

exist

see word

say:

person no

faith

TOP

must amend again

‘Furthermore, whenever we see the wording rằng “kẻ không đạo” in the books we usually read, it must be corrected…’

Thấy chữ rằng… anchors evidence to written text (‘I see/read that X’), a clear reportative evidential bridge: what is ‘seen’ on the page becomes that which is reported or cited in discourse. This metatextual pattern prepares the later generalization of thấy (rằng) + CP as an evidential/belief marker in non-literacy contexts.

(20)

Overtly anchored stance: chúng tôi thấy (rằng)…

mà

chúng

tôi

thấy

hai

lý

đoán thì

lấy

sự

ấy

and

PL

I

see

two

reason

judge

TOP

take

matter that

làm

quái-gở

do

strange

‘And we find the two grounds for that judgment bizarre.’

First-person plural subject plus thấy yields an intersubjective stance construction (‘we consider/find that…’), typical of collective petitions and deliberative letters. Here ‘intersubjective’ is used in Traugott’s sense: the stance is explicitly oriented toward an addressee and toward negotiating shared alignment, not merely expressing a private speaker belief. In this epistolary setting, chúng tôi thấy ‘lit: we see’ frames the judgment as a group-endorsed, publicly accountable evaluation and functions persuasively to invite the recipient to accept the assessment as jointly warranted (‘we’ as an epistemic community). On this interpretation, thấy has shifted from reporting an individual perceptual experience to encoding collectively ratified evaluation and argumentative positioning—i.e., subjectification extended to addressee/community-oriented intersubj ectification (Traugott, 2014).

The 18th-century letters consolidate thấy as a stance/evaluative predicate and as a reportative gateway tied to reading/writing (thấy chữ rằng). Structurally, we see stable templates (thấy + sự/việc NP; [chúng tôi] thấy (rằng) + CP), consistent with Dik and Hengeveld’s complement hierarchy (1991): complement scope broadens from NP to CP as semantic abstraction rises. Functionally, these uses push thấy along the PERCEIVE → KNOW/THINK and PERCEIVE → SAY/CLAIM tracks, setting up the 19–20th-century expansion into generalized epistemic, evaluative, and reportive markerhood.

4.7. Thấy in the 19th Century

The 19th-century stage is represented by two Romanized prose sources that bridge late premodern usage and fully modern Vietnamese: Sách sổ sang chép các việc (Philippe Bỉnh, 1822/1968, 160 tokens of thấy) and Thầy Lazaro Phiền (Nguyễn, 1887; 57 tokens). Two distributional facts matter for theory. First, in Philippe Bỉnh, thấy… thì is highly productive (≈58 tokens), yielding a clause-linking template that packages an experiential trigger → knowledge/conclusion event schema (Heine, 1993), a prime bridge from PERCEIVE→KNOW. Second, in Nguyễn Trọng Quản, first-person stance (tôi thấy) becomes routine (≈32/57 tokens), marking a step toward subjectification and epistemic/evaluative construals (Traugott & Heine, 1991; Traugott, 2014). Across both texts, serial/resultative perception complexes (e.g., xem thấy ‘lit: watch see’) remain robust, and thấy participates in sensory packaging consistent with the vision > hearing > touch > taste/smell hierarchy: vision leads, but auditory imagery and inner feeling inferences become more frequent, paving the way for epistemic and discourse uses.

4.7.1. Sách Sổ Sang Chép Các Việc (1822): Experiential Trigger → Knowledge via thấy… thì; Resultative xem thấy

(21)

khi

tôi

đến

macao

mà

xem

thấy

thầy

Liễn

cùng

when

I

arrive

Macao

PRT

look

see

priest Liễn and

thầy

Nhân

thì

giật

mình

lên

priest

Nhân

TOP

startle self

up

‘When I reached Macao and, looking, saw Father Liễn and Father Nhân, I started in surprise.’

The combination xem thấy ‘look-see’ profiles controlled visual inspection → achieved perception (result state). With thì, the whole [experience] → [reaction] sequence is grammaticalized into a clause-linking construction. Frequent thấy… thì in Philippe Bỉnh entrenches a procedural template that later supports evidential and discourse-structuring functions.

(22)	cụ	chấn	thì	về	nói	rằng	tôi	thấy	vèo vèo		trên
	elder Chấn TOP			return say		that	I	see	ONOM.REDUP on
	đầu	tôi,	cái	đét	cái,	mà	đã	bảy	chức	rồi.
	head	I	then	crack	CLF	PRT	PFV	seven order already
	‘Elder Chấn came back and said: ‘I saw (things) whizzing over my head, then a sharp ‘crack!’, and I had already received seven orders.’’

The first-person tôi thấy conveys direct experiential access. The sentence reports a narrated experience, a discourse configuration that facilitates reportative/evidential readings in later texts.

4.7.2. Thầy Lazaro Phiền (1887): Routinized First-Person Stance Tôi Thấy and Experiential > Epistemic Shift

(23)

ngày

kia

tôi

thấy

bạn

tôi

mệt

lắm

thì

tôi

biết

day

DEM

I

see

friend

I

tired

very

TOP

I

know

đã

gần

giờ

nó

phải

lìa

khỏi

thế

rồi,

nên

tôi

PFV

near

time

3SG

must

leave

out.of

world

already

so

I

ở

gần

một

bên

bạn

tôi

luôn.

stay

near

one

side

friend

I

continuously

‘One day I saw my friend was very weak; then I knew it was near his time to leave this world, so I stayed right by his side.’

The matrix tôi thấy scopes over a stative VP (bạn mệt lắm), yielding a “find/feel that” reading rather than pure visual detection. The thấy… thì tôi biết… pairing lexicalizes the PERCEIVE → KNOW trajectory, with biết encoding the epistemic conclusion. This tightly matches Heine’s event-schema pathway and Traugott’s subjectification: the experiencer-subject’s assessment becomes the discourse point of view.

Overall, the 19th century completes the shift from literal seeing to a general experiential-epistemic predicate that organizes discourse and encodes stance—precisely the bridge predicted by cross-linguistic grammaticalization pathways and constructional change.

4.8. Thấy in the 20th Century

In Sống mòn (Nam Cao, 1944/1956), we register 327 tokens of thấy. The profile in this modern realist prose is decisively non-perceptual: literal visual readings are outnumbered by subjective stance, epistemic evaluation, and discourse-pragmatic uses. This distribution matches our theoretical expectations about late-stage grammaticalization—namely, advanced subjectification/intersubjectification and constructional layering (cf. Hopper & Traugott, 2003; Traugott & Trousdale, 2013). Two macro-developments are especially salient. First, thấy is entrenched as a psychological predicate (‘feel, find’), often with stative or evaluative complements. Second, thấy participates in higher-order stance constructions, including combinations with modals and exclamative templates, signaling the culmination of the sensory→epistemic→discourse cline documented earlier.

Below we illustrate four high-frequency constructions that typify this late stage.

(24)

Stative psych predicate (thấy + internal-state NP + ADJ)

Mỗi

buổi chiều

thứ bảy,

Thứ

thấy

lòng

nhẹ

bỗng.

every

afternoon

Saturday

Thứ

see

heart

light

light

‘Every Saturday afternoon, Thứ felt his heart lighten.’

Here thấy profiles interoceptive affect, not vision. The complement is an internal-state noun (lòng ‘heart, feeling’) modified by an adjective/adverbial (nhẹ bỗng ‘light’). This is a canonical subjectified use: the experiencer (Thứ) presents an evaluative state as directly ‘felt/seen’, aligning with the sensory hierarchy’s drift from vision → (internal) feeling and our pathway PERCEIVE → KNOW/JUDGE → STANCE. The constructional template is: [Experiencer NP] thấy [STATE-NP] [EVAL-ADJ].

(25)	Lexicalization to cognition (nhận thấy ‘realize, recognize’)
	Nhưng	Thứ	nhận	thấy	tính nết	y	đột nhiên	đổi	hẳn.
	but	Thứ	recognize	see	character	3SG	sudden	change complete
	‘But Thứ realized that his character had suddenly changed completely.’

The combination nhận thấy is a cognitive verb (‘realize’), with thấy bleached and lexicalized inside a tighter V-V complex. The complement (tính nết y đột nhiên đổi hẳn) is propositional in force (≈a CP), even if not overtly rằng-marked. This instantiates raising to judge: the experiencer asserts a belief state about a proposition. Constructionally, this belongs to the cognitive judgment family and corroborates the drift perception → cognition (Heine & Kuteva, 2002; Kuteva et al., 2019), now entrenched in the lexicon.

(26)	Stance + modality (thấy + modal necessity clause)
	Thứ	cũng	thấy	cần	phải	ra	ngoài.
	Thứ	also	see	need	must	go	outside
	‘Thứ also felt (it) necessary to go out.’

Thấy scopes over a modalized infinitival (cần phải ‘need must’ + VP), yielding ‘judge as necessary’. This is a stance construction: [Experiencer] thấy [MODAL necessity + VP], where thấy no longer encodes perception but evaluative commitment (epistemic-deontic interface). Such complements are strong evidence of constructionalization: thấy functions like an evaluative matrix predicate selecting modal content.

(27)	Exclamative/pragmatic particle (thấy + EVAL-ADJ)
	Thấy		ghét!
	See		annoying
	‘So annoying!’

This subjectless exclamative is a pragmaticized use: thấy contributes an experiential immediacy (‘I directly feel/judge X’) while the adjective (ghét) carries the evaluation. The construction [thấy + EVAL-ADJ] resembles degree/exclamative patterns (‘so X!’), marking discourse-level stance and affective intensity. Functionally it is at the pragmatic endpoint of the pathway, where thấy has become a discourse marker of immediacy and evaluation.

The 20th-century evidence shows thấy fully grammaticalized beyond vision: (i) psychological predicate with internal-state complements, (ii) lexicalized cognitive verb (nhận thấy), (iii) stance-modal matrix with cần phải, and (iv) exclamative particle. We treat these developments as semantic bleaching in the specific sense that the original ‘see-with-the-eyes’ requirement of thấy is no longer needed to interpret the construction: with internal-state nouns (lòng), evaluative adjectives (ghét), and modal predicates (cần phải), thấy contributes a schematic ‘experience/assess’ meaning rather than visual perception.

These late-stage patterns also highlight Vietnamese-specific constructional outcomes that are typologically noteworthy. In particular, Vietnamese makes heavy use of (a) the narrative/argumentative pivot construction thấy… thì… (‘on seeing/once finding X, then Y’), (b) optional-complementizer propositional frames thấy (là/rằng/Ø) P that allow rapid scope expansion without overt subordination morphology, and (c) productive serial/resultative perception complexes (xem/coi/nghe + thấy) that package ‘attempted perception → attained perception’ within a multi-verb template. These constructions provide multiple, formally distinct ‘bridging sites’ for the PERCEIVE → KNOW/THINK → DISCOURSE trajectory, a profile that is especially compatible with Vietnamese isolating morphosyntax and is less dependent on inflectional evidential morphology than in many languages.

Repeated tokens in comparable contexts strengthen a form–meaning pairing and turn it into a readily accessible default in production and interpretation. In Sống mòn, patterns such as tôi thấy (là)…, thấy + Adj, and thấy cần phải… are not occasional stylistic choices but high-frequency, routinized resources for stance and evaluation, indicating that these constructions have become entrenched in modern prose and, by extension, in modern Vietnamese usage.

5. Discussion

These patterns collectively instantiate the late stages of our general trajectory and satisfy our research questions: they document semantic bleaching (from seeing to feeling/knowing), syntactic reanalysis (from NP objects to clausal/modal complements), and discourse-pragmatic expansion (exclamative stance marking), thereby aligning Vietnamese with cross-linguistic perception→epistemic→discourse pathways while highlighting language-specific constructional entrenchment in modern prose. This study set out to trace how Vietnamese thấy evolved from a concrete visual-perception verb to a polyfunctional marker encoding epistemic stance, evidentiality, affect, and discourse organization. Three findings emerge that jointly address our research questions and speak to broader issues in grammaticalization and constructional change.

First, with respect to the first research question concerning semantic change, the diachrony reveals a layered cline from perception to cognition and discourse. In the earliest stratum (12–13th c. Nôm Buddhist prose), thấy is overwhelmingly literal and transitive (S–V–O), with only incipient ‘experiential’ readings in contexts that join vision to another sense, a configuration that supplies ideal bridging contexts for reanalysis (Heine et al., 1991; Traugott & Dasher, 2002). From the 14th to 15th centuries, two new semantic domains stabilize: (i) experiential/subjective predication (e.g., thấy + ADJ/VP) and (ii) propositional evaluation with clausal scope (thấy rằng/là + CP). These extensions instantiate the well-attested sensory hierarchy and semantic trajectory VISION → KNOW/THINK → EVIDENTIAL/DISCOURSE (Viberg, 1984, 2001; Sweetser, 1990; Evans & Wilkins, 2000; Heine & Kuteva, 2002; Kuteva et al., 2019). By the 16–17th centuries, reportative and inferential uses appear (‘thấy rằng…’; narrative thấy… thì pivots), alongside movement down the sensory hierarchy (e.g., thấy mùi ‘perceive smell’), confirming that higher senses provide the primary launchpad for epistemic/evidential development while lower senses grammaticalize later and more narrowly. The 18–19th centuries expand speaker-anchored and evaluative meanings (chúng tôi thấy…, ‘we find/think…’), in line with subjectification in semantic change (Traugott, 1989; Hopper & Traugott, 2003). By the mid-20th century prose, non-literal uses dominate: thấy functions as a stance/evidential predicate, an affective/psych verb (‘feel, find’), an intensifier, and a discourse organizer—exactly the upper end of the cross-linguistic cline predicted for ‘see’ verbs (Aikhenvald, 2004; Heine & Kuteva, 2002; Kuteva et al., 2019).

Second, concerning the second research question about syntactic reanalysis, the Vietnamese data show constructional pathways predicted by functional–typological work on complementation hierarchies (Dik & Hengeveld, 1991; Noonan, 2007; Palmer, 2001; Nuyts, 2001). We observe a stepwise expansion of complement scope: from NP/VP objects (direct perception) to raising-like uses with inanimate subjects (resultative readings), then to CP complements (thấy rằng/là P), and ultimately to matrix-like reportatives and discourse operators. Postverbal perception complexes (e.g., coi/xem/nghe + thấy) already frequent in the 14th–15th centuries, strengthen the experiential layer and create tight form–meaning pairings where ‘perceived result state’ is grammaticalized. The thấy… thì frame emerges as a narrative pivot linking perceived trigger to subsequent event, a construction that accumulates discourse-structuring function. Across the successive stages, thấy broadens its complement options (host-class expansion), relaxes the selectional and argument-structural constraints characteristic of a concrete perception verb (decategorialization), and becomes increasingly schematic as a stance/discourse operator (semantic bleaching). At the same time, the development is not a clean semantic “reset”: perceptual residue persists and continues to license later epistemic and interactional interpretations, consistent with continuity-based accounts of grammaticalization and constructionalization (Heine, 1993; Traugott & Trousdale, 2013; Lehmann, 2015).

Third, regarding cross-linguistic alignment, Vietnamese thấy closely tracks the universal pathway from perception to epistemic/evidential and discourse functions documented for ‘see’ verbs in diverse families (Viberg, 1984, 2001; Sweetser, 1990; Aikhenvald, 2004; Heine & Kuteva, 2002; Kuteva et al., 2019).8 At the same time, several language-specific properties shape the timing and form of change. The analytic profile of Vietnamese and the optionality of complementizers (rằng, là) facilitate loose subordination, yielding fertile bridging sites for scope widening from NP → CP. Serial/complex predicates (V + thấy) provide periphrastic event schemas that ease reanalysis from ‘perceive X’ to ‘experience/find that X,’ in line with cognitive–conceptual accounts of grammaticalization (Traugott & Dasher, 2002). Finally, discourse routines such as first-person tôi/chúng tôi thấy accelerate subjectification and interpersonal alignment, fitting predictions that items move from objective perception to encoded stance (Traugott, 1989; Hopper & Traugott, 2003). The increased prominence of first-person frames (tôi thấy, chúng tôi thấy) is consistent with broader findings that epistemicity and evidential stance frequently correlate with person and speaker engagement, particularly in contexts of epistemic authority and responsibility (Bergqvist & Kittilä, 2017; Bergqvist, 2025; Keinänen, 2025).9

Together, these results show that Vietnamese thấy exemplifies a canonical perception-to-stance pathway while revealing construction-specific solutions—especially thấy… thì and V+thấy resultatives—that are characteristic of Mainland Southeast Asian discourse and predicate serialization. The case thus supports an integrated view where grammaticalization proceeds within constructions but retains explanatory value in terms of source semantics, event schemas, and hierarchies (Heine et al., 1991; Traugott & Trousdale, 2013).

6. Conclusions

By leveraging a seven-century corpus, this paper has documented how Vietnamese thấy advanced along a robust cross-linguistic cline from visual perception to knowledge/stance and discourse management, with intermediate stages anchored in experiential and evidential uses. The semantic trajectory aligns with the sensory hierarchy (vision > hearing > touch > taste/smell) and with established event-schema pathways (PERCEIVE → KNOW/THINK → SAY/CLAIM/DISCOURSE) driven by bridging contexts (Viberg, 1984, 2001; Heine et al., 1991; Heine, 1993; Traugott & Dasher, 2002; Hopper & Traugott, 2003). Syntactically, increases in complement scope, the rise in raising-like and CP complements, and the consolidation of thấy… thì and V + thấy complexes instantiate constructionalization dynamics (Traugott & Trousdale, 2013; Lehmann, 2015) while preserving source–target continuity. Typologically, Vietnamese contributes a detailed non-Indo-European case showing how an isolating, tone-bearing language repurposes a single perception verb to encode evidential basis, epistemic commitment, affect, and discourse pivoting—thus enriching comparative models of perception-verb grammaticalization (Aikhenvald, 2004; Heine & Kuteva, 2002; Kuteva et al., 2019; Noonan, 2007; Mélac, 2023).

Methodologically, the study underscores the value of combining distributional trends (rising non-visual uses across periods) with micro-constructional analysis of contexts that license reanalysis. Theoretically, the findings support a complementary stance: grammaticalization retains independent explanatory force, but the units of change are constructions, whose evolving form–meaning pairings can be modeled in a network (Traugott & Trousdale, 2013). Future work can extend this approach to related Vietnamese perception verbs (e.g., nghe ‘listen’, ngửi ‘smell’, nếm ‘taste’, sờ/chạm ‘touch’), testing whether lower-sense verbs replicate, lag behind, or diverge from the thấy trajectory, thereby refining sensory-hierarchy predictions for Mainland Southeast Asian languages.