1. Introduction
Decades of linguistic literature have been dedicated to the remarkable way in which children acquire complex structures at very early ages. Consider a question such as the one in (1), an instance of a so-called filler–gap dependency, where the wh element (the filler) has been extracted from its original position in the embedded clause (the gap – signalled by “___”).
Even though filler–gap dependencies are structurally complex (Chomsky, Reference Chomsky1986), taxing from a processing perspective (Omaki & Lidz, Reference Omaki and Lidz2015) and relatively rare in the input (de Villiers et al., Reference de Villiers, Roeper, Vainikka, Frazier and de Villiers1990), children as young as 3 or 4 years old have been shown to comprehend (and produce) such structures (de Villiers, Reference de Villiers and Rice1996; de Villiers & Roeper, Reference de Villers, Roeper, McDaniel, McKee and Cairns1996; Roeper & de Villiers, Reference Roeper, de Villiers, Weissenborn, Goodluck and Roeper1992, among others).
At first sight, filler–gap dependencies seem to be unbounded, in that they can be established across several embedded clauses, as seen in (2) next. While costly from a processing point of view, such structures are grammatical.
However, there are limits to these dependencies – certain configurations, known in the literature as islands for extraction, ever since Ross’s seminal work (1967), ban long-distance movement: a question such as the one in (3), where the embedded clause is introduced by a wh element, is considered ungrammatical (as marked by the asterisk).
Islands come in many guises and span across a wide variety of constructions (see Szabolcsi & den Dikken, Reference Szabolcsi, den Dikken, Cheng and Sybesma2003, for an overview): some are more categorical, strong islands, banning extraction across the board, irrespective of whether the filler is an argument or an adjunct, as illustrated in (4) next, with an example of a relative-clause island, while others are more permissive, they are weak, allowing, for instance, argument extraction, while still banning the extraction of adjuncts, as illustrated in (5) by means of a factive island.
This article focuses on the acquisition of factive islands, those configurations where adjunct extraction is banned out of the clausal complements of factive verbs (e.g., know, regret, remember).
Factive islands are puzzling not only from a theoretical perspective, but also from the point of view of language acquisition. Even though knowledge of some types of islands (e.g., interrogative islands and negative islands) has been (initially)Footnote 2 argued to be acquired as early as 3–4 years old (de Villiers, Reference de Villiers and Rice1996; Roeper & de Villiers, Reference Roeper, de Villiers, Weissenborn, Goodluck and Roeper1992, among others), factive islands were said to fall into the class of (very) late acquired phenomena. Previous research has shown that, up until the age of 6, children do not observe the barriers on movement imposed by factive verbs. While there are several studies which indicate that children of this age and younger have difficulties with respect to extraction out of the clausal complement of factive verbs (Philip & de Villiers, Reference Philip, de Villiers and Clark1992; Roeper & de Villiers, Reference Roeper, de Villiers, Weissenborn, Goodluck and Roeper1992), to my knowledge, there is only one which determines that knowledge of these restrictions is acquired between the ages of 6 and 8 (de Villiers et al., Reference de Villiers, Curran, DeMunn, Philip, Greenhill, Hughes, Littlefield and Walsh1998). This study is meant as an expansion of this work.
The aim of this article is to investigate the acquisition of factive islands by Romanian monolingual children, with a focus on (the ban on) adjunct extraction from the post-verbal clause of factive verbs. Three main questions will be addressed: (i) Do Romanian monolingual children observe the restrictions imposed by factive verbs on long-distance movement and, if so, at what age? (ii) Are there any differences between cognitive factive verbs (e.g., remember, forget, realize) and emotive factive verbs (e.g., regret, be sad, be happy) with respect to the acquisition of long-distance movement? and, more generally, (iii) What does the late acquisition of the restrictions imposed on long-distance movement in such cases tell us about the theoretical approaches to factive islands?
The remainder of this article is organized as follows: first, I focus on previous research on the acquisition of (factive) islands. Then, I will take a brief detour for a discussion on the differences between cognitive and emotive factive verbs. Against this background, I will present an experimental study on the acquisition of factive islands by Romanian monolingual children, meant to expand on the findings of de Villiers et al. (Reference de Villiers, Curran, DeMunn, Philip, Greenhill, Hughes, Littlefield and Walsh1998), where I will show that factive islands are indeed a late acquired phenomenon and that children make no distinction between cognitive and emotive factive verbs in terms of adjunct extraction. Last, but not least, I will evaluate these findings against some of the most prominent theoretical analyses of factive islands, arguing that the data from acquisition provide evidence in favour of a complex developmental path of factivity, in general, and of factive islands, in particular.
2. The acquisition of (factive) islands – previous research
When and to what extent do children gain knowledge of both long-distance movement and the restrictions it needs to abide by? Such questions have been addressed in the literature on language acquisition over the last three decades. While there is a wide body of literature that suggest that children as young as 3 or 4 years old seem to observe island constraints (de Villiers et al., Reference de Villiers, Curran, DeMunn, Philip, Greenhill, Hughes, Littlefield and Walsh1998; Philip & de Villiers, Reference Philip, de Villiers and Clark1992; Roeper & de Villiers, Reference Roeper, de Villiers, Weissenborn, Goodluck and Roeper1992), there are also researchers who claim that what seems to be an adult-like behaviour could in fact be accounted for in terms of processingFootnote 3 (Goodluck, Reference Goodluck, Hoff and Shatz2007; Omaki et al., Reference Omaki, White, Goro, Lidz and Philips2014). These approaches will be discussed in turn.
Extensive work carried out by de Villiers, Roeper, and colleagues (de Villiers, Reference de Villiers, Fletcher and McWhinney1995; de Villiers et al., Reference de Villiers, Roeper, Vainikka, Frazier and de Villiers1990; de Villiers et al., Reference de Villiers, Curran, DeMunn, Philip, Greenhill, Hughes, Littlefield and Walsh1998; Philip & de Villiers, Reference Philip, de Villiers and Clark1992; Roeper & de Villiers, Reference Roeper, de Villiers, Weissenborn, Goodluck and Roeper1992) indicates that, when given a scenario such as the one in (6) next, children give answers corresponding to both short-distance movement (where the filler has been dislocated from the main clause, giving rise to a so-called upstairs reading – “When did she say that?”) and to long-distance movement (where the filler has been extracted from the embedded clause, leading to a downstairs reading – “When did she rip her dress?”).
Question: When did she say ___ she ripped her dress___?
Answers: (a) At night (upstairs reading) (b) That afternoon (downstairs reading) (example taken from de Villiers, Reference de Villiers, Fletcher and McWhinney1995, p. 524)
This type of experimental design, known in the literature as The Question after Story design, has since then been used extensively in the literature on the acquisition of both filler–gap dependencies and of islands. In one of the first instantiations of this design, de Villiers et al. (Reference de Villiers, Roeper, Vainikka, Frazier and de Villiers1990) wanted to test, among other things, whether children ranging in age between 3;7–6;11 observed the restrictions imposed on movement in the case of a so-called interrogative island (i.e., where the embedded clause is introduced by a wh element). As illustrated next, while argument extraction is possible when the embedded clause is introduced by an adjunct wh medial word (7a), but not by an argument wh medial word, adjunct extraction is deemed ungrammatical by adult native speakers of English, regardless of the type of wh medial word (argument (7c) or adjunct (7d)).
While the authors themselves point out that “it is not the case that children have a completely adult-like grammar at this stage” (de Villiers et al., Reference de Villiers, Roeper, Vainikka, Frazier and de Villiers1990, p. 275),Footnote 5 they report that children produce very few “downstairs reading” answers in contexts where long-distance movement is banned, suggesting that knowledge of the restrictions found in wh islands is available at a very young age.
Wh islands are not the only types of islands which have received significant attention in the literature. The restrictions imposed on movement out of relative clauses have also been investigated. Consider the context in (8) next. As can be seen, only an upstairs reading is possible in adult grammars.
Question: What did the boy paint a bird that had long wings with?
Answers: (a) With a blue brush (upstairs reading) (b) *With feathers (downstairs reading) (example taken from Otsu, Reference Otsu1981, p. 90)
Otsu (Reference Otsu1981) set out to experimentally investigate whether pre-school children obeyed these constraints and showed that, at that age, children violated the barrierhood imposed by relative clauses in around 30% of cases. Different results are however reported by de Villiers (Reference de Villiers, Fletcher and McWhinney1995), who tested adjunct extraction out of relative clauses (in contrast with Otsu, Reference Otsu1981, where the experimental stimuli focused on argument extraction). Using the same Question after Story design, the authors tested extraction out of object relatives, out of subject relatives, and out of extraposed subject relatives, all illustrated in (9).
The authors report that, in the case of object relatives (9a) and subject relatives (9b), there were no answers that would have violated the restrictions imposed on movement, while in the case of extraposed subject relatives (9c), 10.5% of the answers provided by the 3-/4-year-old children and 2% of the answers provided by the 4-/5-year-old children would be considered ungrammatical in adult grammar.
Very few errors are also reported by Goodluck et al. (Reference Goodluck, Foley, Sedivy, Goodluck and Rochemont1992), who tested whether children observed the restrictions on movement imposed by adjunct clauses, as illustrated in (10) next, where the downstairs reading is considered ungrammatical.
Question: What did the fox eat before whistling?
Answers: (a) An ice-cream cone (upstairs reading) (b) *A tune (downstairs reading) (example taken from Goodluck et al., Reference Goodluck, Foley, Sedivy, Goodluck and Rochemont1992, p. 182)
The data collected from 24 three-/four-year-old children indicate a preference for the upstairs reading, with less than 15% of answers corresponding to a downstairs reading. However, as the authors also point out, this error rate should not be brushed off (nor can be it accounted for in terms of performance), and seems to indicate that, while in general restrictions on movement out of adjunct clauses seem to be observed, at this age, “not all children may have mastered movement in the syntax.”
So far, what these studies argue is that, at the age of 3/4 years old, there are very few instances where children provide answers corresponding to a “downstairs reading,” in contexts where the adult grammar bans long-distance movement, seemingly abiding by the constraints imposed on movement. An important question has been raised though: Do children truly observe a grammatical principle, or do they simply prefer an upstairs reading, corresponding to a short-distance movement, due to processing limitations?
Such a view is supported by Goodluck (Reference Goodluck, Hoff and Shatz2007), who starts from the assumption that language processing occurs in a step-wise fashion: small fragments are built as soon as the sentence is heard/read and these smaller units coalesce into larger ones, up to the point where they form sentences. According to Goodluck (Reference Goodluck, Hoff and Shatz2007), “sentence boundaries are the point at which material is recoded into a non-immediate memory representation.”
But processing is not free of constraints. For instance, with respect to long-distance movement, Goodluck postulates the following constraint:
Consider again the sentence in (9a), repeated for convenience below. If the filler were linked to a gap within the relative clause, it would give rise to an incomplete proposition, as indicated in (12b).
Contrast this with the sentence in (13). Extracting “how” from the first possible gap-site gives rise to a complete proposition and it is therefore possible.
A similar line of reasoning could be followed to account for the apparent obedience to the constraints imposed on extraction from adjunct islands, reported by Goodluck et al. (Reference Goodluck, Foley, Sedivy, Goodluck and Rochemont1992). It might be the case that children mostly provided upstairs reading not necessarily because they already abide by the grammatical principles which govern extraction, but rather because extracting from the main clause would give rise to a complete proposition, as illustrated in (14) next.Footnote 6
Further evidence for such a rationale comes from cross-linguistic data. Goodluck et al. (Reference Goodluck, Saah and Stojanović1995) show that a similar preference arises for speakers of Akan, a language where adjunct clauses are not islands for extraction, while Omaki et al. (Reference Omaki, White, Goro, Lidz and Philips2014) test long-distance movement in English and Japanese and argue in favour of a short-distance movement bias in both languages, even in contexts where long-distance dependencies are, in fact, grammatical.
Therefore, while knowledge of island constraints was initially claimed to be acquired early, it is not clear whether children’s answers truly reflect knowledge of grammatical principles or a mere preference, stemming from a smaller processing load.
However, regardless of the lens through which the data are accounted for, the picture becomes even more complicated when looking at factive islands, as children seem to allow island violations (allowing long-distance dependencies) in such configurations even as late as 8 years old.
2.1. The acquisition of factive islands
In a series of experimental studies, de Villiers and colleagues show that, up until the age of 6, children fail to observe the barriers imposed on long-distance movement out of the clausal complement of a factive verb and that, at this age, they make no difference between bridge verbs (those that allow extraction) and non-bridge verbs (verbs that restrict extraction) (Philip & de Villiers, Reference Philip, de Villiers and Clark1992; Roeper & de Villiers, Reference Roeper, de Villiers, Weissenborn, Goodluck and Roeper1992). When presented with a context such as the one in (15), children offered answers corresponding to both short- and long-distance movement, regardless of the verb.
Question 1 (non-factive verb): When did the boy say he got a bruise?
Answers: (a) At night (upstairs reading) (b) That afternoon (downstairs reading)
Question 2 (factive verb): When did the boy know he got a bruise?
Answers: (a) At night (upstairs reading) (b) *That afternoon (downstairs reading) (example taken from Roeper & de Villiers, Reference Roeper, de Villiers, Weissenborn, Goodluck and Roeper1992, p. 202)
However, Philip and de Villiers (Reference Philip, de Villiers and Clark1992) also show that other aspects of factivity itself are problematic for children in this age group. The authors argue that, at age 4, children fail to recognize the inferences of the verb itself. Consider the context in (16) next:
Question: Did Jim forget that his aunt was coming? (example taken from de Villiers, Reference de Villiers, Fletcher and McWhinney1995, p. 520)
When asked the question earlier, 4-year-old children answer “Yes,” unlike adults. The authors thus argue that only once children have acquired knowledge of these inferences, do they observe the barriers imposed on extraction.
In a later study that used a similar design, de Villiers et al. (Reference de Villiers, Curran, DeMunn, Philip, Greenhill, Hughes, Littlefield and Walsh1998) identify the age of 7 as the plausible cut-off point where knowledge of these properties of factive verbs is acquired. The authors test the availability of adjunct extraction out of clausal complements headed by the factive verbs forget and remember (alongside other mental verbs) and show that, at least in the case of forget, there is a clear age effect: the 8-year-old children in their study seem to observe the restriction on long-distance movement, while the 6-year-old children do not. However, the data coming from the verb remember are nowhere as clear, as even adults seem to allow extraction out of factive islands, contrary to the theoretical predictions. The authors account for this surprising finding by arguing that remember might have a dual nature, being used either factively or non-factively, depending on what the speaker chooses to emphasize.
Another potential explanation might be that, in such cases, the complement of the verb remember has been re-analysed as an extended event, identified as such with the event denoted by the predicate in the main clause.Footnote 7 Such an approach has been proposed in the literature by Truswell (Reference Truswell2007) to account for contexts such as the ones in (17), which showcase an apparent case of adjunct island violation.
Briefly, Truswell argues that, in (17), the event in the adjunct clause describes the event in the matrix clause and, consequently, it could be seen as being part of the argument structure of the verb in the matrix clause. Another key ingredient in Truswell’s analysis is telicity. According to him, extraction will only be possible if the event in the main clause is telic, while the event in the adjunct clause is atelic, giving rise to a contrast between the sentences in (18) next. In (18a), both the event in the matrix clause and the one in the adjunct clause are telic, whereas in (18b), the event in the matrix clause is telic, but the one in the subordinate clause is atelic.
Going back to the case of remember in the study of de Villiers et al. (Reference de Villiers, Curran, DeMunn, Philip, Greenhill, Hughes, Littlefield and Walsh1998), while such an extension could be possible for the event in the clausal complement of the verb remember, what is unclear is why a verb such as forget could not have undergone a similar reinterpretation. Both factive verbs used in the experiment are telic, and yet the results reported by the authors clearly show a difference between the two. Without access to the full set of stimuli used in the study – that could reveal any potential differences in telicity with respect to the events in the clausal complement – one can only speculate.Footnote 8
The acquisition of factive verbs (and their syntactic properties) has also been investigated by Schulz (Reference Schulz2003), who builds on the findings of de Villiers et al. (Reference de Villiers, Curran, DeMunn, Philip, Greenhill, Hughes, Littlefield and Walsh1998), as well as on data coming from a series of corpora analyses and comprehension tasks, and argues that the acquisition of factivity unfolds “stepwise, not in an all-or-nothing fashion.” In summary, children first need to distinguish between factive and non-factive verbs and the inferences that arise from this distinction. They then need to understand that factivity arises mainly compositionally, from the interaction of a factive verb with a (usually) finite complement, marked for tense and aspect. Once they have done that, they must learn that the proposition in the clausal complement is presupposed and true and that it refers to “an already established event in the discourse” (p. 113). In Schulz’s model, only once they realize that the factive complementizer binds the event variable, will they also master the syntactic properties that factive verbs evince. Following de Villiers et al. (Reference de Villiers, Curran, DeMunn, Philip, Greenhill, Hughes, Littlefield and Walsh1998), Schulz argued that, around the ages 7–8, children observe the island effects imposed by factive verbs.
In light of these findings, there is at least one question that still needs answering: do these findings apply across the board to all factive verbs or is there variation, as indicated by the study put forth by de Villiers et al. (Reference de Villiers, Curran, DeMunn, Philip, Greenhill, Hughes, Littlefield and Walsh1998)? Are there differences between sub-types of factive verbs? This question can also be seen as rooted in a much-discussed distinction between cognitive factive verbs (e.g., know, remember, forget) and emotive factive verbs (e.g., resent, be happy), the former having been shown to allow adjunct extraction more readily. Before I present an experimental study which addresses precisely this distinction, a small detour is needed in order to discuss some semantic and syntactic differences between these two sub-types of factive verbs.
3. Factive islands – a theoretical detour
3.1. Cognitive and emotive factive verbs
Ever since Karttunen (Reference Karttunen1971), a distinction has been made between factive verbs (e.g., regret, be sad) and semi-factives (e.g., know, realize), or, in more recent terminology, between emotive and cognitive factive verbs, respectively.
Cognitive and emotive factive verbs are said to differ both from the point of view of their meaning, and from that of their syntactic behaviour.Footnote 9 For instance, while factive verbs presuppose the truth value of their complement, this presupposition can be more readily cancelled in the case of cognitive factives, as seen in (19) next:
Another difference between the two sub-classes of factive verbs is related to their (in)ability to take a null complementizer: while cognitive factives seem to allow complementizer omission, emotive factives do not (Bîlbîie et al., Reference Bîlbîie, De La Fuente and Abeillé2023; de Cuba, Reference de Cuba2007).
What is, however, of interest for this article is that cognitive and emotive factives have been argued to differ from the point of view of extraction as well. More specifically, cognitive factive verbs seem to allow adjunct extraction, at least in some cases, as seen in (22) and (23) nextFootnote 10:
According to the authors, extraction is in fact possible when the information under discussion is new and not part of the common ground, or, in other words, when the verb does not behave like a factive verb. Alternatively, if the proposition is clearly part of the common ground, extraction is banned.
3.2. Factive islands in Romanian
Factive verbs give rise to weak island effects in Romanian as well, as illustrated in (24):
At first sight, in Romanian, cognitive factive verbs, however, seem to allow adjunct extraction – in an experimental task testing the barrierhood of manner of speaking verbs, where factive islands were used as distractors, a sentence such as the one in (25) was accepted by 72% of respondents.
Importantly though, this sentence was presented without any preceding context.
4. The acquisition of factive islands – an experimental study
4.1. Rationale and aim
To my knowledge, the acquisition of islands in Romanian in general has not been tested. Consequently, there are no data on the acquisition of factive islands either. There were several reasons behind this study: first, I wanted to expand on the work put forth by de Villiers et al. (Reference de Villiers, Curran, DeMunn, Philip, Greenhill, Hughes, Littlefield and Walsh1998) and to investigate the behaviour of a wider range of factive verbs. Not only did I wish to see whether the verb remember “a-și aminti” is truly different from other factive verbs, but also to investigate any potential differences between cognitive factive verbs and emotive factive verbs.
Second, as will be seen next, I also slightly altered the design of the experiment put forth by de Villiers et al. (Reference de Villiers, Curran, DeMunn, Philip, Greenhill, Hughes, Littlefield and Walsh1998). Their study was a typical instantiation of the Question after Story design, used extensively in the study of the acquisition of long-distance movement and island effects. Recall that in this design, children read/listen to a story, accompanied by pictures, at the end of which they are asked a question. The answer they provide can correspond to either a downstairs reading or to an upstairs reading. Importantly though, absence of evidence does not equal evidence of absence: if a child only provides a short-distance answer, does this mean they truly observe a linguistic restriction or is it simply because it might be less taxing from a processing perspective (a plausible alternative, given the short-distance movement bias mentioned earlier)?
Roeper and de Villiers (Reference Roeper, de Villiers, Weissenborn, Goodluck and Roeper1992) recognize the possibility of this issue and, in some instantiations of their experiments, they use a pragmatic bias in order to encourage a downstairs reading. In their study on interrogative islands, children did not provide answers corresponding to long-distance movement, not even when they were “encouraged” to do so, which led the authors to conclude that they indeed observe the restrictions which arise in those instances. To my understanding, in the study on factive islands, no such bias was used.
The aim of this study is therefore twofold. Broadly, I wanted to investigate whether monolingual Romanian children observe the restrictions imposed by factive verbs with respect to adjunct extraction out of the clausal complement and whether they make any difference between factive and non-factive verbs in that respect. The existing literature predicts that only older children (7-year-old children) will observe these restrictions, whereas younger children (5-year-old children) will not. Recall that the data are very clear with respect to the restrictions imposed by forget, but it seems to be less clear with respect to the behaviour of remember. Moreover, I was also interested in testing whether there is any distinction between cognitive and emotive factives with respect to the unavailability of adjunct extraction out of clausal complements in the case of either of the two age groups.
Method: The test was a comprehension task, modelled after the Question after Story design. The test crossed two main factors: type of verb (factive vs. non-factive) and distance (long-distance movement vs. short-distance movement). There were 16 test items, 8 of which targeted adjunct extraction out of the clausal complement of non-factive verbs (a crede “believe,” a se gândi “think”), and 8 tested adjunct extraction out of the clausal complement of factive verbs (a-și aduce aminte “remember,” a-și da seama “realize,” a afla “find out,” a descoperi “discover,” a se enerva “be annoyed,” a regreta “regret,” a se întrista “be sad,” a se bucura “be happy”). Of the 8 test items with factive verbs, 4 included cognitive factive verbs and 4 included emotive factive verbs. Când (when) and unde (where) extraction were tested.Footnote 11
The task had the following set up: children, together with a Mickey Mouse puppet, were invited to watch a series of illustrations on a screen and listen to a story. At the end of the story, the puppet was asked a question, which it answered, as seen in (26) next (and illustrated in Figure 1), for a question with a cognitive factive verb. The children were then asked to state whether the puppet had provided the right answer and, if not, to provide the correct alternative. The answers the puppet provided varied between a downstairs reading and an upstairs reading. The reason for which children judged Mickey’s answer, rather than directly providing their own, was motivated by the need to obtain a clearer picture on the level of acceptability of these structures (or lack thereof). Had children simply provided their own answer (and assuming this answer had been the one corresponding to the short-distance movement), it would have been impossible to tell whether this was a mere preference or truly a matter of (un)grammaticality.Footnote 12
Illustration of a story which tested extraction out of the clausal complement headed by a cognitive factive verb (example (26))
Question: Mickey, când și-a dat Maria seama că și-a pierdut ursulețul?
Answer: (a) Când a ajuns acasă (upstairs reading) (b) *Când se plimba prin parc. (downstairs reading)
This is Maria. Maria has a teddy bear named Bruno, whom she always keeps attached to her backpack. One day, when she was strolling in the park with her grandparents, the teddy bear fell off, but she didn’t notice. When she got home, she wanted to take Bruno to bed, but she couldn’t find him, so she said to herself: Oh, no, I’ve lost Bruno. She quickly ran to her Mom and asked her: “Mommy, what are we going to do?”
Question: Mickey, when did Maria realize she had lost her teddy bear?
Answer: (a) When she got home (upstairs reading) (b) *When she was strolling the park (downstairs reading)
4.2. Participants
Seventy typically developing monolingual Romanian children participated in the task. Participants were split into two distinct age groups: 35 five-year-old children (M = 5;9, SD = 1;10) and 35 seven-year-old children (M = 7;10, SD = 1;68). The results of three children were excluded from the analysis, as their answers suggested that they had not understood the task. Fifteen adults, native speakers of Romanian, also took part in the task, as a control group.
5. Results
5.1. Descriptive results
The results, summarized in Figure 2 next, show that, overall, Romanian children have difficulty in observing the restrictions imposed on adjunct extraction in the case of factive islands – 5-year-old children performed at chance level and, even though there were significant improvements with the 7-year-old group, at this age, they still do not showcase complete adult-like competence.
Percentages of acceptability per age group and condition.
As for the difference between cognitive and emotive factive verbs discussed earlier, no clear distinction arose. More specifically, as shown in Figure 3, the acceptability rates for extraction from the clausal complement of cognitive factive verbs were 45.92% and 31.51% for the 5-year-old and 7-year-old children, respectively. In contrast, long-distance movement from the clausal complement of emotive factive verbs yielded acceptability rates of 57.21% for the 5-year-old group and 25.08% for the 7-year-old group.
Percentages of acceptability of LDM per age and sub-class of factive verb.
Importantly though, there were differences between the verbs used in the study, as illustrated in Figure 4 next. While a more in-depth analysis of the differences between the individual verbs would require more data, note that, while the 5-year-old children were mostly at chance with respect to many of these verbs, for the 7-year-old group, remember stands out as (incorrectly) allowing long-distance movement out of its clausal complement at a higher rate. Recall that it is precisely this verb that also stood out in de Villiers et al.’s (Reference de Villiers, Curran, DeMunn, Philip, Greenhill, Hughes, Littlefield and Walsh1998) study. Importantly though, in their study, adults allowed extraction as well, whereas in this study, adult speakers of Romanian never allowed extraction in the case of remember. However, I leave for further research a closer look at the behaviour of this verb and the possibility of it being either associated with a non-factive verb or, as one of the reviewers suggested, it being interpreted as part of an extended event.
Percentages of acceptability of LDM per age and individual verb.
5.2. Statistical analysis
To test how children fared with respect to adjunct extraction out of the clausal complement of a factive verb in contrast to that of a non-factive one, I employed a logistic regression model (conducted in Jamovi (2021), using the gamlj module; Galluci, Reference Galluci2019), with Response as a dependent variable, the fixed effects Verb Type (factive vs. non-factive), Group (Children vs. Adults) and the interaction between Verb Type and Group. The model revealed significant effect of Verb Type (ß = 1.370, SE = 0.169, Z = 8.119, p < 0.05) and interaction between Verb Type and Age (ß = −0.573, SE = 0.210, Z = −2.728, p < 0.05), but no effect of Age individually (ß = −0.191, SE = 0.215, Z = −0.889, p = 0.3). A series of pairwise t tests further showed a significant difference between children and adults with respect to adjunct extraction from the clausal complement of a factive verb (p < 0.05), but not from that of a non-factive one (p = 1).
However, I wanted to further investigate any potential differences between the 5-year-old group and the 7-year-old group. To do so, I conducted a logistic regression on the child data with Response as the dependent variable, the fixed effects Verb Type (factive vs. non-factive), Age (5-year-old vs. 7-year-old), and the interaction between Verb Type and Age. The model revealed significant effect of Verb Type (ß = 0.773, SE = 0.184, Z = 4.20, p < 0.05), of Age (ß = 0.797, SE = 0.186, Z = 4.32, p < 0.05), and interaction between Verb Type and Age (ß = −0.823, SE = 0.159, Z = −5.17, p < 0.05).
A series of pairwise t tests further showed that while 7-year-old children evince more adult-like behaviour (p = 0.6), there are statistically significant differences with respect to the acceptability of adjunct long-distance movement out of the clausal complement of a factive verb between 5-year-old children and adults (p < 0.05).
I wanted to further see whether there is any statistically significant difference between cognitive and emotive factive verbs, with respect to adjunct extraction. I performed a logistic regression (conducted in Jamovi, using the gamlj module), with Response as a dependent variable, the fixed effects Class (cognitive factive vs. emotive factive), Age (Children vs. Adults), and the interaction between Class and Group. The model revealed significant effect of Age (ß = 2.503, SE = 0.606, Z = 4.131, p < 0.05) and interaction between Class and Age (ß = −3.027, SE = 0.608, Z = −4.975, p < 0.05), but no effect of Class itself (ß = 0.154, SE = 0.215, Z = 0.614, p = 0.5). In order to see whether the cognitive/emotive factive verb distinction played any role in the child data, I conducted one more logistic regression with Response as a dependent variable, the fixed effects Class (cognitive factive vs. emotive factive), Age (5-year-olds vs. 7-year-olds), and the interaction between Class and Age. The model once more revealed significant effect of Age (ß = 0.9815, SE = 0.264, Z = 3.719, p < 0.05) and interaction between Class and Age (ß = −0.9603, SE = 0.227, Z = −4.266, p < 0.05), but no effect of Class itself (ß = 0.09, SE = 0.264, Z = 0.345, p = 0.7). A series of pairwise t tests further showed that the distinction between cognitive and emotive factive verbs was not significant for the 5-year-old group (p = 1), nor for the 7-year-old one (p = 0.8).
6. Discussion
The results of the experimental task confirm the findings from previous studies: there is a difference between 5-year-old children and 7-year-old children, the latter exhibiting significantly more adult-like behaviour and rejecting adjunct extraction at a higher rate than the former. Seven-year-old children were also shown to distinguish between factive and non-factive verbs with respect to long-distance movement, readily allowing extraction in the case of the latter, but not of the former. However, at the age of 7, the barrierhood of factive verbs is still not fully mastered. As predicted, my data support the claim made by de Villiers et al. (Reference de Villiers, Curran, DeMunn, Philip, Greenhill, Hughes, Littlefield and Walsh1998) and Schulz (Reference Schulz2003), according to which the acquisition of the barrierhood of factive verbs is therefore a phenomenon that occurs late.Footnote 13
The data revealed no difference between cognitive and emotive factive verbs with respect to adjunct extraction from the clausal complement, neither overall nor in the two sub-groups of children. This is not surprising, as cognitive factive verbs are said to facilitate extraction only in those cases where the information is interpreted as not being presupposed (i.e., uttered out of the blue), as proposed by Djarv (Reference Djarv2019). In this task, the information was clearly presupposed and had been mentioned in the context.Footnote 14
In her dissertation, Schulz (Reference Schulz2003) argued that the acquisition of the (syntactic) properties of factive verbs takes place “on a verb-by-verb” basis and this study too showed differences between individual verbs with respect to the acceptability of adjunct extraction.Footnote 15
7. Consequences for linguistic theory
Ever since Ross’s (Reference Ross1967) seminal work, islands have been at the core of countless studies, many of which focus on one main question: what is the source of these island effects? Several directions of investigation have been explored in the theoretical literature with respect to islands in general and to factive islands in particular.
On the one hand, there are researchers who argue that these island effects can be accounted for in purely syntactic terms. Probably the most influential analysis of factive islands stems from Kiparsky and Kiparsky’s (Reference Kiparsky, Kiparsky, Steinberg and Jakobovits1971) work, where the authors argue that factive verbs have a more complex structure than non-factive verbs (think, believe, etc.), as they contain a silent noun fact in their structure, as illustrated in (27)Footnote 16:
In this line of analysis, the ban on extraction can be readily accounted for in terms of Chomsky (Reference Chomsky, Anderson and Kiparsky1973) – factive islands would be very similar to complex NP islands and movement would be blocked due to the moved constituent crossing more than one bounding node (NP and IP). However, one issue is that, under such an analysis, factive verbs are expected to give rise to strong island effects, contrary to the standard literature.
Factive islands have also been analysed from the point of view of information structure. Erteschik-Shir (Reference Erteschik-Shir1973) argued that extraction is only possible out of semantically dominant constituents, where a constituent is seen as semantically dominant either if it is not part of a presupposition, or if it not contextually determined. In the case of factive islands, the filler is part of a presupposition, long-distance movement being therefore banned. However, she also acknowledges certain exceptions, where extraction is, in fact, allowed, provided that the clause is interpreted as dominant (Erteschik-Shir, Reference Erteschik-Shir1973). This line of analysis would readily explain the surprising behaviour of remember, reported by de Villiers et al. (Reference de Villiers, Curran, DeMunn, Philip, Greenhill, Hughes, Littlefield and Walsh1998). It might be the case that, for the adults in their experiment, the clausal complement in those items was interpreted as “dominant.” However, as one of the reviewers pointed out, an important question that arises is whether dominance is in any way mirrored on the structural make-up of these sentences as well. One possible solution that was suggested was that the clausal complement might be reattached to an escape hatch, a configuration which could occur if it were extraposed. This would be an interesting avenue to explore, but it would require a more careful design of the stimuli, where dominance is purposefully factored in. I leave such an investigation for further research.
An account in a similar vein was put forth by Ambridge and Goldberg (Reference Ambridge and Goldberg2008), who argue that extraction is only possible out of focus domains. Backgrounded constituents, on the other hand, will serve as islands for extraction. Crucially though, the authors argue that, depending on how backgrounded or focused an environment is, one can expect a gradient of acceptability of extraction. For them, extraction is not syntactically ill-formed per se, but rather infelicitous.
The argument–adjunct asymmetry is however problematic for such accounts – they would predict that extraction is infelicitous, irrespective of the function of the filler. While this study focused solely on adjunct extraction, argument extraction is reported in the literature to be grammatical (Szabolcsi & den Dikken, Reference Szabolcsi, den Dikken, Cheng and Sybesma2003).Footnote 17
Importantly, such analyses argue that context can in fact facilitate extraction – depending on the context, non-bridge verbs (e.g., factive verbs) could be seen as less dominant, the focus being primarily on the subordinate clause and therefore a filler–gap dependency could be established.
Last, but not least, there are also analyses which explain factive island effects on account of the low(er) frequency of the verb itself (factive vs. non-factive verb), on the one hand, as well as the low(er) frequency of the construction (wh question vs. declarative sentence). In an experimental study on factive islands and manner of speaking verbs islands, Liu et al. (Reference Liu, Ryskin, Furtell and Gibson2022) reported effects of extraction and effects of verb frequency, but not an effect of interaction between these two factors. The authors thus claim that “there is no true island effect for such structures” – in other words, factive verbs themselves are not island inducers at all, but the unacceptability of structures such as (5) above stems from the fact that they are less common. However, in their two experiments, Liu et al. (Reference Liu, Ryskin, Furtell and Gibson2022) tested 24 and 48 verbs respectively (of which only six and eight verbs, respectively, were factive). However, Huang et al. (Reference Huang, Almeida and Sprouse2025) conducted an experimental study meant to investigate the behaviour of a “nearly exhaustive” list of verbs (484 verbs were analysed) with respect to extraction. The authors report statistically significant differences between bridge and non-bridge verbs (including factive verbs), contradicting therefore Liu et al.’s (Reference Liu, Ryskin, Furtell and Gibson2022) findings. According to their results, factive verbs give rise to island effects, though not of high magnitude. It is important to mention that the authors targeted argument extraction, not adjunct extraction, contradicting, once more, the standard view according to which factive islands are weak.
Each of these analyses, individually, seem to account for some facets of the behaviour of factive verbs with respect to extraction. However, each of them also leaves questions unanswered. Perhaps a more efficient approach would be to advocate for a layered analysis of factive verbs, where all these different factors contribute and give rise to their specific properties. The data from the acquisition of factive islands seem to support such a view. This study confirmed that factive islands are indeed a late acquired phenomenon.
If one were to account for the data on island acquisition in general purely in terms of syntactic constraints, then factive islands would stand out, as being, at least at first sight, acquired later than the other types of islands mentioned in this article. If, on the other hand, only a processing route were to be adopted, the data from this experiment still raise a challenge: taking into account the short-distance bias mentioned in the previous sections, it is surprising that the younger group of children accepted long-distance movement at a high rate. A possible explanation could be that the younger children consider the information in the embedded clause to be “the main point” (in the sense of Simons, Reference Simons2007) and give their answers accordingly. As this possibility was not specifically controlled for in this experiment, it remains a speculation for now, but something worth exploring in the future. While experimental confounds cannot and should not be excluded,Footnote 18 the data from this experiment seem to provide evidence in favour of the view put forth by de Villiers et al. (Reference de Villiers, Curran, DeMunn, Philip, Greenhill, Hughes, Littlefield and Walsh1998) and Schulz (Reference Schulz2003), according to which this late acquisition is the result of a more complex path of factivity itself.
8. Concluding remarks
This study showed that factive islands are a late acquired phenomenon – 5-year-old children quite freely allow long-distance movement out of the clausal complement of a factive verb. Not even at the age of 7 do we see a full mastery of the phenomenon, but, at this age, children do evince more adult-like behaviour. Additionally, the results suggest that there is no difference between cognitive factive verbs and emotive factive verbs in this respect.
This study therefore adds to the existing literature on late acquired phenomena, providing data from a language where, to my knowledge, the acquisition of islands has not yet been investigated.
Funding statement
This research was funded by ICUB (The Research Institute of the University of Bucharest), by means of the Young Researchers Grant.
Competing interests
The author declares none.
Open access
