1 Introduction
This study investigates a construction that has become particularly frequent in recent decades but so far has not received attention in the literature. It takes the elliptical form of ‘last I + Verb’, as exemplified in (1) and (2), and has evidential meaning. For ease of reference, we refer to it here as ‘evidential last I’ fragment, or ELI fragment for short.
(1) I thought he was in Australia or Japan, last I heard. (COCA:2000:FIC:Ploughshares)
(2) But really, this country, last I checked was a free country (COCA:2006:SPOK:MSNBC_Cosby)
What makes this construction interesting, apart from its relative novelty, is, on the one hand, its fragmentary form and, on the other, its specific use in discourse, which fulfills a number of different pragmatic functions. The present article aims to shed light on both their use and form by drawing on data from the Corpus of Contemporary American English (COCA; Davies Reference Davies2008–), a one-billion-word corpus covering the years 1990 to 2019. It yields a total of 1,218 instances, which were analyzed both quantitatively and qualitatively.
The study shows that the underlying evidential meaning lends itself to the expression of a range of more specific discourse functions, namely that of a downtoner, a booster and ironic use. What triggers these different functions are specific contextual factors. Although fragmentary form is not exclusively associated with evidential meaning, the data suggest that the choice between full and reduced form is significantly affected by its function, viz. evidential versus non-evidential. Fragmentary form is thus an important signal for evidentiality. It is also argued that ELI fragments cannot be explained by simple ellipsis, but that their grammatical status is best captured by a Construction Grammar account which takes them to be emerging (constructionalizing) constructions.
The article is organized in the following way. Section 2 first delimits the category of ELI fragments, highlighting their links to different full (i.e. non-elliptical) forms. It discusses their potential reconstruction as a specificational sentence (section 2.1), as a temporal adjunct (section 2.2) or as both (section 2.3). Section 3 provides a detailed analysis of the corpus data with regard to their frequencies, preferred predicates, position and other formal features. Section 4 homes in on the discourse functions of ELI fragments, distinguishing between downtoner (section 4.1), booster (section 4.2) and ironic use (section 4.3). Section 5 widens the scope to include also full (i.e. non-fragmentary) forms, comparing them with their reduced counterparts, and section 6 discusses their grammatical status from a constructional perspective. Section 7 offers a brief conclusion.
2 Defining the class
Before investigating their frequencies and specific functions (in sections 3 and 4 respectively), ELI fragments need to be defined more clearly as a category and delimited from related constructions. The following three properties can be identified.
First, there is their fragmentary form. ELI fragments take the form given in (3), where brackets indicate optionality.
(3) (the) last I/we VERB (adjunct)
Accordingly, ELI fragments may include an optional determiner in the form of the definite article the (illustrated by examples (4) and (5)), followed by last and a first-person pronoun, which is typically singular I but may also be plural we (as in example (6)).Footnote 2 The verb is typically in past tense although present tense is possible with remember and recall, as in example (7). The verb may also, infrequently, be followed by an optional adjunct, as in (8).Footnote 3
(4) And, the last I looked, obesity is a major epidemic in this country. (COCA:2010:SPOK:PBS_NewsHour)
(5) Elon Musk, last I checked, still has a lot of balls he is juggling in the air (COCA:2019:SPOK:CNN_Quest)
(6) We think it's highly relevant to ask his campaign director what they feel about gay relationships when, last we heard, she's in one. (COCA:2004:SPOK:CNN_Insight)
(7) Well, last I remember, fifteen of the 21 9/11 terrorists were Saudi citizens (COCA:2019:SPOK:CBS_FaceNation)
(8) And from there, the last I saw from the satellite forecast, it was heading toward Alabama. (COCA:2004:SPOK:NPR_ATCW)
The fragmentary nature, in other words, takes two different forms. One, shared by all ELI fragments, is the lack of a head noun such as time or thing (e.g. the last [time] I looked; the last [thing] I remember). The other, which is typically but not always found, is the lack of a determiner (e.g. [the] last I checked).
Second, although their syntactic status may not always be so clear (see discussion below), ELI fragments are best analyzed as disjuncts (or supplements, extra-clausal constituents; see Dik Reference Dik1997; Huddleston & Pullum et al. Reference Huddleston and Pullum2002). This means that prototypical ELI fragments have extra-clausal status with regard to the clause they are associated with, that is, their host clause. Thus, they are pragmatically but not syntactically linked to their host clause (e.g. Dik Reference Dik1997: 379–409; Peterson Reference Peterson, Collins and Lee1999; Kaltenböck et al. Reference Kaltenböck, Keizer, Lohmann, Kaltenböck, Keizer and Lohmann2016: 1–26). As such, they cannot be questioned, as illustrated in (9a), and cannot be the focus of a corresponding it-cleft, as illustrated in (9b). Similarly, if negated, the scope of the negation will be understood as excluding the ELI fragment, as illustrated in (9c) (e.g. Erteschik-Shir & Lappin Reference Erteschik-Shir and Lappin1979: 46; Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 612–13; Espinal Reference Espinal1991: 729–35; Haegeman Reference Haegeman, Chiba, Ogawa, Yamada, Fuiwara, Koma and Yagi1991; Brinton Reference Brinton2008: 7–9).
(9) For example, last I checked, heroin and cocaine are illegal, right? (COCA:2013:SPOK:Fox:TheFive)
(a) When are heroin and cocaine illegal? – *Last I checked.
(b) *It is/was last I checked that heroin and cocaine are illegal.
(c) No, that's not true. (= Heroin and cocaine are legal; ≠ I didn't check)
As a reflection of their syntactic independence, ELI fragments can also be expected to be prosodically independent, exhibiting their own intonation contour. In writing, this is often indicated orthographically with the help of commas, as in (9) above. Their lack of syntactic integration is also reflected by their positional mobility, allowing for clause-initial position, as in (9), medial, as in (5), and final position, as in (1).Footnote 4 ELI fragments may also be used as a free-standing response signal (see section 3).
Third, ELI fragments – as their name suggests – have evidential function. Evidentiality is defined here in its narrow sense, as indicating a source of information (e.g. Aikhenvald Reference Aikhenvald2004: 3), although, as we will see in section 4, they allow for epistemic extension in different ways, marking speaker attitude (e.g. Chafe Reference Chafe, Chafe and Nichols1986; Dendale & Tasmowski Reference Dendale and Tasmowski2001: 341–3, Aikhenvald Reference Aikhenvald2004: 5). ELI fragments, in other words, refer to the speaker's evidence for making a particular claim in the host clause. This function is owed to the semantics of the ELI predicate, which is limited to expressions of sensory perception (e.g. see, hear), cognition (e.g. remember, recall) and activities of information gathering (e.g. check, count). What they all have in common is that they denote a form of ‘knowledge update’ (cf. ‘predicates of knowledge and acquisition of knowledge’ in Noonan (Reference Noonan and Shopen1985: 129)).
As evidential disjuncts, ELI fragments need to be distinguished from related, but nonetheless different, constructions: specificational sentences and temporal adjuncts. These are related in representing structurally ‘complete’ (i.e. non-fragmentary) reconstructions of ELI fragments (cf. Bowie & Aarts Reference Bowie, Aarts, López-Couso, Méndez-Naya, Núñez-Pertejo and Palacios-Martínez2016). As illustrated in (10), different verb semantics allow for different reconstructions, either as specificational construction or temporal adjunct, or both. In the following we will look at each of these three options in turn.
(10)
2.1 Reconstruction as specificational sentence
Extension to a specificational construction is demonstrated for the ELI fragments in (11), with the reconstructed specificational versions given in (12) (and the reconstructed elements underlined).
(11)
(a) Last I remember, las Reinas didn't respond well to polite. (COCA:2005:FIC:Bk:DirtyLittleSecrets)
(b) Last we knew, you were suspended midair, … (COCA:2016:SPOK:NPR_FreshAir)
(c) Last I recall, you were studying the law. (COCA:2017:TV:TURN:Washington'sS…)
(12)
(a) The last thing (that) I remember is (that) las Reinas didn't respond well to polite.
(b) The last thing (that) we knew was (that) you were suspended midair.
(c) The last thing (that) I recall is (that) you were studying the law.
Reconstruction of a specificational sentence thus requires the addition of the shell noun thing (although other general nouns such as information also qualify; cf. Schmid (Reference Schmid2000)) and the addition of the copula be together with the optional addition of a that-complementizer and/or of a relativizer. In cases where the ELI fragment lacks a determiner, the definite article is also added. In this way the ELI fragment can be expanded into the subject noun phrase of a specificational construction (e.g. Huddleston Reference Huddleston1984: 458; Keizer Reference Keizer1992: 53–82), which allows for syntactic reversal (cf. That las Reinas didn't respond well to polite is the last thing I remember).Footnote 5
ELI fragments differ from the reconstructed specificational sentences not only in their elliptical form but also in their syntactic status: ELI fragments are extra-clausal disjuncts, whereas the noun phrase of a specificational construction is firmly part of sentence grammar, functioning as subject of the copular structure and specifying the semantic role of a variable for the value expressed by the that-clause (e.g. Higgins Reference Higgins1979: 155; Huddleston & Pullum et al. Reference Huddleston and Pullum2002: 266–72; Van Praet Reference Van Praet2022: 76–80).
A structural criterion for distinguishing between specificational and disjunct noun phrases is that of positional mobility. While a disjunct such as (13) is positionally mobile (as a result of its lack of syntactic integration), a specificational NP, as in (14), is fixed in its sentence-initial position (allowing only for syntactic reversal). This lack of positional mobility seems to apply even in cases where the specificational NP has no explicit copula, as in (15). Although native speaker judgments vary on the acceptability of these structures and thus may take the examples in (15) as a disjuncts,Footnote 6 final position with thing is not attested in COCA.
(13)
(a) Last I remember, John was in London.
(b) John was in London, last I remember.
(c) John, last I remember, was in London.
(14)
(a) The last thing I remember is we were driving around Patterson's neighborhood (COCA:1990:MOV:Bad_Influence)
(b) *We were driving around Patterson's neighborhood, the last thing I remember is.
(15)
(a) I mean, the last thing I remember they were fighting. (COCA:1998:SPOK:CBS_48Hours)
(b) ?They were fighting, the last thing I remember.
Structurally, disjunct status is thus associated with positional mobility. In terms of discourse function, it is associated here with evidential meaning. Accordingly, disjuncts such as in (13) can be replaced by other evidential disjuncts, such as as far as I know/remember, which is not possible with specificational subjects, such as (14a). Conversely, only the latter allow for questioning with what? (e.g. What is the last thing you remember?).
2.2 Reconstruction as temporal adjunct
Extension to an adjunct of time is demonstrated for the ELI fragments in (16), with the reconstructed noun phrase adjunct given in (17) and the clausal adjunct given in (18) (reconstructed elements are underlined).
(16)
(a) Last I checked, actually, Romney was the candidate White House advisers most feared (COCA:2011:SPOK:NBC_MeetPress)
(b) Where is Violet? Last I looked, she was puking in the toilet I cleaned just this afternoon. (COCA:2016:TV:Mom)
(c) The last I counted, there was 126 things on there. (COCA:011:TV:KitchenNightmares)
(17)
(a) The last time (that) I checked, Romney was the candidate White House advisers most feared
(b) The last time (that) I looked, she was puking in the toilet I cleaned just this afternoon.
(c) The last time (that) I counted, there was 126 things on there.
(18)
(a) When last I checked, Romney was the candidate White House advisers most feared
(b) When last I looked, she was puking in the toilet I cleaned just this afternoon.
(c) *When the last I counted, there was 126 things on there.
As illustrated by the examples, reconstruction of a temporal noun phrase requires addition of the head noun time (plus the possible addition of a definite article and a that-relativizer). Reconstruction of a temporal adverbial clause, on the other hand, requires the addition of the adverbial subordinator when, and is restricted to ELI fragments without a determiner (as evidenced by the unacceptability of (18c)).
The difference between the expanded and the unexpanded forms is that the former are more clearly temporal adjuncts, whereas the latter are probably best analyzed as evidential disjuncts. However, the difference is far from clear-cut and depends on the type of verb as well as the specific context of use. We will look at this distinction in more detail in the following.
Clear cases of temporal adjuncts are given in (19) and (20). They locate the event of the main clause in time and allow for questioning with when? (e.g. When did you try to throw me off a building?) and can be the focus of an it-cleft (e.g. It was last time we met that you tried to throw me off a building) (on the properties of adjuncts, see e.g. Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 504). As illustrated by the example in (20), such uses can also occur in reduced (i.e. elliptical) form.
(19) The last time we met, you tried to throw me off a building. (COCA:1998:TV:TheNewBatmanAdven…)
(20) I'm a big fan of your movies, as I told you last I was on your show. (COCA:2015:SPOK:ABC:TheView)
To exclude such clear-cut cases of location in time, the category of ELI fragments is restricted to only those predicates that identify a source of information, that is, with verbs of cognition (e.g. remember, recall), sensory perception (e.g. see, hear), and activities of information gathering (e.g. check, count) (as noted above). Within the category of ‘sensory perception’, we also exclude verbs with a direct object (or complement), as in (21), and within the category of ‘information gathering’ verbs of speaking (e.g. talk, speak), as in (22), since their main function is to ground an event in time rather than identify the source of information.
(21) All right, Last I saw you, you were winging off to Chicago to interview… (COCA:2008:SPOK:NBC_Today)
(22) OK, last we talked you were driving around Fort Pierce and describing some terrible damage that you were viewing (COCA:2004:SPOK:CNN_Event)
However, even with such a restrictive definition, the exact semantic and syntactic status of ELI fragments may not be unambiguously clear. Thus, the examples in (16) allow for questioning with both When? (i.e. temporal) and How do you know? (i.e. evidential), as illustrated in (23), and substitution with both yesterday (i.e. temporal) and as far as I know (i.e. evidential), as illustrated in (24). They also allow for the ELI fragment to occur as the focus of an it-cleft, as in (25); this indicates adjunct rather than disjunct status (cf. Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 504, 612–13), although the pragmatic acceptability of (25) may be somewhat doubtful.
(23)
(a) Q: When was Romney the candidate White House advisers most feared? A: Last I checked.
(b) Q: How do you know Romney was the candidate White House advisers most feared? A: I checked
(24)
(a) Yesterday Romney was the candidate White House advisers most feared.
(b) As far as I know, Romney was the candidate White House advisers most feared.
(25) ?It was last I checked that Romney was the candidate White House advisers most feared.
Where the status of ELI fragments as evidential disjuncts is unambiguously clear, however, is in those cases where there is disagreement in tense between the ELI fragment and the host clause, as illustrated by the examples in (26) to (28).
(26) For example, last I checked, heroin and cocaine are illegal, right? (COCA:2013:SPOK:Fox:TheFive)
27) There are, last I counted, 563 billion cats in need of good homes. (COCA:2018:NEWS:ChicagoTribune)
(28) And, the last I looked, obesity is a major epidemic in this country. (COCA:2010:SPOK:PBS_NewsHour)
In these cases the ELI fragment cannot be construed as grounding an event or state of affairs in time, only as reporting on the speaker's source of information. Its disjunct status is also clearly illustrated by its inability to be questioned by when? (e.g. *When are heroin and cocaine illegal?) or to be made the focus of an it-cleft (e.g. *It was/is last I checked that heroin and cocaine are illegal).
2.3 Ambiguous reconstruction
In addition to the possible reconstruction as specificational structures (section 2.1) or temporal adjuncts (section 2.2), some ELI fragments are ambiguous in that they allow for both. Such a dual reconstruction is found only with a restricted number of verbs, such as read, hear and see, as illustrated in (29) to (31).
(29)
(a) Last I read, Zynga reported about 750 current employees in December. (COCA:2012:BLOG:gamasutra.com)
(b) The last thing I read was that Zynga reported about 750 current employees in December.
(c) The last time I read (about it), Zynga reported about 750 current employees in December.
(30)
(a) Last I heard, this issue was being contested in Federal court right now. (COCA:2012:WEB:reason.com)
(b) The last thing I heard was that this issue was being contested in Federal court right now.
(c) The last time I heard (about it), this issue was being contested in Federal court right now.
(31)
(a) Last I saw, he was detoxing at the bottom of a frozen lake. (COCA:2002:MOV:EyeSeeYou)
(b) The last thing I saw was that he was detoxing at the bottom of a frozen lake.
(c) The last time I saw him, he was detoxing at the bottom of a frozen lake.
Cases such as these, with more than one reconstruction option, call into question a simple ellipsis account for ELI fragments; even more so since hear is the most productive of all the verbs found in ELI fragments, accounting for close to 50 percent of all tokens. In the following section we will look at the frequencies of ELI fragments in more detail, and in section 6 we will return to the question of their grammatical status.
3 Forms and frequencies
Based on the discussion in section 2 above, retrieval from the corpus was carried out with the search string ‘last + I/we + VERB’ and subsequently sifted manually to implement the restrictions outlined in section 2.2, viz. excluding predicates that do not lend themselves to an evidential interpretation (i.e. not expressing cognition, sensory perception, or activities of information gathering). Sensory perception verbs, moreover, were restricted to those without a direct object or complement, but adjuncts were allowed for.
Table 1 shows that ELI fragments are substantially represented in COCA with a total of 1,218 instances, most of which occur in initial position with respect to their host clause. Initial position was identified as pre-subject position, while medial includes positioning anywhere between subject and a final complementation or adjunct (except if the adjunct is clausal, in which case it was taken to be final). The different positions are illustrated in (32)–(34).
(32) Last I heard, little Teddy's still working the piers. (COCA:2014:TV:Law&Order:SpecialVictimsUnit)
(33) The LGBT community, last I checked, was composed of human beings. (COCA:2012:WEB:…10dimes.blogspot.com)
(34) The 32GB WiFi only iPad accounts for some 65% of all iPad sales last I saw (COCA:2012:WEB:androidcentral.com)
Frequency of ELI fragments in COCA
An example of a free-standing use is given in (35), which is only rarely attested (cf. Hirota & Brinton Reference Hirota and Brinton2023: 7). It covers cases where the ELI fragment is used in reply to a question, similar to a reaction signal (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 852), thus underlining its disjunct status. Note that this use also allows for negation, as in (36).
(35) A: Is he still in St. Charles?
B: Last I heard. (COCA:1996:MOV:LastDance)
(36) A: Found any family on this kid yet?
B: Not last I heard. (COCA:1997:TV:ER)
Not only are ELI fragments frequently attested in the corpus, they have also become more frequent in recent decades, as demonstrated by figure 1, which shows a doubling since 1990–4. In the interest of a more balanced picture, these figures exclude the text categories Web and Blog, which only feature in the corpus for the year 2012.
Development of ELI fragments in COCA (excluding web and blog): normalized per million words (vertical axis) and raw frequencies (above the bars)
In terms of their distribution in text types, ELI fragments exhibit a preference for informal, speech-like genres (TV/movie, recreation of speech in fiction) and those frequently used for the expression of opinion (blog, web), as evinced in figure 2.Footnote 7 The specific uses of ELI fragments in the expression of speaker stance will be examined in more detail in section 4.
Normalized frequencies of ELI fragments in COCA text types (per million words)
Compared to the overall frequency of ELI fragments, the number of different predicate types attested is rather low. As indicated in table 2, there is a total of only 13 different predicates, with heard and checked outnumbering the others by far. The range of possible predicates is of course restricted by their evidential use, but also suggests a moderate level of productivity.
Predicates attested for ELI fragments in COCA
The majority of predicate types allow for reconstruction as temporal adjunct (with time or when), as discussed in section 2: viz. 6 out of 13 (checked, looked, counted, was measured, tested, did the math). Three predicates allow for specificational reconstruction (knew, remember, recall), and four both (heard, saw, read, noticed). In terms of tokens, however, the most productive category is that of dual reconstruction, accounting for 620 instances out of the overall total of 1,218.
As noted in section 2, ELI fragments may occur without the definite determiner: (the) last I checked. Determiner omission is, in fact, the most frequent option with 83.7 percent (see table 3).
ELI fragments with and without determiner in COCA
Another characteristic identified in section 2.2 was that of potential tense disagreement between the ELI fragment and its host clause. This was investigated in COCA for the predicates allowing extension to a temporal adjunct,Footnote 8 with the results in table 4 revealing a slight preference for disagreement (52.5%), underlining its disjunct status. Instances were identified as agreeing in tense if there was an exact match in tense between the ELI fragment (typically in past tense) and its host clause, as in (37). All other cases were classified as tense disagreement, such as (38).
(37) Last I checked it was easy to get the Google apps back on Kindles. (COCA:2012:BLOG:asymco.com)
(38) While its recent gains may be somewhat slim, last we checked, a win is a win. (COCA:2012:BLOG:engadget.com)
Tense (dis)agreement with the host clause in COCA for predicates allowing extension to a temporal adjunct. (verbless host clauses or free-standing uses are excluded)
4 Discourse functions
Having delimited the category of ELI fragments in section 2 and identified their corpus frequencies in section 3, we now turn to their functional use. As noted above, ELI fragments are, by definition, evidential in function, rather than specificational or temporal. This basic evidential meaning gives rise to three more specific pragmatic functions depending on contextual use: (i) downtoner, (ii) booster, (iii) ironic use. We discuss these functions in turn, illustrating them with examples mainly from spoken data.
4.1 Downtoner
The discourse function that seems to be most commonly associated with ELI fragments is that of a ‘downtoner’. The term downtoner is borrowed from Holmes (Reference Holmes1984: 346, 359), who uses it for linguistic devices that attenuate illocutionary force (e.g. You're a bit of a fool you know). In the present context, the notion of downtoning is equivalent to expressing epistemic modality. The speaker, in other words, uses the ELI fragment to reduce or hedge their commitment with regard to the truth of the proposition in the host clause. In the framework proposed by Prince et al. (Reference Prince, Frader, Bosk and Di Pietro1982: 85), this function corresponds to one particular subtype of hedge referred to as shield, which expresses degrees of speaker commitment. As such, the ELI fragment can be loosely paraphrased by as far as I know/remember or apparently.
The expression of epistemic modality follows naturally from the basic evidential meaning of ELI fragments, as a form of ‘epistemic extension’ (e.g. Aikhenvald Reference Aikhenvald2004: 6): The speaker's commitment to the proposition of the host clause is reduced, based on limited available evidence. This limitation arises from a temporal lag in the information update (as indicated by last) and a potential insecurity associated with the source of information itself, i.e. the mode of knowing, such as ‘hearing’, ‘seeing’, or ‘remembering’. The use of last, importantly, also implies potential change of a state of affairs. Thus, by using an expression such as last I heard or last I checked the speaker conveys: ‘this statement is based on my latest information on something that may have changed in the meantime’. The result is a distancing effect from the actual truth of the proposition in the host clause.
To illustrate the downtoning effect of ELI fragments, let us look at the following examples from the corpus.
(39) They had to drive the chopper three-quarters of a mile with her dangling that way with the kid rescued. Now the 16-year-old, Jason Gelroth, remains in critical condition last -- last we heard. (COCA:1996:SPOK:Ind_Geraldo)
(40) ZARRELLA […] Where we were it was awfully bad. It was awfully bad in Naples. Fort Myers Beach was evacuated. And I have not heard any reports
KING You know,
Sam? CHAMPION: Last I saw they were lined up waiting to kind of get back in towards Sanibel and Fort Myers area. I don't know whether they've opened up to let them get back in or not (COCA:2005:SPOK:CNN_King)
(41) BOB-HENSON# Well, this is a good example of that quandary, because as recently as yesterday, it looked like there was the chance for more significant tornado action today. It now -- at least the last I looked -- was looking more like kind of garden-variety thunderstorms, maybe some severe storms, but probably not the kind of tornadoes like we saw yesterday. (COCA:2013:SPOK:PBS_Newshour)
(42) Matt is not here. He's off on his five cities in five days round the world trip. This morning, day two, and still Al, Ann and I have no idea where he is, even at this very moment. The suspense, though, is killing us. Last we knew on Monday he was in the Pacific Ocean on the Big Island of Hawaii at the Kilauea Volcano. (COCA:2000:SPOK:NBC_Today)
(43) HILL: We did talk about -- someone had made a rifle for her [= Sarah Palin]. They were going to auction off another one. She has not, the last I checked said whether or not she's going to accept it. (COCA:2009:SPOK:CNN_Brown)
In (39), the report of a dramatic rescue, last we heard indicates that the 16-year-old's condition, which was critical, may have changed in the meantime. In (40), a discussion of Hurricane Wilma's impact, last I saw hedges the report of people ‘lining up to get back to Sanibel and Fort Myers’, with the lack of speaker commitment being made explicit in the subsequent sentence: I don't know whether they've opened up to let them get back in or not. In (41), where the speaker discusses the likelihood of further tornadoes, the hedging function of the last I looked is further underlined by the use of at least. In (42) the speaker speculates on Matt's suspected whereabouts, with last we knew indicating that they may have changed in the last few days. In (43), finally, the use of the last I checked expresses reduced speaker commitment with regard to Sarah Palin accepting the rifle or not.
4.2 Booster
Although typically exploited for downtoner uses (as discussed in section 4.1), the evidential meaning of ELI fragments lends itself also to expressing precisely the opposite: conveying emphasis. For such cases we have adopted Holmes’ (Reference Holmes1982: 18–24, Reference Holmes1984: 353–4) term ‘booster’, whose function is that of ‘emphasizing or boosting the illocutionary force of a speech act’ (Holmes Reference Holmes1984: 346). As such, it represents ‘a complementary strategy to that of softening or attenuating’ illocutionary force (ibid.). Although boosting and attenuating are typically associated each with their own specific set of lexical items, Holmes (Reference Holmes1984: 359) shows that both strategies may be expressed by one and the same linguistic unit, and ELI fragments are a case in point.
The boosting effect derives naturally from the evidential meaning of the fragments: They underline the validity of the proposition in the host clause by providing the source of information, thus signaling that the proposition is based on reliable evidence (almost challenging the hearer that they could not disagree with such evidence). In contrast to the downtoning effect discussed above, where reference to the source of information is synonymous with reduced commitment (along the lines of ‘This information is based on an external source which may not be completely reliable; things may have changed in the meantime’), the emphasis here is on the reliability of the evidence. More specifically, the speaker highlights that the view expressed is supported by outside evidence. In contrast to the downtoning function, here the temporal component of last (suggesting that this may not be the most recent, up-to-date information) recedes into the background. The result is that of strengthening the speaker's commitment to the proposition and underlining its importance. Consequently, the ELI fragment can be paraphrased by ‘let me emphasize’ or ‘I emphatically want to point out’.
The following examples from the corpus illustrate this particular use.
(44) We try to be -- refine on those criteria from “DSM- IV” to make it tighter. In fact, I think “DSM-V” will reduce the number of patients who receive diagnoses. Now, for bulimia, I think it's very important to point out that these patients or subjects are highly distressed. It's not just a matter of someone overeating and having a bad meal or a bad day on Thanksgiving. In fact, these -- this disorder is seen commonly in young women. It tends to be associated very commonly with obesity. And, the last I looked, obesity is a major epidemic in this country. And if we are to in fact address and help the society deal with their obesity problems, we have to have a way of defining this pathological overeating. (COCA:2010:SPOK:PBS_NewsHour) video available at: https://archive.org/details/WETA_20100211_000000_PBS_NewsHour/start/2580/end/2640: 7:43pm
(45) PAUL-SOLMAN: […] and on the NASDAQ, the stock exchange for America's fast-growing firms, Amazon's share price has been on a rocky ride for months now. Is it a sign that the Internet stock craze is over, or just another bump on the road to future riches, since Amazon's shares were still valuable enough, last we looked, to buy all of Borders and Barnes & Noble several times over and still have billions of dollars left to play with? (COCA:1999:SPOK:PBS_Newshour)
(46) FERGUSON# Well, here's the other thing. You remember when Barack Obama was running for president the first time and there was the pictures that showed him jumping through a wave. it was in Hawaii. And everybody was like, look how fit the president is. I don't give a rat's rear end about if the president has a six-pack or a keg belly, in my opinion, as long as he knows how to run the country. Because the last I checked, someone that's all fit, you know, is not exactly the exact person that's going to know about foreign policy, about the war on terror, about the economy, about the housing market, about the financial crisis with the banks, about guns in America (COCA:2013:SPOK:CNN:PiersMorganLive)
(47) QUEST): Tesla shares. Now, the stock is that more than 16 percent. Tesla unexpectedly reported third quarter profit. It is the first time the electric car maker has been profitable in three quarters. It says progress at its Shanghai factory is ahead of schedule. Are Tesla's problems behind it?
LA-MONICA): I wouldn't go so far as to say that the problems are behind it. Tesla has obviously had production issues in the past. Elon Musk, last I checked, still has a lot of balls he is juggling in the air because of SpaceX, the Boring Company, that's always going to be an issue as long as Tesla has a bit of a brain drain in their own management ranks. (COCA:2019:SPOK:CNN_Quest)
(48) But I also think Reggie has been with Bees for five years and appears healthy and happy. Even if he's not living the exact way Feline Friends would like to see him living, he's certainly doing far better than he would have as a feral cat. # And if you want proof that Bees cares about the cat, consider this: She's the one that started the legal action after Feline Friends refused to return Reggie. # There are, last I counted, 563 billion cats in need of good homes. I see them wandering the streets and alleys, hungry and skittish. They would all love to be Reggie, to live with cool artists and musicians and become the inspiration for what I assume would be terrible cat-themed folk music. (COCA:2018:NEWS:ChicagoTribune)
In example (44), an interview with a doctor on PBS Newshour, the speaker contradicts the previous speaker's assessment of bulimia (see link for online video). The use of the last I looked emphasizes the importance of strict classification criteria in view of a major epidemic. It serves as a snide remark towards the previous speaker, suggesting something like ‘you seem to forget that there is an obesity problem’ and thus brings out the confrontational nature of the exchange. In (45), the booster function of last we looked ties in with the emphatic point made about Amazon's persistent financial value, viz. that it could still buy all of Borders and Barnes & Noble several times over and still have billions of dollars left to play with. Its medial position, separating the purpose clause (which illustrates Amazon's financial power) from the rest of the sentence, also has a highlighting effect and thus contributes to the emphatic function of the ELI fragment. In (46) the speaker takes a critical stance towards Barack Obama's alleged fitness. The speaker emphatically elaborates on his view, viz. that he does not care about a president's looks, by prefacing it with the last I checked. In (47) the use of last I checked underscores potential problems with Elon Musk's other companies. Its medial position, separating the information structure into topic and comment, contributes to its emphatic function (e.g. Taglicht Reference Taglicht1984: 16, 24). Example (48) is from a newspaper article pleading for the return of the cat Reggie from the cat adoption group Feline Friends to her previous owner, the artist Bees; here the booster function of last I counted ties in with the hyperbolic statement it introduces: 563 billion cats in need of good homes.
With downtoning being the default function for ELI fragments, what are the factors that may trigger the opposite reading, i.e. one of emphasis? Several ingredients can be identified as being conducive to a booster interpretation. One of them is the semantics of the verb in the ELI fragment. As can be seen from the examples above, the verbs express active information gathering (e.g. look, check, count) rather than inert perception (e.g. hear, see) or cognition (e.g. remember, recall). It seems that the idea of an activity on the part of the speaker sits particularly well with the notion of emphasis, which also suggests an active involvement of the speaker (even though no actual activity of, for instance, ‘checking’ or ‘looking’ is carried out).
The most important factor for promoting a booster function, however, is that of co(n)text. One such co(n)text is where the speaker expresses a personal opinion, particularly if it is confrontational and an affective, emotionally charged stance is taken, as in examples (44) and (46). Another type of co(n)text is where the speaker expresses some form of exaggeration or emphasis, as in examples (45) and (48) (buying all of Borders and Barnes & Noble several times over and 563 billion cats). Finally, a booster reading may also be triggered in contexts where the speaker reports on a state of affairs that is generally known and thus does not require evidence (i.e. no checking or looking), as in example (47) (Elon Musk's wide range of investments) and possibly example (44) (obesity being a major epidemic). The latter two contexts may also be conducive to an ironic reading, a function we turn to in the next section.
4.3 Ironic use
ELI fragments are also used for humorous effect, typically for ironic use. Verbal irony, simply put, is characterized by a mismatch between the (implied) meaning of an utterance and its context, with the latter including the speaker's and hearer's shared common ground (for a more precise definition, see e.g. Attardo Reference Attardo, Ruiz-Gurillo and Belén Alvarado Ortega2013: 40). With ELI fragments the ironic effect arises from an incongruence between the implied meaning of ‘potential change of a situation’ (as noted for the downtoner function) and a host clause expressing an immutable ‘fact’. As was discussed in section 4.1, the use of last in a sentence such as Last I heard, John lives in London implies that John's residence is subject to change and may indeed have recently changed. Such an implication of variability does not square well with relatively permanent states (e.g. ‘Gaddafi being dead’ in (51) below) or facts (e.g. ‘Hillary Clinton being a woman’ in (50)).Footnote 9 It is precisely this mismatch of the implied meaning of the ELI fragment (viz. variability) and a specific host clause content (viz. persistence, generally accepted and obvious fact) that is exploited for the expression of irony.
Let us look at some examples to illustrate the ironic use of ELI fragments.
(49) COLMES: Mark, we're just about out of time. You know, last I checked, in this country, Mark Levin, you're innocent until proven guilty. Your case has to be proven against somebody. We're jumping to conclusions and rushing to judgment here. (COCA:1998:SPOK:Fox_HC)
(50) But do you think she [=Hillary Clinton]'s playing the woman card like he said? I know you didn't, apparently, but how about him, her?
RICK-PERRY# Well, I mean, last I checked she's a woman, so I mean, it's kind of like she's not faking it. So this is a nonissue and one that, you know, frankly we need to get focused on the real important issues that face this country. (COCA:2016:SPOK:ABC:TheView)
(51) Not only has he done a lot here -- the green revolution I might point out, the chances of us participating in green revolution coming out of it were near zero. And the idea that we would get ourselves mixed up in that -- and by the way, Libya last I saw, Gaddafi was dead. (COCA:2012:SPOK:Fox_Five)
(52) You can't even get close to the place.” B. J. shot her an odd glance. “What do you mean,’ can't?’ Last I noticed we've each got two good legs.” (COCA:2008:FIC:Analog)
(53) BERNIE-SANDERS-(D)# We are going to take this to the convention. We have real momentum behind us. And last I heard, there are 50 states in the United States of America and we're going to compete in every one of them. (COCA:2016:SPOK:Fox:OnTheRecord:Susteren)
In example (49) the host clause evokes the presumption of innocence, a firmly established and well-known legal principle that is not subject to change and therefore at odds with the implied variability of last I checked. Similarly in (50), Hillary Clinton's gender is typically regarded as an immutable fact that jars with the notion of change suggested by last I checked. In example (51) it is the generally known fact of Gaddafi's death that is incongruent with the temporary nature of the ELI fragment and in (52) it is the clearly observable fact of both interlocutors having healthy legs. In (53), finally, the number of US states is a well-known fact (especially to politicians) and unlikely to change any time soon, while this is precisely what last I heard implies. The effect of this obvious semantic mismatch between host clause and ELI fragment is, on the one hand, the canceling of any literal meaning of the ELI fragment (e.g. there is no actual ‘checking’ or ‘hearing’ involved) and, on the other, the flagging as humorous, ironic use.
5 The role of fragmentary form: comparing full and reduced forms
Having investigated the use of ELI fragments in some detail in the sections above, in this section we now widen the scope to include full (i.e. non-fragmentary) forms. More precisely, the question is what role fragmentary form plays in signaling evidential function and disjunct status.
This is important since evidential meaning and disjunct status are not restricted to ELI fragments alone but can also be expressed by full forms. Thus, we find full forms such as the ones in (54) and (55) with the head noun time. Although the explicit use of the noun time may be indicative of a temporal rather than unambiguously evidential meaning, the tense disagreement with the host clause suggests disjunct status (see section 2.2).
(54) Last time I looked, you need fifty votes. (COCA:2018:SPOK:CBS_NewsEve)
(55) But last time I checked, you are the president of the United States. (COCA:2017:SPOK:Fox:YourWorld:Cavuto)
Similarly, we also find complete when-clauses, as in (56), where the verb (e.g. checked) is suggestive of evidential meaning, although its syntactic status is perhaps less clearly disjunct.
(56) And by the way, when last I checked, Juan, he had a pretty successful presidency. (COCA:2008:SPOK:Fox_HC)
Only for specificational reconstructions (see section 2.1) no ‘complete’ evidential disjuncts can be identified, since full forms with the head noun thing, as in (57), do not allow for positional mobility, as indicated by the questionable acceptability of (58).
(57) Last thing I remember, I was coming back from a run feeling a little lightheaded (COCA:2017:TV:ChicagoMed)
(58) ?I was coming back from a run feeling a little lightheaded, last thing I remember.
While evidential meaning and disjunct status can thus also be found with full forms (i.e. outside ELI fragments), the converse may also hold: Although rare, reduced forms are not excluded from expressing temporal adjuncts or specificational subjects either, as the following examples attest.
(59) I'm a big fan of your movies, as I told you last I was on your show. (COCA:2015:SPOK:ABC:TheView)
(60) Oh, last I saw them, they were down in the basement. (COCA:2018:TV:DC'sLegendsofTomorrow)
(61) The last we see of our family, they are walking single-file down a dirt path toward a touch of woods (COCA:1994:SPOK:NPR_ATC)
Examples (59) and (60) are temporal in meaning (on account of the predicate being on your show and the transitive use of saw them). Example (61) is best analyzed as a specificational subject (lacking the head noun thing and a copula). Thus, fragments are not exclusively wedded to the expression of evidential disjuncts but may also occasionally have non-evidential function.
To investigate the role of fragmentary form for the expression of evidential and non-evidential meaning, the spoken COCA data were analyzed as outlined in table 5. The category ‘full form’, which included either a head noun (time or thing) or the subordinator when, was classified as (i) evidential, if it contained an evidential predicate without object or complement (as discussed in section 2 and exemplified in (54)–(56) above), or as (ii) temporal/specificational: temporal adjuncts are cases with non-evidential predicate (e.g. last time I was on your show, when last we met) or evidential predicate with an object or complement (e.g. last time I saw you); non-copular specificational subject are cases with the head noun thing without a copula, e.g. (57). The category ‘reduced form’ was equally classified as (i) evidential (i.e. ELI fragments) or (ii) temporal/specificational, as exemplified in (59)–(61) above.
Analysis of COCA Spoken: categories and examples
The results for the four different categories in COCA Spoken, given in table 6, have tested as statistically highly significant (χ2 = 167.39, df 1, p < .01, φ = -0.50). This means that the choice between full and reduced form is significantly affected by its function, viz. evidential versus non-evidential. In other words, full forms are more likely to have non-evidential meaning (as temporal adjunct or specificational subject), while fragments are more likely evidential disjuncts. Reduced form thus plays an important role in signaling evidential function.
Analysis of COCA Spoken: frequencies
a The different subtypes of full forms are attested as follows: evidential with time (114 instances), with thing (0), with when (2); non-evidential with time (436), with thing (10), with when (24).
6 Grammatical status: a constructional perspective
As discussed in section 2, ELI fragments can be related to essentially two types of full forms: specificational constructions, on the one hand, and temporal adjuncts (with time or when), on the other.Footnote 10 It thus seems plausible to derive ELI fragments from these more complete structures and identify them as parent constructions in a Construction Grammar sense (e.g. Goldberg Reference Goldberg1995, Reference Goldberg2006; Goldberg & Herbst Reference Goldberg and Herbst2021; Kaltenböck Reference Kaltenböck2021).
From the preceding discussion, the following additional observations have emerged: (i) Fragmentary form signals evidential disjunct status although the two are not exclusively associated with each other, with evidentiality cutting across fragmentary and full forms (see section 5). (ii) As evidential disjuncts the fragments have adopted specific discourse uses (see section 4).Footnote 11 (iii) In terms of verb tokens, the most productive subtype is the one that allows for ambiguous reconstruction (viz. heard, read, saw, noticed), rather than reconstruction as either temporal adjunct or specificational subject (see section 3).
This latter point is particularly interesting as it rules out a simple ellipsis account of ELI fragments. This is because a decisive criterion for ellipsis, according to Quirk et al. (Reference Quirk, Greenbaum, Leech and Svartvik1985: 884), is that of verbatim recoverability, which means that the ellipted words must be precisely recoverable. Given that a substantial number of ELI fragments do not allow for unambiguous (but rather for multiple) reconstruction, this criterion does not apply.
Instead, ELI fragments bring together under one umbrella, as it were, two (or three) different constructions, while at the same time neutralizing the differences between them, much like a syntactic blend. The recent increase in frequency (noted in section 3) suggests that ELI fragments may be emerging as an independent construction. Whether they are already ‘sufficiently frequent’ to qualify for constructional status is unclear (on the criterion of ‘sufficient frequency’ for constructions, see e.g. Goldberg Reference Goldberg2006: 5). We therefore adopt a more prudent approach and take them to be ‘constructionalizing’ (Traugott & Trousdale Reference Traugott and Trousdale2013: 22), i.e. in the process of creating a new ‘type node’ in the speaker's grammar or ‘construct-i-con’ (e.g. Hilpert Reference Hilpert2014). The reason for seeing this process as ongoing rather than completed rests on the observation that neither their fragmentary form nor their evidential function is uniquely idiosyncratic to ELI fragments (on idiosyncrasy as a criterion for constructional status, see e.g. Croft & Cruse Reference Croft and Allan Cruse2004: 263; Goldberg Reference Goldberg2006: 5, Reference Goldberg2009: 94): As noted in section 5, the reduced form may occasionally be found in specificational or temporal uses, and evidential function is also attested for full forms.
While it is difficult to identify clear idiosyncratic features for ELI fragments, they do exhibit some formal characteristics that set them apart from their unreduced counterparts, specificational subjects and temporal adjuncts: unlike specificational subjects (even those lacking a copula, e.g. The last thing I saw, John was having a good time), ELI fragments are fully positionally mobile, allowing also for non-initial position. Unlike temporal adjuncts (e.g. Last time I saw you), ELI fragments are more clearly disjuncts, as evidenced by their propensity for tense disagreement (see sections 2.2 and 3).
Drawing on the framework of Construction Grammar (e.g. Goldberg Reference Goldberg1995, Reference Goldberg2006), the grammatical status of ELI fragments can thus be captured as in figure 3. Here ELI fragments are represented as constructionalizing (hence the dashed line) and as overlapping with (i.e. emerging from) two different full forms: On the one hand, the ‘non-copular specificational construction’, that is, specificational subjects lacking a copula link to their complements, such as The last thing I remember, it touched its claw to my chest (COCA:2018:TV). On the other hand, the ‘evidential disjunct construction’, whose syntactic status may not always be clearly distinguishable from adjuncts, except in cases of tense disagreement. This category comprises two types: noun phrases with the head noun time and when-clauses. ELI fragments are depicted as overlapping with all three of these constructions to indicate their blending of different structures as well as their emerging status (compare Goldberg & Herbst Reference Goldberg and Herbst2021). Note that these constructions themselves are instantiations of different parent constructions (e.g. Croft & Cruse Reference Croft and Allan Cruse2004: 262–4; Fried & Östman Reference Fried, Östman, Fried and Östman2004: 12; Diessel Reference Diessel2019): full ‘specificational constructions’ (i.e. with a copula) or ‘temporal adjunct’ constructions (either in the form of an NP with time or a when-clause).Footnote 12
Taxonomic network links for ELI fragments (cxn = construction)
7 Conclusion
This article has investigated a fragment that has seen increased use in recent decades but has so far not been subjected to a detailed analysis: evidential fragments of the type last I checked (or ELI fragments for short), which were defined as elliptical structures with the form ‘(the) last I/we VERB (adjunct)’. Section 2 has delimited the construction, highlighting its links to different unreduced (or full) forms: viz. specificational sentences and temporal adjuncts. Interestingly, some predicates used in ELI fragments (e.g. hear) are ambiguous in their reconstruction, allowing for both full forms.
Analysis of the corpus data shows that ELI fragments are frequently attested in American English, with a total of 1,218 instances, and a doubling in frequency in the period from 1990 to 2019 (section 3). As a reflection of their disjunct status, ELI fragments are found in initial, medial and final position with regard to their host clause, showing a clear preference for the former. Occasionally, ELI fragments also occur as free-standing discourse units in reply to a question. In terms of its form, the typical ELI fragment prefers omission of the determiner (83.7%) and frequently disagrees in tense with its host clause (52.5% for predicates allowing extensions to a temporal adjunct). Its evidential meaning is expressed by its predicate, which exhibits a rather limited range of 13 verb types, with heard and checked being by far the most frequent in terms of attested tokens. As evidentials, ELI fragments also show a preference for text types that typically express speaker stance (blog, web) as well as informal, speech-like genres (TV/movie, speech in fiction).
The main focus of the investigation, however, has been more qualitative in nature: the discourse function of ELI fragments (section 4). While all ELI fragments, by definition, share a basic evidential meaning, they are also used for a range of more specific discourse functions, all of which can be derived from the underlying notion of evidentiality. The most typical function is that of a downtoner, hedging the speaker's commitment to the truth of the host clause proposition (section 4.1). However, they can also be used to do the exact opposite, namely act as a booster to emphasize the proposition in the host clause (section 4.2). This reading is triggered by specific contextual factors, such as an emotionally charged and confrontational speaker stance. A third type of function identified is that of ironic use, with the irony arising essentially from a clash between the notion of variability implied by the ELI fragment and a host clause expressing an immutable ‘fact’ (section 4.3).
The study has also investigated the role of fragmentary form for signaling evidentiality (section 5). This is interesting, as the reduced form is not restricted to ELI fragments alone but can also be found with temporal adjuncts (e.g. last I was on your show) or specificational subjects (e.g. The last we see of our family, …). Conversely, evidential meaning can also be expressed by full forms (e.g. last time I checked). The statistical analysis suggests that the choice between full and reduced form is significantly affected by its function, viz. evidential versus non-evidential, with fragments more likely to have evidential meaning. As such, reduced form acts as an important signal for evidential meaning.
As regards the grammatical status of ELI fragments, a simple ellipsis account is ruled out since a large number show ambiguity in reconstruction and thus do not meet the criterion of verbatim recoverability. Instead, we proposed a constructional account which identifies ELI fragments as constructionalizing, that is, potentially developing their own ‘type node’ representation, but not yet full constructions in their own right (section 6). To substantiate this claim, however, further research is required, specifically a detailed analysis of diachronic data to establish the exact origin and development of ELI fragments.
Open access
