Skip to navigation – Site map

It looks like this complementizer used to be an adjective

Mathilde Pinson


The conjunction like meaning ‘as if’ is usually considered to derive from the preposition. There exists, however, a striking parallel between the impersonal post-copula use of the conjunction (e.g. It’s likeAS IF he didn’t see it) and the extraposed construction of the now extinct adjective like meaning ‘likely’ with a covert complementizer (e.g. It’s likeADJ ø he didn’t see it). This paper therefore aims at sustaining the hypothesis of the adjectival origin of likeAS IF with the help of the Corpus of Historical American English. By documenting the geographic, semantic and syntactic expansion of likeAS IF, this study suggests that the disappearance of the attributive use of the adjective for reasons of isomorphism triggered the constructionalization of the two remaining adjectival constructions. The constructionalization of likeAS IF can be formalized as follows: [it copula Adj [ø + Clause]] ↔ [probability] > [it copula [Comp + Clause]] ↔ [comparison].

Top of page

Full text


1When adjectives grammaticalize into other parts of speech, they tend to develop either adverbial or prepositional functions. Adjectives which become adverbs typically encode degree, like very (from the French adjective vrai), pretty, bloody, real, pure or dead. Adjectives which behave more or less like prepositions include worth, like, due (to), near, next (to) and opposite.

2Conversely, the direct shift from an adjective to a complementizer is very rare from a typological perspective and is unique in the history of English. Adjectives are not mentioned as typical sources of complementizers in Heine & Kuteva [2007], who cite the noun channel, the verb channel (e.g. say in some English-derived creoles), the demonstrative channel (e.g. that) and the interrogative channel (e.g. French que).

3The emergence of the conjunction like meaning as if is usually attributed to a host-class expansion of the preposition like [López-Couso & Méndez-Naya 2012b]. However, an alternative hypothesis can relate like to the obsolete epistemic adjective like meaning ‘likely’. Although the mutation from adjective to complementizer is not well attested cross-linguistically, it becomes understandable when one considers the full construction involving the adjective. This is why the constructionalization approach devised by Traugott & Trousdale [2013] is used in this study, together with a more traditional grammaticalization framework.

4After preliminary remarks on the various uses of like and the previous studies that have addressed likeAS IF, I will endeavour to establish its geographical origin and its approximate date of birth. Then the adjectival hypothesis will be presented in detail and will be substantiated by a quantitative study based on the Corpus of Historical American English [1810-2009, Davies 2010-] which documents the semantic and syntactic expansion of likeAS IF.

1. Preliminary remarks

1.1. The various uses of like

5Like is undoubtedly one of the most polyfunctional morphemes in the English language. As is well-known, it can be a noun (e.g. You shouldn’t mess with the likes of him), a verb (e.g. I don’t like chocolate), an adjective (e.g. She replied in like manner), a preposition (e.g. She’s not like me), a discourse marker (e.g. He’s like very nice), a suffix (e.g. He’s very childlike), a component in the quotative expression be like (e.g. She was like: ‘I don’t know’) and a conjunction. In fact, there are two different types of conjunctive like: likeAS and likeAS IF.

(1) LikeAS: He sings just like his father did.

(2) LikeAS IF: He looks like he’s getting better.

6Both of the above conjunctions, which are said to stem from an erroneous use of the preposition, have been overtly stigmatized in style books for decades, as shown by Meyers [1995] and D’Arcy [2007, 2017]. Other examples of proscription against their use include Webster [1790], Raub [1897], Fowler [1908] and Curme [1931] for likeAS, and Wood [1962] and Follett [1966] for likeAS IF. Now, however, these two conjunctions are more or less tacitly accepted within the norms of English, as suggested by Huddleston and Pullum [2002: 1158] or Murray and Simon [2004: 243]. Yet, they remain understudied.

1.2. Previous studies

7The semantic notions encoded by likeAS and likeAS IF are usually called Similarity and Comparison, respectively [cf. Quirk et al. 1985: 1110-1111]. These two notions are often considered minor types of semantic relations, as suggested by their noticeable absence from some general works on clause-linking [Givón 2001; Cristofaro 2003]. This can be explained by the fact that Similarity and Comparison lie outside the domain of ‘core’ adverbial relations [Aikhenvald 2009: 381]. Passing reference is made to likeAS and likeAS IF in several grammars of English, such as Biber et al. [1999: 844], Leech & Svartvik [2002: 400], Huddleston & Pullum [2002: 1151-1154]. Especially interesting is Kortmann [1997], whose typology of adverbial subordinators in European languages contains a chapter on the evolution of adverbial subordinators in English [1997: 291ff.]. Other notable exceptions include Bryant [1962], who addresses the social distribution of like, and Bolinger [1986], who focuses mainly on the hypercorrective use of as (if). In addition, there are also numerous theoretical studies dealing with one type of structure in which likeAS IF can appear, namely copy-raising [e.g. Rogers 1971, 1973; Postal 1974; Horn 1981; Potsdam and Runner 2001; Landau 2009; Asudeh 2012; Asudeh and Toivonen 2012]. This subtype of structure involving likeAS IF includes a raising verb followed by a finite clause whose subject is usually coreferential with that of the main clause, as illustrated in the following example:

(3) He seems like/as if/as though he wants to leave.

  • 1 Others treat like-complements as PPs [e.g. Potsdam & Runner 2001; Huddleston & Pullum 2002; Asudeh (...)

8None of the works mentioned above addresses likeAS IF from a diachronic perspective. Other researchers, like Bender & Flickinger [1999] and López-Couso & Méndez-Naya [2012a; 2012b], study the various types of constructions in which like, as if and as though can be used and focus on their use as complementizers. Indeed, after appearance verbs and object-oriented perception verbs, likeAS IF can be considered a complementizer [cf. also Rooryck 2000; Matushansky 2002; Brook 2014], as shown by several syntactic tests [López-Couso & Méndez-Naya 2012a: 174-176]. These types of clauses are obligatory in clause structure; they convey the argument of a semantic predicate; they can be replaced by unambiguous complement clauses; they may appear in coordination with prototypical complements; they cannot be moved to an unambiguously adverbial position and they pronominalize in the same way as complements do.1

9Additionally, Bender & Flickinger [1999: 11] and López-Couso & Méndez Naya [2012a: 177; 2015] demonstrate that the complementizers as if and as though originally derived from adverbial subordinators and these researchers extend this conclusion to likeAS IF. However, they do not actually study the emergence and evolution of likeAS IF per se, notably because “like is both harder to search for and more recent” [Bender & Flickinger 1999: fn. 5].

2. Date of birth and geographical origin of likeAS IF

10Although the Oxford English Dictionary (henceforth OED) mentions a few examples of likeAS IF dating back to Early Modern English (s.v. like def. 6e), it seems that likeAS IF was seldom used before the nineteenth century. It then started to be used extensively, particularly in the South of the United States, as Simpson [1952] observes:

[T]he conjunctive like is more widely prevalent in America outside New England. Though Wentworth, in his American Dialect Dictionary, cites general usage from 1820, most of his citations are drawn from the South and West, and he specifically notes that the construction is ‘almost universal in Texas.’ Moreover, C. Alphonso Smith, southern scholar and historian, […] has supplied seven rules for writing the southern dialect, of which the first is: ‘Like does duty for as if in such a sentence as, He looks like he was sick’ Indeed, look like and feel like plus a clause are used so frequently in the South that they are considered southern idioms. [Simpson 1952: 463-464].

  • 2 The geographical provenance of each instance of like was determined by reading it in its wider cont (...)

11Indeed, many of the early works included in COHA that contain instances of likeAS IF stem from the South of the USA (e.g. Kentucky, Virginia, Alabama, Tennessee, Louisiana and in particular Georgia),2 as shown in the following graph. It is only subsequently that the use of likeAS IF became generalized across the USA.

Figure 1: Geographical extension of likeAS IF [COHA]

Figure 1: Geographical extension of likeAS IF [COHA]

12The Early Modern English examples cited in the OED may not be related to the instances of likeAS IF that we observe today. The Early Modern period corresponds to an era of great linguistic innovation [Kortmann 1997: 302]. It is the period with the largest inventory of subordinators [Kortmann 1997: 294] but 70% of those introduced during that period were rarely used or turned out to be short-lived (compared with 20% of those created during the Middle English period) [Kortmann 1997: 301]. The Early Modern examples mentioned in the OED can therefore be considered instances of a kind of experimental use of the conjunction, which did not catch on in the language used in Britain at the time.

13There are many other cases of extension of prepositions into conjunctions which did not survive, such as:

(4a) From she was twelve yeer of age, she of hir love graunt him made. [Romaunt Rose, 1366, OED]

(4b) To remaine […] during a necessary conveniency might also be had for the repairing of her own ship. [Cloria and Narcissus, 1653, OED]

  • 3 Similarly, an extensive database such as the Old Bailey Proceedings (1674-1913) suggests that the u (...)

14Moreover, the use of likeAS IF in Early Modern English was so rare that no instances can be found in the Helsinki Corpus (1150-1710) [Gisborne & Holmes 2007]. Gisborne and Holmes even go so far as to say that “like introducing a clause did not occur in English till after [their] period (1150-1710)” [2007: 20], which strengthens the idea that the early examples from the OED are of a purely experimental kind.3

15This evidence tends to show that the Early Modern examples mentioned in the OED do not reflect a fully-fledged use of the conjunction. This conjunction did not catch on in British English and the use that can be observed today comes from the South of the United States. Since this usage undoubtedly originated in speech, there was inevitably a time lapse before it was first recorded in writing. One may therefore speculate that it emerged sometime during the eighteenth century, and started to be used in writing during the early nineteenth century. In COHA it first appeared in non-standard fictitious Southern dialogues, often associated with African American Vernacular speakers.

3. The adjectival hypothesis

16The two following sentences contain strikingly similar clauses:

(5a) Their father said something, too, and it sounded like he was angry. [Help wanted: stories, Gary Soto, 2005, COHA]

(5b) [I]t is like she was angry when she heard that thou wert fond of a poor Indian woman of Darien. [The Damsel of Darien, vol. 2, William Gilmore Simms, 1839, COHA]

17Yet, this parallel is only superficial. Sentence (5a), which dates from 2005, includes an instance of likeAS IF, while the use of like in sentence (5b) is of a different type, even though this is not apparent at first sight. It becomes clearer, however, if the sentence is given in full:

(5b’) If she loves my lord, it is like she was angry when she heard that thou wert fond of a poor Indian woman of Darien. [The Damsel of Darien, vol. 2, William Gilmore Simms, 1839, COHA]

18The replacement of like by as if becomes ungrammatical here, although it is possible in (5a) and in the truncated version (5b) mentioned above:

(5b’’) *If she loves my lord, it is as if she was angry when she heard that thou wert fond of a poor Indian woman of Darien.

19In this sentence, as the initial if-clause indicates, like is not a complementizer but an adjective meaning ‘likely’ (from now on likeADJ). Indeed, the if-clause expresses a hypothesis which is followed by a potential consequence. What is at play here is a deduction based on logic; the speaker evaluates the probability of a state of affair, based on a hypothetical premise. The use of likeAS IF in sentence (5a), by contrast, encodes an inference based on perception, here an auditory one.

20Interestingly, in the truncated version (5b), the status and meaning of like is impossible to ascertain and only the context can help to identify the correct meaning of like. This is also the case in another example, from the Compleat Dictionary of English and Dutch by William Sewel, published in 1766:

(6) It is like he did not see it.

  • 4 The translations into Dutch given in the dictionary may also be useful: vermoedelijk / waarschijnli (...)

21This sentence is ambiguous between an adjective and a complementizer interpretation, and it is only because it belongs to a set of unambiguously adjectival examples that we can identify its correct status and meaning.4 These examples include:

(7) He is not like to live long.

(8) It is not like that he should say so.

22In examples (5b’) and (6), the co-text helps to identify the precise value of like, but in many instances, such as (9), even the co-text is insufficient to enable the reader/listener to distinguish between the adjective and the complementizer:

(9) [T]hey all knew that long before the waters could be lowered so that any attempt to save them could be made, the foul air of that small chamber would have done its fatal work. […]. Even as they tried to talk, poor Boodle, saying that he was sleepy, lay down on the bare rock floor, where he was almost instantly fast asleep and breathing heavily. “‘T is like he’ll never wake again,” said one of the miners, gloomily. “Let him sleep, then; ‘t is the easiest way out of it,” responded a comrade. [Derrick Sterling: A Story of the Mines, Kirk Monroe, 1888, COHA]

23In this sentence, there is no objective element that can rule out either interpretation with certainty. In such a case, the difference in meaning between the two interpretations of like has very little communicative relevance. Is the miner saying that Boodle will probably never wake up (likeADJ) or is he implying that Boodle has the appearance of someone who will never wake up (likeAS IF)? These two readings, that of likely imminent death and that of an appearance of imminent death, are identical from a pragmatic perspective. The only nuance that separates them is that the use of likeADJ emphasizes the deduction based on logic, while in the case of likeAS IF the logical deduction is backgrounded in favour of a notion of inference based on perception. Yet, these two types of reasoning are inseparable: in order to be able to make a logical deduction based on what one knows about the danger of toxic gases, one has to use the sensory evidence at hand: sight (i.e. the miner has fallen asleep), sound (i.e. he is breathing heavily), smell (i.e. the air is ‘foul’), etc. In short, both are evidential, as they express “the perceptual and/or epistemological basis for making a speech act”, which is Cornillie [2009]’s definition of evidentiality. In addition, both interpretations express an epistemic modality; in each case, the speaker does not fully endorse the proposition ‘he will never wake up’ but presents it as a probable event. The semantic proximity between the two versions suggests how the mutation from an epistemic adjective to an evidential complementizer was possible.

4. The constructionalization of like

4.1. The obsolescence of likeADJ

24In the OED [s.v. like def. 8 and 9.a.], the use of the adjective like meaning ‘likely’ is illustrated by examples dating from 1380 to 1896. The adjective appeared in three micro-constructions:

(a) The attributive construction: Cloudy, a like change of weather [1757]

(b) The predicative raised construction: He was like to fail

(c) The predicative extraposed construction: It is like (that) he will fail.

25The attributive micro-construction died out earlier than the other two, undoubtedly because like could easily be confused with the other meaning of the adjective, namely ‘similar’. For instance, the phrase “a like fate” could both mean “a likely fate” and “a similar fate”. In the attributive construction like was therefore replaced by the monosemous adjective likely for reasons of isomorphism, i.e. the tendency towards biunique mapping between form and meaning. No instances of the attributive micro-construction can be found in COHA, which suggests that it had already disappeared (at least in American English) by the end of the eighteenth century.

26The raised construction survived longer but had steadily lost ground by the nineteenth century. Its demise is illustrated in the following graph, which represents the proportion of the forms {be like to; am like to; is like to; are like to} as opposed to {be likely to; am likely to; is likely to; are likely to} in COHA.

Figure 2: Proportion of be like to as opposed to be likely to in COHA

Figure 2: Proportion of be like to as opposed to be likely to in COHA

27During the nineteenth century, speakers were still sporadically confronted with the raised micro-construction, but had no longer access to the attributive micro-construction, which had already died out. As they could no longer associate the raised construction with the corresponding adjective, many speakers became unable to analyze this construction appropriately. Their hesitation was doubtless also increased by the great polyfunctionality of the word like. Consequently, some speakers reanalyzed it as a verb, as in the following example:

(10) Jest then some bilin hot steam come up into my throte that liked to blow’d my nose rite out by the roots. [Major Jones’s Sketches of Travel, William Tappan Thompson, 1848, COHA]

28Others started to treat like to as a kind of adverb, liketa, replaceable by almost:

(11) And it liketa scared him to death! [Feagin 1979]

29These two uses are avertive and mean ‘be about to do’, ‘almost do’ (cf. the French modal faillir) [Romaine & Kytö 2005]. The semantic connection with the epistemic adjective is quite clear here, as the expression is used for an event that was highly probable but did not actually happen. This (ad)verb still occurs sporadically in some non-standard varieties spoken in the South of the United States and the Appalachians.

4.2. Severance of link and reanalysis

30As the attributive construction of likeADJ had disappeared, speakers/hearers found the two micro-constructions left (the raised and the extraposed constructions) opaque, because they were unable to relate them with the corresponding adjective. The severance of the link between the adjective and the extraposed micro-construction led to a different kind of reanalysis of likeADJ. The following sentence, with an adjective followed by an extraposed nominal that-clause with a covert complementizer:

(12a) It’s likeADJ [ø he’ll never wake again].

was rebracketed as follows:

(12b) It’s [likeCOMP he’ll never wake again].

with like turning into an overt complementizer.

31The constructionalization, which can be formalized as [it copula Adj [ø Clause]] ↔ [probability] > [it copula [Comp Clause]] ↔ [comparison], is summarized in Figure 3.

Figure 3: Change in inheritance

Figure 3: Change in inheritance

32The constructionalization led to a change in inheritance of like. It completely ceased to be treated as an adjective and acquired conjunctive properties. By analogy with as if and as though it subsequently also became an adverbial subordinator, introducing adjuncts.

4.3. A case of primary grammaticalization

  • 5 The idea that conjunctions are more grammatical than prepositions is somewhat debatable, notably be (...)

33The emergence of likeAS IF and that of likeAS therefore illustrate two different types of grammaticalization. The appearance of likeAS constitutes a case of secondary grammaticalization, from preposition to conjunction, hence from a grammatical to an even more grammatical item.5 By contrast, the emergence of likeAS IF results from a case of primary grammaticalization, from an adjective to a complementizer, hence from a lexical to a grammatical item.

  • 6 On the various parameters of grammaticalization, see for instance Lehmann [1982]; Heine, Claudi & H (...)

34The parameters used to identify cases of grammaticalization are present in the case of likeAS IF.6 First, the process exhibits decategorialization [Hopper 1991: 22] of likeADJ. As a conjunction, like loses the properties that it displayed as an adjective. It is now incompatible with a raised subject (*He is likeAS IF to fail), an adverb of degree (*It is very likeAS IF he will fail), a comparative or superlative (*It is more likeAS IF he will fail) and the prefix un- (*It is unlikeAS IF he will fail).

35Second, likeAS IF lost its former paradigmatic variability [Lehmann 1982]. Although likeADJ can alternate with several other adjectives, such as likely, expected, probable, possible, plausible, believable, conceivable, etc., likeAS IF can only be replaced with a closed list of items, namely as if, as though and sometimes that or zero.

36Third, the rebracketing had an impact on the placement of a prepositional phrase encoding the experiencer. Compare for instance the two following sentences, where like is used in comparable contexts, after the copula verbs look and appear:

(13a) Looks like [to me] they need it mighty bad. [Judith of the Cumberlands, Alice McGowan, 1908, COHA]

(13b) “‘Pears [to me] like you’re mighty slow,” she said, complainingly. [A Campfire Girl’s First Council Fire, Jane L. Stewart, 1914, COHA]

  • 7 One cannot totally rule out the adjectival interpretation in (13b), since it is not incorrect to sa (...)

The placement of the PP after like in (13a) seems to orient the interpretation toward an adjectival status, while it is quite improbable to find intervening material between likeAS IF and the subject of the like-clause (cf. Pinson [2015]). This tends to suggest that like is a complementizer in (13b).7

37Also noticeable is an increase in token frequency, all the more so because likeADJ was a rare adjective. It became much more frequent when it acquired conjunctive properties. Subsequently, we also observe an increase in type frequency, since likeAS IF was later extended to other syntactic contexts (see Section 5.2.). This increase in type frequency goes hand in hand with semantic generalization (or desemanticization). In the case of likeAS IF, we observe a mutation from a content-word, describing the properties of a referent as being ‘within the realm of credibility’, to a non-factual clausal connective.

  • 8 Note, however, that the diphthong /aɪ/ is typically realized as a monophthong in Southern American (...)

38Another feature of grammaticalization is erosion, which may (or may not) affect grammaticalized items in the final stages of their development (cf. Heine & Kuteva [2007: 42-43]). As is well-known, adjectives are stressed while monosyllabic conjunctions are not, although diphthongs, such as /aɪ/ here, usually retain some prominence.8 A few examples in COHA can potentially be considered a sign that like is unstressed, owing to the use of a particular graphic rendering. By removing the final letter in like, some authors may have attempted to transcribe its deaccentuation:

(14) I dun nussed dat man an’ his baby thru’ yaller fever, ‘pears lik’ he thinks he neber can do ‘nuff for ole Aunt Savannah […]. [Thirty Years of Freedom, Katherine Davis Chapman Tilman, 1902, COHA]

39Similarly, in the novel Bad Boy at Home, the spelling lik is sporadically used for the preposition or the conjunction, but never for the verb like, as illustrated by the comparison between (15) and (16):

(15) Maria […] begun a cryin lik her hart wuld brak. [Bad Boy at Home, Metta Victoria Fuller Victor, 1885, COHA]

(16) I alwus like to help my ’mployers outer a tite place. [Bad Boy at Home, Metta Victoria Fuller Victor, 1885, COHA]

40One element which seems at first to contradict general findings on grammaticalization is that low frequency items are not good candidates for grammaticalization, and yet the adjective like meaning likely was quite rare. However, given that like is very polyfunctional and very frequent in its other uses, particularly as a preposition, the low frequency of the adjective did not prove to be a hindrance. In other words, it is probably an interplay between the syntax of the adjectival construction and the frequency of the preposition that led to the emergence of a highly productive clausal connector.

5. Semantic and syntactic extension

5.1. Semantic extension

41The conjunctions as if, as though and like are compatible with two types of meaning: the evidential/epistemic meaning and the irrealis/counterfactual meaning (see Pinson [2018] for more detail). The evidential meaning reflects an inference based on sensory stimuli, as in:

(17) He looks like he’s getting better.

while the irrealis meaning corresponds to a comparison between an actual situation and an imaginary one, as in:

(18) The moonlight makes the park hills, lakes, trees, and meadows look like they’ve been dipped deep in blue light and purple shadow [The Mayor of Central Park, Avi, 2005, COHA]

  • 9 The periods compared vary greatly in length, but this is due to the dramatic increase in the use of (...)

42If the adjectival hypothesis is true, it implies that the conjunction originally had an evidential/epistemic meaning and that it is only later that it acquired an irrealis meaning. The corpus study shows that indeed the proportion of epistemic readings has decreased over time while irrealis cases have increased.9

Table 1: The semantic evolution of likeAS IF

  • 10 There were also a few indeterminate cases, which are not presented here.




COHA [1820-1879]

63.7% [N=79]

36.3% [N=45]


COHA [1880-1899]

55.2% [N=95]

44.8% [N=77]


Sample from COHA [2000-2010]

39.1% [N=108]

60.9% [N=168]


43As Table 1 shows, the proportion of irrealis readings has dramatically increased between the 19th century and the early 21st century and the difference is statistically significant (χ2 test: p < 0.0001).

5.2. Syntactic extension

44The conjunction likeAS IF has also extended its range of syntactic uses over time. A syntactic comparison between the nineteenth-century part of COHA and a sample of occurrences from the 2000s in COHA suggests two stages, which can also help to sustain the adjectival hypothesis.

5.2.1. Anteriority of structures with impersonal subjects

45The adjectival hypothesis proposed here postulates that likeAS IF’s first context of appearance involves an impersonal subject:

(19) It + be/seem/appear/look… + like-clause

because the adjective from which it derives cannot be used with a referential subject together with a finite clause:

(20) *He looks like(ly) he will be sick.

  • 11 Asudeh & Toivonen [2012]’s study on copy-raising with verbs of appearance shows that there exist fo (...)

46If the adjectival hypothesis is true, it implies that copy-raising structures, which involve a referential subject and a finite clause, appeared only subsequently.11 Therefore, the proportion of impersonal subjects was probably higher in older texts than it is now. To test this hypothesis, I have compared the percentage of impersonal subjects relative to the number of post-copular uses of likeAS IF in two different periods of COHA.

Table 2: The syntactic evolution of likeAS IF [1]

  • 12 The distinction between the two percentages is statistically significant, as shown by a χ2 test (p (...)

Impersonal subjects


Percentage of impersonal subjects relative to the number of post-copular uses of likeAS IF12

19th century

30.22% [N=81]


46.28% [81/175]

Early 21st century

18.1% [N=49]


33.33% [49/147]

47The decrease in the proportion of uses with an impersonal subject can be interpreted as a sign that this construction predates copy-raising.

5.2.2. Anteriority of the complementizer over the adverbial subordinator

48Today, likeAS IF can be used both as a complementizer, after a copula verb, as in:

(20) It looks like he will be sick.


(21) He looks like he will be sick.

but it can also be used as an adverbial subordinator, as in example (15), repeated here as (22):

(22) Maria […] begun a cryin lik her hart wuld brak. [Bad Boy at Home, Metta Victoria Fuller Victor, 1885, COHA]

49The adjectival hypothesis implies that the complementizer predates the adverbial subordinator, since it involves that likeAS IF’s context of emergence is in post-copular position.

Table 3: The syntactic evolution of likeAS IF [2]

Table 3: The syntactic evolution of likeAS IF [2]
  • 13 The decrease in post-copular uses is statistically significant (χ2: p = 0.0004).
  • 14 This development runs counter to the unidirectionality of grammaticalization, as complementizers ar (...)

50As can be seen in Table 3, the proportion of use of like as a complementizer has diminished over time,13 while that of like as an adverbial subordinator has increased. This result can be interpreted as evidence that the use of likeAS IF started in a post-copular position, before being extended to more peripheral functions. This suggests that, contrary to what is postulated in Bender & Flickinger [1999] and López-Couso & Méndez Naya [2012a], likeAS IF has followed the opposite path from that of as if and as though. While as if and as though were primarily adverbial subordinators and later became complementizers, likeAS IF seems to have evolved from a complementizer to an adverbial subordinator.14 This is not surprising, given the polyfunctionality and polysemy of like, which make it unlikely to have appeared directly in a peripheral position. Indeed, like is very hard to process unless there are clear cues as to its grammatical status (cf. Pinson [2015]). That is why its proximity to a copular verb is a much needed favouring context.


51The adjective like meaning ‘likely’ was originally used in three micro-constructions: the attributive construction, the raised construction and the extraposed construction. As shown by its absence from COHA, the attributive construction disappeared before the other two, undoubtedly for reasons of isomorphism. The disappearance of the attributive use has triggered the constructionalization of the two remaining uses of the adjective. In other words, the decreasing exposure to the attributive use led to a lack of entrenchment; speakers-hearers then became unable to draw the link between epistemic adjectives and the two micro-constructions that remained. It was particularly difficult for speakers-hearers to relate the two micro-constructions to the adjective given the wide array of functions of the word like.

52In raised constructions, like gave rise to a non-standard avertive marker diversely treated as a verb or as the adverb liketa. By contrast, the extraposed construction provided the bridging context allowing for the grammaticalization of like into a complementizer. The impersonal construction was rebracketed from ‘It’s likeADJCOMP he’ll never wake again]’ into ‘It’s [likeCOMP he’ll never wake again]’. This paper substantiates this hypothesis by documenting the semantic expansion of likeAS IF (from epistemic to irrealis), as well as its syntactic expansion (from impersonal to copy-raising and then adverbial constructions). This suggests that it is indeed the impersonal epistemic use which predates the other ones, pointing to an adjectival origin of likeAS IF. The constructionalization can be formalized as follows: [it copula Adj [ø + Clause]] ↔ [probability] > [it copula [Comp + Clause]] ↔ [comparison].

53This paper thus reassesses the role of the adjective channel for the grammaticalization of complementizers, giving centre stage to a long-gone marginal adjective which turns out to have somehow survived through the use of its very popular conjunctive offspring. This study also adds to the general picture of the polygrammaticalization of the word like, which can be summarized as follows:

Figure 4: Summary of the polygrammaticalization of like

Figure 4: Summary of the polygrammaticalization of like
Top of page


Aikhenvald Alexandra Y., 2009, “Semantics and Grammar in Clause Linking”, in Dixon R. M. W. & Aikhenvald Alexandra Y. [Eds.], The Semantics of Clause Linking, Oxford: Oxford University Press, 380-402.

Asudeh Ash, 2012, The Logic of Pronominal Resumption, Oxford: Oxford University Press.

Asudeh Ash & Toivonen Ida, 2012, “Copy raising and perception”, Natural Language and Linguistic Theory 30.2, 321-380.

Biber Douglas, Johansson Stig, Leech Geoffrey, Conrad Susan & Finegan Edward, 1999, Longman Grammar of Spoken and Written English, London: Longman.

Bender Emily & Flickinger Dan, 1999, “Diachronic Evidence for Extended Argument Structure”, in Bouma Gosse, Hinrichs Erhard, Kruijff Geert-Jan M. & Oehrle Richard [Eds.], Constraints and Resources in Natural Language Syntax and Semantics, Stanford, CA: Center for the Study of Language and Information, 3-19.

Bolinger Dwight, 1986, “As Strikes Back”, American Speech 61.4, 332-336.

Brook Marisa, 2014, “Comparative complementizers in Canadian English: Insights from early fiction, University of Pennsylvania Working Papers in Linguistics 20.2, Article 2.

Bryant Margaret M., 1962, Current American Usage, New York: Funk & Wagnalls Company Inc.

Bybee Joan, Perkins Revere & Pagliuca William, 1994, The Evolution of Grammar: Tense, Aspect and Modality in the Languages of the World, Chicago/London: University of Chicago Press.

Cornillie Bert, 2009, “Evidentiality and epistemic modality: On the close relationship between two different categories”, Functions of Language 16.1, 44-62.

Cristofaro Sonia, 2003, Subordination, Oxford: Oxford University Press.

Curme George O., 1931, Syntax: A Grammar of the English Language, Boston: D. C. Heath and Company.

D’Arcy Alexandra, 2007, “Like and language ideology: Disentangling fact from fiction”, American Speech 82.4, 386-419.

D’Arcy Alexandra, 2017, Discourse-Pragmatic Variation in Context: Eight Hundred Years of Like, Amsterdam/Philadelphia: John Benjamins.

Davies Mark, 2010-, The Corpus of Historical American English: 400 million words, 1810-2009, available at

Feagin Crawford, 1979, Variation and Change in Alabama English: A Sociolinguistic Study of the White Community, Washington D.C.: Georgetown University Press.

Follett Wilson, 1966, Modern American Usage, New York: Hill and Wang.

Fowler Henry Watson, 1908, The King’s English, 2nd edition, Oxford: Clarendon Press.

Gisborne Nikolas & Holmes Jasper, 2007, “A history of English evidential verbs of appearance”, English Language and Linguistics 11.1, 1-29.

Givón Talmy, 2001, Syntax: An Introduction, vol. 2. Amsterdam: John Benjamins.

Heine Bernd, Claudi Ulrike & Hünnemeyer Friederike, 1991, Grammaticalization: A Conceptual Framework, Chicago: University of Chicago Press.

Heine Bernd & Kuteva Tania, 2007, The Genesis of Grammar: A Reconstruction, Oxford: Oxford University Press.

Hopper Paul J., 1991, “On some principles of grammaticalization”, in Traugott Elizabeth & Heine Bernd [Eds.], Approaches to Grammaticalization vol. 1, Amsterdam: John Benjamins, 17-35.

Hopper Paul J. & Traugott Elizabeth, 2003 [1993], Grammaticalization, 2nd edition, Cambridge: Cambridge University Press.

Horn Laurence, 1981, “A pragmatic approach to certain ambiguities”, Linguistics and Philosophy 4, 321-358.

Huddleston Rodney & Pullum Geoffrey K., 2002, The Cambridge Grammar of the English Language, Cambridge: Cambridge University Press.

Kortmann Bernd, 1997, Adverbial Subordination: A Typology and History of Adverbial Subordinators Based on European Languages, Berlin: Mouton de Gruyter.

Kytö Merja & Romaine Suzanne, 2005, “‘We had like to have been killed by thunder and lightning’: The semantic and pragmatic history of a construction that like to disappeared”, Journal of Historical Pragmatics 6, 1-35.

Landau Idan, 2009, “This construction looks like a copy is optional”, Linguistic Inquiry 40.2, 343-346.

Leech Geoffrey & Svartvik Jan, 2002, A Communicative Grammar of English, 3rd Edition, London: Longman.

Lehmann Christian, 1995 [1982], Thoughts on Grammaticalization, Köln: Arbeiten des Kölner Universalien-Projekts 48.

López-Couso María José & Méndez-Naya Belén, 2001, “On the history of if- and though-links with declarative complement clauses”, English Language and Linguistics 5.1, 93-107.

López-Couso María José & Méndez-Naya Belén, 2012a, “On the use of as if, as though and like in Present-day English complementation structures”, Journal of English Linguistics 40.2, 172-195.

López-Couso, María José & Méndez-Naya Belén, 2012b, “On comparative complementizers in English: Evidence from historical corpora”, in Vázquez Nila, Creation and Use of Historical Corpora in Spain, Cambridge: Cambridge Scholars Publishing, 311-334.

Maling Joan, 1983, “Transitive adjectives: A case of categorical reanalysis”, in Heny Frank & Richards Barry [Eds.], Linguistic Categories: Auxiliaries and Related Puzzles, vol. 1, 253-289.

Matushansky Ora, 2002, “Tipping the scales: the syntax of scalarity in the complement of seem”, Syntax 5, 219-276.

Meyers Walter E., 1995, “Linguistics in Textbooks: A Forty-Year Comparison”, American Speech 70.1, 30-68.

Murray Thomas E. & Lee Simon Beth, 2004, “Colloquial American English: grammatical features”, in Kortmann Bernd, Burridge Kate, Mesthrie Rajend, Schneider Edgar W. & Upton Clive [Eds.], A Handbook of Varieties of English, 2nd volume: Morphology and Syntax, Berlin: Mouton de Gruyter, 221-244.

Pinson Mathilde, 2015, « Linéarité et interprétation de like », in Amary Valérie & Roussel Emmanuelle [Eds.], Linéarité et Interprétation 1, Travaux Linguistiques du CerLiCO 28, Rennes: PUR, 119-136.

Pinson Mathilde, 2018, « Perception et comparaison : de l’évidentiel au contrefactuel », in Digonnet Rémi, Pour une linguistique sensorielle, Paris : Honoré Champion, 97-121.

Postal Paul M., 1974, On Raising, Cambridge, Ma: MIT Press.

Potsdam Paul M. & Runner Jeffrey T., 2001, “Richard Returns: copy raising and its implications”, in Andronis Mary, Ball Chris, Elston Heidi & Neuvel Sylvain [Eds.], CLS 37: the main session, vol. 1, Chicago: Chicago Linguistic Society, 453-468.

Quirk Randolph, Greenbaum Sidney, Leech Geoffrey & Svartvik Jan, 1985, A Comprehensive Grammar of the English language, London: Longman.

Rogers Andy, 1971, “Three kinds of physical perception verbs”, Papers from the seventh regional meeting of the Chicago Linguistic Society, Chicago: Chicago Linguistic Society, 206-222.

Rogers Andy, 1973, Physical Perception Verbs in English: A Study in Lexical Relatedness. Ph.D. thesis, UCLA, Los Angeles.

Romaine Suzanne & Lange Deborah, 1991, “The use of like as a marker of reported speech and thought: a case of grammaticalisation in progress”, American Speech 66.3, 227-279.

Rooryck Johan, 2000, Configurations of Sentential Complementation: Perspectives from Romance Languages, London: Routledge.

Simpson Harold, 1952, “A Harmless Crime?”, College English 13.8, 463-464.

Simpson John A. & Weiner Edmund S. C. [Eds.], The Oxford English Dictionary on CD-ROM, 2nd edition. Oxford: Oxford University Press.

Traugott Elizabeth C. & Trousdale Graeme, 2013, Constructionalization and Constructional Changes, Oxford: Oxford University Press.

Webster Noah, 1790, Rudiments of English Grammar, Hartford: Elisha Babcock.

Wood Frederick T., 1964 [1962], Current English Usage, London: Macmillan.

Top of page


1 Others treat like-complements as PPs [e.g. Potsdam & Runner 2001; Huddleston & Pullum 2002; Asudeh 2012; Asudeh & Toivonen 2012]. According to these researchers, like/as though/as if- complements are headed by prepositions with clausal complements. For instance, Asudeh [2012] supports the prepositional status of these connectors by evidence related to modification: these three connectors can be modified by the same type of constituents as prepositions (e.g. He appears just/almost as though he has been swimming), while these adverbs cannot modify complementizers. This argument is convincing but assumes that prepositions can have clausal complements. If one does not accept this premise, then the argument is not valid. Indeed, it is considered here that prepositions can only govern NPs including gerund clauses, not finite clauses.

2 The geographical provenance of each instance of like was determined by reading it in its wider context, thanks to the Gutenberg Project. In each case, the location where the story takes place is clearly mentioned.

3 Similarly, an extensive database such as the Old Bailey Proceedings (1674-1913) suggests that the use of likeAS IF was virtually absent from British English during the Late Modern period, even from rather informal types of discourse. While likeAS appears numerous times in the Old Bailey, I only found three occurrences of likeAS IF, in two different passages from the 19th century. The word appears in italics, which suggests that it was considered a mistake by the clerks of the court. LikeAS, conversely, never appears in italics in the Old Bailey. In addition, it has been shown that the use of likeAS IF is more frequent in American than in British Present-day English [Quirk et al. 1985: 1110; López-Couso and Méndez-Naya 2012a: 177] and Brook [2014]’s corpus study suggests that likeAS IF has been attested in Canadian English from the 1890s onward. The fact that it had been widely used in the United States by 1820 but only reached Canadian English later also suggests that the use started in the US, not in the UK.

4 The translations into Dutch given in the dictionary may also be useful: vermoedelijk / waarschijnlijk mean ‘probably’.

5 The idea that conjunctions are more grammatical than prepositions is somewhat debatable, notably because prepositions can vary greatly in this respect (e.g. notwithstanding is less grammatical than of). However, the notion that the shift from preposition to conjunction constitutes a case of grammaticalization is mentioned for instance in Heine and Kuteva [2007: 324].

6 On the various parameters of grammaticalization, see for instance Lehmann [1982]; Heine, Claudi & Hünnemeyer [1991]; Bybee, Perkins & Pagliuca [1994]; Hopper & Traugott [2003]; Heine & Kuteva [2007].

7 One cannot totally rule out the adjectival interpretation in (13b), since it is not incorrect to say It appears [to me] likely you’re mighty slow. However, a rapid search on COCA, iWeb and the NOW corpus shows that the string “likely to me” is much more frequent than “to me likely” (the respective ratios are: 15:3 in COCA; 228:25 in the NOW corpus and 895:54 in iWeb).

8 Note, however, that the diphthong /aɪ/ is typically realized as a monophthong in Southern American English, which may facilitate the deaccentuation of the word like.

9 The periods compared vary greatly in length, but this is due to the dramatic increase in the use of likeAS IF.

10 There were also a few indeterminate cases, which are not presented here.

11 Asudeh & Toivonen [2012]’s study on copy-raising with verbs of appearance shows that there exist four different groups of speakers: i) those who only accept expletive subjects in such structures (e.g. It seems like Alfred hurt Thora.), ii) those who accept copy-raising only when the subject of the sentence is co-referential with the subject of the like-clause (e.g. Alfred seems like he hurt Thora.), iii) those who accept copy-raising regardless of the function of the resumptive pronoun, and iv) those who accept everything, even when there is no copy pronoun at all (e.g. Alfred seems like Thora hurt Jane.). Although Asudeh & Toivonen [2012] do not provide an apparent-time analysis and do not mention the age of their informants, such groupings could potentially reflect a diachronic pattern. This would suggest that the structure involving an expletive subject is the oldest one and that the other structures only appeared subsequently.

12 The distinction between the two percentages is statistically significant, as shown by a χ2 test (p = 0.018).

13 The decrease in post-copular uses is statistically significant (χ2: p = 0.0004).

14 This development runs counter to the unidirectionality of grammaticalization, as complementizers are more grammatical than adverbial subordinators, since they belong to a much more restricted paradigm and are on the whole more semantically bleached. Compare for instance the meaning of if as an adverbial subordinator (I’ll go if it doesn’t rain) and as a complementizer (I don’t know if it will rain). However, the shift from complementisers to adverbial subordinators is well attested cross-linguistically, as indicated by Heine & Kuteva [2007: 254].

Top of page

List of illustrations

Title Figure 1: Geographical extension of likeAS IF [COHA]
File image/png, 9.3k
Title Figure 2: Proportion of be like to as opposed to be likely to in COHA
File image/png, 43k
Title Figure 3: Change in inheritance
File image/png, 14k
Title Table 3: The syntactic evolution of likeAS IF [2]
File image/png, 14k
Title Figure 4: Summary of the polygrammaticalization of like
File image/png, 25k
Top of page


Electronic reference

Mathilde Pinson, « It looks like this complementizer used to be an adjective »Lexis [Online], 15 | 2020, Online since 13 June 2020, connection on 07 July 2020. URL:; DOI:

Top of page

About the author

Mathilde Pinson

Sorbonne Nouvelle University

By this author

Top of page
  • Logo Centre d’études linguistiques
  • Logo Université Jean Moulin – Lyon 3
  • Logo DOAJ - Directory of Open Access Journals
  • OpenEdition Journals