1.1. Evolution of pluperfect semantics

This subsection describes the semantic evolution of the pluperfect forms. I first proposed and described the model below in Balla-Johnson (2020), building upon previous work on the compound tenses in Romance, especially that of Squartini and Bertinetto (2000).

The three structures with active indicative pluperfect semantics have several different aspectual interpretations available to them. These include: i) resultative-in-past, a stative reading where a resulting state of a prior action is described, ii) perfect-in-past, and iii) perfective-in-past. Examples of each are listed below in tokens (3)–(5). The latter two both represent eventive predicates, but they sometimes can be differentiated by an optional temporal adverb. A reference time adverbial indicates a perfect-in-past reading, while an event time adverbial a perfective-in-past (Squartini, 1999). All three interpretations were available for the -ra forms from their earliest occurrences in Romance.

The question of how exactly these semantic interpretations evolved in the pluperfect has not attracted nearly as much attention as has that of the semantic evolution of the CP and the simple past in Romance. The semantics of the compound structures of the CPP and the PA, however, must have progressed along a cline from resultative-in-past to perfective-in-past during the development of the medieval language. The equivalent structure whose auxiliary appears in the present (ex. Spanish yo he cantado ‘I have sung’) is referred to here as the compound past, rather than the term perfect, here reserved for a specific semantic category. Just as in the development of the CP, the CPP structures with HAVE/HOLD must have originally carried a resultative value and only later extended their aspectual interpretations to include eventive predicates (a process dubbed the perfect-to-perfective shift or aoristic drift described by Bybee et al., 1994; Squartini & Bertinetto, 2000, 2016, and many others). This latter category can be further subdivided into perfects and perfectives.

The progression of a Romance variety whose compound pluperfect form has completed this semantic shift (i.e., all except Galician, Asturian, and literary Portuguese) must have passed through the grammaticalization stages shown below in Figure 2.⁴ This progression was originally proposed in Balla-Johnson (2020), based upon the analogous development in the aoristic drift of the CP.

Figure 2

Aspectual development of the pluperfect.

As Adams (2013) asserts, the CP (and the CPP) structure in Latin was purely stative and not yet able to represent events. Indeed, in discussions of this structure, no attempt to treat the CPP (auxiliary in the imperfect) and the CP (auxiliary in the present) differently is made. In many works, the CP structure in Latin is even explained with examples that appear in the CPP. Thus, at the start of this process of grammaticalization, only the descendants of the Latin synthetic pluperfect—the -ra forms—could be used with eventive predicates (i.e., perfect-in-past/perfective-in-past aspect), and the compound structure was restricted to resultative-in-past interpretations (Stage 1). Then the CPP was extended in a limited context, able to represent an event with perfect-in-past aspect (Stage 2). It then expanded into perfective-in-past contexts (Stage 3). Once the CPP was generalized and established as the default, unmarked form, the -ra forms became more specialized and restricted (Stage 4). The indicative -ra forms were then lost entirely, leaving the CPP as the only option for the eventive uses (Stage 5). Once the CPP becomes the generalized eventive pluperfect, a new structure is needed to represent the resultative-in-past aspect. Varieties such as French and Spanish have turned to the structure of lexical verb meaning ‘to hold/grasp’ plus a participle to communicate this meaning. It is important to note that the chronological duration of each of these stages can differ greatly by variety.

Furthermore, due to the often-unclear division between the perfect-in-past and perfective-in-past (much hazier than the already complicated distinction between the perfect and the perfective) it is not clear that stages 2, 3, and 4 were necessary or present in each language. There must have existed variation of the type seen in stages 2–4, but a progression from stage 1 > 2 > 4 > 5 is just as possible as stage 1 > 3 > 5. Maintaining distinct morphological forms for each of these aspects may not be common cross-linguistically, but it has been shown to exist in 19th century literary Portuguese (Squartini, 1999) and in some Indo-Aryan varieties (Ashwini Deo, personal communication). The relevant fact for this paper is that there was a much more evident distinction between a resultative-in-past (namely, stative) construction and one of either a perfect-in-past or perfective-in-past (namely, eventive) construction.

Three example tokens from the Corpus do Português which represent each aspectual reading appear below in (3) – (5). Semantically, token (3) corresponds with resultative-in-past, token (4) perfect-in-past, and token (5) perfective-in-past. The CPP is used in (3) and (4), and the -ra form in (5).

N.B. Each of the three semantic aspects can appear with each of the three formal structures at different points in the history of Portuguese. For brevity I only provide three, rather than nine, examples here.

1.2. The Portuguese pluperfect

This section describes in turn each of the three most common forms which are, or have been used with, one of the above pluperfect aspects in Portuguese, beginning with the oldest, the -ra form, followed by the CPP, and finally the PA.

1.3. Summary of the Portuguese pluperfect

The -ra pluperfect is maintained in modern Galician, and still exists in Portuguese but is restricted to formal literary registers. The Portuguese PA was originally the most frequent compound form, but it was restricted and lost relatively early on. Portuguese has developed a CPP which still exhibits slight auxiliary variation, but generally prefers ter over haver. As in many other Romance varieties, there were conditional/modal interpretations of the -ra form, but in Portuguese these fell into disuse by the 19th century. The Portuguese -ra forms and the CPP were not exactly in complementary distribution, but rather could make use of different aspectual interpretations at different points in the history of the language. Finally, unlike the unique Portuguese CP, there is no iterative interpretation for the CPP.

3.1. Diachronic expansion of auxiliary ter

Auxiliary ter was generalized as the compound tense auxiliary over haver and ser. Figure 3 above shows this diachronic development (in the CPP) using adjusted token-per-million values calculated from random samples of 100 CPP tokens per century. This data shows the auxiliary ter was clearly the preferred auxiliary by the 16^th century.

Although that figure shows the CPP, the CP follows a similar trajectory. Ter is the most frequently preferred auxiliary by the 16th century, although diminished variation with haver is maintained in the CPP into the present (see Table 2).

Since it has been shown that the development of the compound tenses in Portuguese was indeed different than that of Spanish, let us turn to the question of how exactly they differ and what consequences this has had for the development of the aspectual interpretations of the CP and CPP in Portuguese. First, let us examine the chronology of the generalization of auxiliary ter over haver as a past auxiliary.

In order to generate a visual representation of the changing auxiliary preferences in Portuguese, a query was crafted in the Corpus do Português to return tokens of the CPP and the CP using the auxiliaries haver and ter. Ser was not examined. Given that the PA all but disappears and thus does not seem to participate in the change in auxiliary selection in Portuguese, it will not be examined here. Using the corpus query interface, I searched for the auxiliary followed immediately by the participle, with no separation between the two. It would have been possible to add separation using the collocates feature, but this would not return a century’s tokens which could be sampled randomly. Moreover, this was not found to be a problem because the vast majority of the tokens of a compound verb form appear in exactly that order with no intervening material. Excluding other structures may have missed some relevant tokens, but it also minimized the chance of false inclusions where the participle is not associated with the auxiliary (ex. when the participle and auxiliary are not in the same sentence). To illustrate this, Table 3 below shows the number of additional tokens returned when the query is extended up to four lexical slots in each direction for a frequent verb (tomar ‘take’) in the Portuguese CPP. The column labeled +1 represents the tokens which were included in the token pool and then sampled randomly for the investigations to follow.

In the 15th and 16th century there are some additional tokens found between two and four spaces to the right of the auxiliary, while the same is true for the 15th century to the left. These are not enough to have a large confounding effect on the overall trends shown by the token counts returned by the corpus in Figure 9. Note that the results of these searches are presented in the exact tokens-per-million, not adjusted tokens-per-million as was generated by the random sample examinations. This method of display was chosen due to the varied number of words included in each of the centuries (i.e., comparing raw token counts between the 2.8 million words of the 15th century and the 4.3 million of the 16th would be unhelpful).

Figure 9

Portuguese compound pluperfect auxiliary comparison in tokens-per-million from the 13th–19th century.

Several observations can be made from the data displayed here. First and most importantly, haver is used at a higher rate than ter in the 15th century, but the opposite is true in the 16th. It was in the 16th century that ter became the generalized auxiliary for the CPP. Continuing, the use of haver with the CPP is shown to rise after almost disappearing in the 16th century. Why there is so little haver CPP in the 16th century is unclear and may be a peculiarity of this specific corpus.

3.2. Auxiliary ter’s impact on the development of the Portuguese compound tenses

Whatever was the exact cause of this generalization of auxiliary ter, the fact that it maintained its full lexical meaning of ‘hold/possess’ into this period (and also into the present) led to an increase in the amount resultative (stative) interpretations of the CP and the CPP, while these related structures were used as perfects (and increasingly perfectives) in neighboring Spanish. This can be imperfectly shown at scale by examining the trends in participle–complement agreement with each auxiliary.

In order to do this, another series of queries were crafted and run in the Corpus do Português. The auxiliary followed by the past participle was queried, for both auxiliaries in the CP and the CPP for five centuries, beginning with the 14th.⁹ Each of these queries was entered into the chart search option, which returned in one group every token which fit the criteria (ex. haver in the present followed immediately by any past participle, etc.). Both the absolute number of tokens and the tokens-per-million value was recorded. Within the group of tokens, I used the random sample generator to obtain a 50-token sample, which were then extracted and examined thoroughly. This resulted in a total of 987 tokens used for the following analyses (the ter CP in the 14th century only returned 37 tokens. This was the only category with less than 50 tokens collected).

First, since the clearest agreement between participle and complement is able to be recognized when the participle appears in a form other than the masculine singular, Tables 4 and 5 show the breakdown of the participle morphology for each auxiliary in first the CP and then the CPP. Again, note that for the auxiliary ter in the 14th century, there were not 50 tokens to collect. Only 37 were gathered so the percent, not the token number should be used in comparing that century’s data with the others.

The results in Table 4 show very similar rates in the participle morphology of the Portuguese CP between the two auxiliaries. The largest discrepancy is eight percentage points (four tokens in the random sample of fifty) between the non-masculine singular participles in the 16th century. Ter is shown to appear slightly more with a masculine plural or feminine participle, which is more likely to be resultative as signaled by the agreement.

The differences between the participle morphology of the auxiliaries are slightly more pronounced in the CPP. The ter CPP in the 14th century appears with non-masculine singular participles at a higher rate than either the haver CPP or the ter CP. There is also a larger difference in the 16th century CPP than in the CP. In Table 5, ter is recorded to appear with non-masculine singular participles 16% more (8/50 tokens) than haver. This is at the time when the overall rates of the ter CPP is about 16 times higher than the haver CPP. Thereafter, the haver CPP is only found with a masculine singular participle, whereas ter shows uses with other participle numbers and genders, albeit rarely.

However, while the morphology of the participle can be used as a proxy for agreement, there is not an exact correspondence between the two. More specifically, not all participles which appear in the masculine singular do not agree with their complement – many do agree. Therefore, the question is posed: Within the contexts of the collected tokens where there was an explicit NP or pronoun with which the participle could agree in number and gender, how often was agreement found?

Using the same data, overt participle and complement agreement was examined. Shown below are the numbers of tokens, divided by the morphology of the participle and the auxiliary, which show overt agreement (as in example (20)) or overt non-agreement (21).

Tokens which were classified as ambiguous were excluded since it could not be clearly established if there was participle–complement agreement or not, e.g. token (22).

Table 6 shows the results for the CP, when the auxiliary appears in the present, and Table 7 does the same for the CPP, with the auxiliary in the imperfect. The tokens were recorded from random samples of 50 per tense, per auxiliary, per century. Due to the exclusion of the ambiguous tokens which showed neither overt agreement nor overt non-agreement, there is not a consistent number of tokens shown in each century.

These results show clearer trends. In general, overt non-agreement between the participle and an explicit complement began a century earlier with haver than with ter (14th and 15th c., respectively), and overt agreement between the participle and the complement persisted for about a century later in compound forms with ter rather than haver. This is demonstrated in Table 8 above where, in the 16th century, there were only three cases of agreement found in the CPP with haver (outside of the masculine singular), compared to 13 in the ter CPP. Next, cases of clear non-agreement do not clearly outnumber the cases of agreement until the 18th century. The actual cases of non-agreement are doubtlessly higher because it is impossible to establish with certainty when the number and gender of masculine singular participles cease to be determined by an overt masculine singular complement. For example, in (21) one cannot be sure if the participle simply appears in the masculine singular as the default form, or if it is controlled by the masculine singular relative pronoun ho que.

Finally, it is important to remember that these are artificially identically sized samples. Fifty tokens were collected for each auxiliary for each century for the sake of comparison. However, the raw results show: (i) the CP is much more frequent than the CPP, and (ii) auxiliary ter is much more frequent than haver in both the CP and the CPP after the year 1500. The trends seen in the tables, like for there to be more agreement (and in particular more non-masculine singular agreement) in ter than haver, become exacerbated when taking the overall frequencies of these auxiliaries into account (shown for the CPP in Figure 9). For example, in the 16th century, 13 out of the 50 possible CPP ter tokens show participle–complement agreement, while only 3 of the 50 haver tokens do. This is already a large difference in the samples (20%), but when applied to the actual number of occurrences per-million it becomes even clearer. The 6% of haver CPP tokens which show participle–complement agreement in the non-masculine singular equates to 1.69 tokens-per-million (.06 × 28.15 total haver CPP tokens-per-million), while the 26% agreement with the ter CPP translates into 161.05 tokens-per-million (.26 × 619.41 total ter CPP tokens-per-million). This 20% difference in the sampled tokens represents the ter CPP actually appearing with non-masculine singular participle–complement agreement almost one hundred times as frequently as the same agreement in the haver CPP. The values presented in Tables 6 and 7 are shown converted into the estimated tokens-per-million according to the method explained above, below in Figures 10 and 11. Due to the difficulty of knowing whether the cases of a masculine singular participle appearing with a masculine singular complement are true cases of agreement (described above), these tokens have been excluded. It bears repeating that the CP (Figure 10), especially in the later centuries is much more frequent than the CPP (Figure 11).

Figure 10

Estimated CP tokens-per-million showing agreement (non-masculine singular).

Figure 11

Estimated CPP tokens-per-million showing agreement (non-masculine singular).

These two figures show that as haver declines in frequency, it also declines in showing overt agreement between the participle and complement. In contrast, ter shows an increase in overall tokens showing this overt agreement between the 14th and 16th centuries. However, this agreement declines sharply and is much less frequent in the 17th century, and all but disappeared in the 18th. Therefore, as the compound forms with ter increase, so does the number of CPP tokens showing agreement. However, even though ter continues its rise in frequency from the 16th century on (shown in Figure 9), participle–complement agreement is quickly lost.

The logical question arises, is participle–complement agreement a reliable indication of a state of lesser grammaticalization? This is a tricky issue since there are varieties (such as French and standard Italian) where the CP and the CPP are clearly fully grammaticalized and yet there are still environments where participle–complement agreement is present. Therefore, recently an idea has been proposed that it is not the presence or lack of agreement (a synchronic notion) which signals advanced grammaticalization, but rather the loss of agreement (a diachronic notion) which can show progression towards a more grammaticalized state (Smith, 2022). In Portuguese, there is a loss of participle–complement agreement with auxiliary haver, but a retention of it for at least another century in constructions with ter. This retention of agreement, breaking the trend seen with haver and extant in the neighboring variety of Spanish, I propose can safely be taken as a signal of lesser grammaticalization.

Therefore, if one can accept that the participle–complement agreement is a proxy (albeit an imperfect one) for stative/resultative semantics of the compound forms, then the data presented above show clearly the generalization of the ter auxiliary did lead to a resultative interpretation of the CP and CPP being maintained longer (into the 17th century) than in other languages. This fact has affected the development of both the CP and the CPP in Portuguese, the former being the more extreme case. These will be examined in turn.

First of all, the development of the CPP in Portuguese was thoroughly treated in Sections 1 and 2. The Portuguese -ra forms persisted as a perfect-in-past longer than in Spanish and French. Then, after being restricted to the perfective-in-past, the -ra forms existed there in variation with the CPP. The restriction happened later in Portuguese than in Spanish. In Spanish, the CPP was expanding as the -ra forms were being used more and more frequently as a conditional or imperfect subjunctive. These two processes exerted both a push and pull effect which put pressure upon the aspectual system of the pluperfect to reorganize. Portuguese did not experience either of these pressures, at least at first. The -ra form did not overwhelmingly become a conditional/subjunctive, and the CPP was being pulled even further away from perfective-in-past uses due to the resultative interpretations associated with the new generalized auxiliary ter. Therefore, whereas languages like Spanish were experiencing two related pressures causing the CPP to become the sole eventive pluperfect, Portuguese lacked one of these (the default interpretation of -ra becoming a conditional/subjunctive) and was experiencing the exact opposite of the other (CPP used more with resultative-in-past meaning). Therefore, the CPP was not semantically extended all the way into the perfective-in-past category until much later than in Spanish and in French. This was, at least in part, due to the generalization of ter which reinforced a resultative reading in the 16th century. Finally, it was only recently that the use of the -ra forms as a pluperfect indicative has been restricted to formal written registers, leaving the CPP as the only unmarked form of the pluperfect indicative.

Second, the CP of Portuguese has developed in quite a unique way within Romance. Amaral and Howe (2009) offer the most thorough and convincing explanation for this innovation.

We claim that the resultative construction is the historical precursor of the [CP] in Portuguese. The emergence of the iterative interpretation of the [CP] requires that the construction ter + Past Participle entails the prior occurrence of the event denoted by the verb, as the [CP] in synchrony denotes event iteration over a time interval. As the prior occurrence of the event denoted by the verb becomes part of the encoded meaning of the form, we expect to see an expansion in the aspectual verb classes that may occur in the construction: the construction is no longer restricted to telic verbs, as was the case in the purely stative resultative construction. We argue that the aspectual shift from the entailment of existence of a single prior event to the entailment of existence of multiple prior events was induced by contexts in which the interpretation of the VP was compatible with both a single and multiple event interpretation (Amaral & Howe, 2009, p. 395, emphasis original).

In summary, according to Amaral and Howe (2009, 2012), this development of the iterative CP arose directly out of the resultative (not perfect) interpretation of the Portuguese CP construction. Therefore, the triumph of ter as the auxiliary of the CP and pluperfect yielded a higher proportion of resultative interpretations for the CP than in other languages (like Spanish). This persistence and reinforcement of the resultative interpretation provided the opportunity for a reanalysis into the modern iterative/durative perfect to take place. The generalization of ter and the development of the iterative CP are thus directly connected. The first change created conditions which allowed for the second to take place.

In order to support their proposal, Amaral and Howe (2009) examine participle agreement in particular as evidence of resultative constructions. Their results for agreement are reproduced below in Table 8. The construction type is classified as “eventive” if there was overt non-agreement between the participle and the complement, ‘resultative/stative’ if there overt agreement between the participle and the complement, and “ambiguous”.¹⁰ The final category contains tokens which have no overt complement or whose participle and complement both appear in the masculine singular (in which case it is unclear whether the participle is in its default form or if it is actually showing agreement with the complement).

These results show a large increase in the eventive interpretations of the CP structure, based on overt non-agreement between the participle and the complement. Using the same criteria, I consulted my random samples of imperfect ter + participle structure from the Corpus do Português in the same way. My own results, shown here in Table 9 and Figure 12 display the same general pattern for the CPP that Amaral and Howe found for the CP.

Figure 12

Portuguese ter CPP tokens by construction type from 14th–18th century.

Just as in Amaral and Howe (2009), the resultative/stative use decreases over time as the eventive uses increase. The 17^th century is when the eventive use of the ter CPP becomes more frequent than the stative¹¹ use. While logical, the documentation of this finding is an important piece in the picture of the Portuguese pluperfect.

This section has shown that the increase in frequency and generalization of ter as a compound form auxiliary led to a greater number of resultative interpretations, recognized by participle–complement agreement, of the CP and CPP than with haver. Quantitative evidence was provided, documenting a greater tendency for this participle–complement agreement in the CPP with ter than with haver until the 17th century. This larger proportion of resultatives was shown to affect the CP and the CPP in different ways. First, the CPP was delayed in its progression to become a generalized perfective-in-past. Second, the CP developed from the resultative interpretation into the modern iterative Portuguese CP (Amaral & Howe, 2009).

3.3. Markedness and Partial Blocking

It has been shown that the CPP with ter generalized over haver by the 16th century. The question remains: why would this generalization of ter lead to a strong persistence of the pluperfect indicative -ra forms? Namely, the haber CPP generalized in Spanish and the indicative -ra forms were mostly lost by the 15th/16th century. Why did this not happen contemporaneously in Portuguese? These different processes in Spanish and Portuguese will be addressed here by appealing to Markedness Theory.

This idea was originally developed by Roman Jakobson in a series of articles, but in particular Jakobson (1957/1971). Andrews (1990) describes the original Markedness Theory proposed by Jakobson (1957/1971) thus: “The marked term gives the statement of a property A; the unmarked term can be divided into two components: (1) a general meaning = non-statement of A; (2) a specific meaning = statement of non-A” (Andrews, 1990, p. 10). The general meaning use is shown in Figure 13 below.

Figure 13

Jakobson’s theory of markedness with general meaning (Andrews, 1990, p. 10).

Horn’s Iconicity Principle states that (un)marked forms are mapped with (un)marked meanings (Grønn, 2007, p. 17). Therefore, meaning A would usually be expressed with a marked form. When “two morphological forms compete for the same semantic space (due to partial overlap in meaning) the more specific form may block the application of the less specific form wherever it can apply” (Hollenbaugh, 2019). Namely the hearer believes that there must be a reason the speaker used the specific form rather than the more general form, that is, the hearer expects that wherever the speaker can use the more specific form, the speaker will use the more specific form.

In historical Portuguese, the marked meaning of a resultative state (rather than an event) was most often represented with the marked form ter + participle which agrees with complement (marked form due to a higher morphological complexity), corresponding to meaning A in Figure 13 above. The -ra form was unmarked since it had a more general (larger range of) meaning and less complex morphology, corresponding to meaning B in Figure 13.

This situation of variation between the marked CPP and the unmarked -ra forms in the Portuguese pluperfect led to a delay in the semantic expansion of the former. Since the CPP originally had a more specific meaning (stative rather than eventive) it would have been reinforced in these contexts and disfavored in more general eventive interpretations. Since the CPP had this specific resultative interpretation available, the -ra forms would have been used less and less to express the same meaning because of the more specific available CPP form. Eventually, once the resultative meaning had given way to the eventive interpretation (seen in the sharp decline in participle–complement agreement in the CPP from the 17th century on), the CPP and the -ra forms overlapped in their use. It was not until the -ra form became restricted to formal written genres (ostensibly due to extra-linguistic factors dealing with language ideology and standardization of Portuguese since Galician maintains the -ra form) that the CPP generalized as the perfective-in-past. This process is called a markedness shift, where the originally marked (usually innovative) variant swaps places with the originally unmarked (usually traditional) variant, in this case the -ra form (Andersen, 2001). Visualized as the meanings in Figure 13, the CPP began as meaning A and the -ra forms as meaning B. As speakers generalized the CPP, that form became the unmarked (meaning B), leaving the -ra forms to be more restricted and only used in specific contexts for specific reasons (meaning A).

The exact opposite situation was found in the Spanish pluperfect. The use of the -ra form as a pluperfect indicative became marked since the -ra form was much more frequently used as a conditional/subjunctive by the 15th century. At this point, if a hearer heard a speaker use the -ra form, they would be forced to decide which meaning was intended. No such cognitive load would follow hearing a speaker use a CPP. Thus, the CPP was unmarked and the “better” form (“more economic; more harmonic; more salient etc.” (Grønn, 2008, p. 122)) to express the pluperfect.

Therefore, we see a different development between the Spanish and Portuguese CPP because between the 14th and 16th centuries, the marked Spanish -ra forms were increasingly blocked from occurring with the indicative pluperfect meaning since there was a more transparent option available (the CPP). During this same period, the marked Portuguese CPP was blocked from occurring with the perfective-in-past meaning until it lost the preference for resultativity (shown in participle–complement agreement). Simply put, the Spanish CPP generalized more quickly because it was the unmarked pluperfect form, and the Portuguese CPP took longer to generalize because it was originally the marked pluperfect form.

3.4. Summary

This section addressed the question of what role the generalization of auxiliary ter played in the development of the Portuguese pluperfect. Quantitative data from the Corpus do Português was collected to show that the generalization of ter was well established in both the CP and the CPP by the 16th century. It was then empirically shown that there was more participle–complement agreement in the constructions with ter than with haver, indicating a greater proportion of stative (i.e., resultative), rather than eventive, verbal meanings. This increased use of the resultative affected the CP and the CPP in different ways.

This examination of historical Portuguese and how the development of the CP and CPP were affected by the generalization of the auxiliary ter have shown that the resultative interpretations were maintained longer than in languages where the CP and the CPP were mostly eventive (either perfect or perfective) early in their textual history. This one change has been shown to influence the semantic development of the CP (Amaral & Howe, 2009). This paper has presented evidence that this change likewise influenced the semantic development of the CPP in the following way: the Portuguese CPP (formed with auxiliaries ser and haver) was originally interpreted as a resultative-in-past. Then it gained eventive interpretations, alternating in some contexts with the -ra forms. However, at this point in the 16th century, the auxiliary ter was generalized, yielding more resultative-in-past interpretations. Gradually, the ter CPP disfavored this resultative interpretation, becoming more eventive, and it entered into more direct competition with the pluperfect indicative -ra forms. Finally, the ter CPP became the unmarked generalized pluperfect form, and the -ra forms were restricted into the few domains where they still exist in modern Portuguese. Thus, the unique development of the Portuguese pluperfect can be attributed directly to the generalization of auxiliary ter.