Tense and aspect in English and Portuguese:
a contrastive semantical study
Diana Maria de Sousa Marques Pinto dos Santos
This page contains the translation into English of the Portuguese
extended abstract of my PhD dissertation which I needed to deliver at IST because the
dissertation itself was written in English... Oh well, I am used to
parallel texts by now. This was a preliminary version, but never got to be rewritten.
This thesis originated from the need to describe Portuguese so that it could be processed by computers.
The study of tense and aspect in Portuguese was chosen because it was related to the knowledge representation of time by Portuguese native speakers, and because it was expected to be useful both in narrative understanding and in the formalization of reasoning about actions in general.
In order to be able to make use of the huge literature on the subject
concerning English while studying Portuguese in its own, I decided to
make a contrastive study that focussed on the similarities and
differences between the two languages concerning temporal-aspectual
information. Such a task would make the present study, in addition, of
relevance for machine translation between the two languages, a subject
whose need I knew well the need for both theoretical and practical
studies concerning the pair English/Portuguese I was well aware of.
Given that my proficiency in English is quite far from native, the use
of original texts (and their translations) was absolutely
necessary. Bringing in corpora, however, even if I had studied
Portuguese alone, would have always been the case, since I consider their use a methodological must in language engineering.
This thesis encompasses work in areas which are traditionally
separate: semantics (one might actually say "computational
semantics"), translation, and corpus processing. One of the expected contributions of the present work is the integration of perspectives and methods from different disciplines.
In what follows, I describe first the contents of each chapter, and
then the main results of the thesis as a whole.
Organization of the dissertation
The thesis is divided in three parts, plus an introductory chapter
(chapter 1) and a small chapter of final remarks (chapter 15).
The first part (chapters 2 to 4) features the theoretical
background, trying to give a (partial) answer to the following broad
questions: What is a contrastive study? What is the meaning of tense
and aspect? How does one analyse translation in general (and
particular translations in particular)?
The second part (chapters 5 to 8) contains the actual proposals and its application to the description of Portuguese and its contrast with English. In particular, two computational models are suggested which are then employed to describe a large number of translation pairs in the corpus.
Finally, the third part (chapters 9 to 14) presents several detailed empirical studies done on the corpus, one about Portuguese only, the others concerning the contrast between the two languages.
Follows a description of each chapter, encosing in parentheses the number of the relevant sections:
- Chapter 1, Introduction, deals briefly with the
philosophical problems of using translation as semantic data (1.1),
introduces the tense and aspect subject matter (1.2) and presents my
view about Natural Language Processing (or Computational Linguistics)
(1.3). In addition, it describes succintly the corpus used, discussing
critically some of the options taken in its constitution
(1.4). Finally, the readers are made aware of the problems raised by
the need of presenting an English "translation" of the Portuguese
member of every translation pair displayed as example in the
dissertation, and my general policy on the matter is presented (1.6).
- Chapter 2, Languages in contrast, deals with the problem
of language comparison. There I argue that, in theoretical semantics
studies and natural language processing systems alike, the usual
assumption that languages express the same meaning by different means
is deeply misconceived (2.1). I describe shortly (2.2) three basic
perspectives on the contrast of two or more languages: the
universalist, the typologist and the relativist's views, illustrating
them in the tense and aspect domain. After criticising the two first
ones, I suggest and motivate a relativist methodology (2.5). I argue
that the fact that the two languages at stake belong to the same
language family (Indo-European) does not bear upon the methological
issue (2.4) and make a brief reference to the discipline of Contrastive Studies (2.3).
- Chapter 3 is called simply Translation. Its main goal is
to investigate how to compare two languages through their translation,
without making the assumption that translation
preserves meaning. The first part of the chapter contains an overview
to "meaning across languages" in machine translation (3.1) and general
linguistics (3.2). The second part is devoted to the question of
characterizing translation pairs: It investigates the translation
mismatch concept (3.3) and shows that, no matter how useful such a
concept may be in the context of machine translation, it is
unappropriate for the analysis of human translations, which are the
object of the present study. As an alternative, a typology of
translation pairs is suggested according to the degree of information
preservation in translation (3.4). Then, interference of the source
language in the translated texts - in the target language, thus -
translationese, is presented (3.5) and the difficult question of
translation quality is mentioned (3.6). A
discussion of some of the consequences of these
phenomena rounds up this second part: I state that most translation
pairs do not preserve meaning, and that it is hard to believe that the
analysis of (existing) human translations is easier than the
implementation of (future) machine translations, notwithsatnding the optimism of some researchers. Finally, the third and most important part of this chapter suggests the use of parallel (translated) corpora for the comparison and contrast of two languages with a novel argument (3.8.1), and sketches a general methodology for this kind of studies. Such a proposal constitutes a reflection on the research possibilities of parallel corpora; it does not, however, reflect the empirical studies described in the thesis, since it is a consequence of them. I conclude (3.9) stating that the only correct way of comparing two languages is by using independent translations and suggest that languages differ more in what they say than in what they can say.
- Chapter 4, about Tense and aspect semantics, gives some background in the matters that seem more relevant to me in the huge literature on the subject. It is meant to organize my personal opinions, not to give an unbiased overview. First (4.1), I distinguish three types of questions: definition issues, description / explanation issues, and computation issues. I discuss them in turn. Two main concerns are ontology (what one speaks about) (4.3.2 and 4.3.4) - I propose an irreducible distinction between events, states, and properties - and the question of vagueness (4.4.3). This I claim to be one of the essential properties of natural language as an information transmission device. For the impatient reader, I conclude (4.5) with a list of the more important points of the chapter.
- Chapter 5 presents A model for the description of tense and aspect and their translation. First, it includes my redefinition of the aspectual network proposed by Moens (1987) (5.1). The English tense and aspect system is then briefly described in that framework (5.2). Finally, a model for the description of translations is suggested which is based on the mapping of two different aspectual networks, the translation network (5.3). The core of these proposals is presented below.
- The whole of Chapter 6 is devoted to A sketch of the
Portuguese tense and aspect system. It contains, however, some
sections of more general character: I criticize the use of the
Vendlerian (1967) categories for languages other than English, while
defending Vendler's method to discover the underlying categories for
each language (6.1). Such method is followed in the chapter, arguing
(6.2 to 6.5) for three main aspectual categories in Portuguese:
Qualidades (properties), Estados (temporary states) and Acontecimentos
(events). The last ones are either seen as Mudanças or Obras
(Mudanças are events which have a result; Obras are events
which take time). Qualidades can be further distinguished by
linguistic criteria between Propriedades essenciais and Propriedades
sociais. Finally, other classes, of a compound nature, are identified
and discussed: Aquisições, Moradas, Séries,
Obra+Mudança, and Mudança+Obra. The problem of point of
view (perspectival aspect) and quantification, mentioned in Chapter 4
as generally relevant, are discussed and applied to Portuguese (6.6
and 6.7). Follows a description of some Portuguese tenses (6.8),
culminating with an aspectual network description (6.9). It should be noted that, although quantification phenomena have not been dealt with primarily in the thesis, a suggestion for extending the model in rder to encompass them is done here (6.9.2). The chapter ends with an explanation, based on the analysis of the Portuguese tense and aspect system proposed, of why the classes and tests of Vendler are not appropriate for Portuguese (6.10).
- Chapter 7, the core chapter of the dissertation, presents a significant sample of the problems that arise Contrasting English and Portuguese, divided in four classes:
Two translation networks - from English to Portuguese, and from Portuguese to English - are presented that summarize the cases of aspectual difference pointed out (7.3). Some limitations of the model are also discussed.
- the source language is aspectually vague regarding a distinction required by the target language (7.1);
- the source language is aspectually compact vis-à-vis the target language, i.e., the former links together information pieces that are conveyed separately by the latter;
- the contrast is due to differences in the temporal reference (7.4);
- different general preferences in the subjects expressed in each language.
- Chapter 8, in turn, presents some Formal considerations
divided in three parts. First (8.1) it discusses practical applications of the work described in the thesis, presenting an ideal translation browser after a short overview of the ones found in the literature, and suggesting several applications built on top of such a browser. Second (8.2), the aspectual and translation network concepts are investigated with a view to formalization. Two kinds of approach are followed and eventually compared: automata theory and the algebraic specification of abstract data types. The goal of these digressions is to make more precise the informal specification in the previous chapters. Furthermore, some properties of the translation relation as a mathematical relation are made explicit. Finally (8.3), I adapt Carlson's (1993) tense and aspect theory to the formalization of the two networks, analysing some problematic translation pairs presented in the previous chapter.
- Chapter 9 initiates the reporting of the empirical studies which preceded and fed the conclusions presented in the previous chapters. It describes the Preliminary studies: the preparation of the corpus and the annotation decisions (9.1), the gathering of detailed information about the translation of particular tenses and about the source of a particular tense in a translation (9.2). The first conclusions are briefly mentioned to motivate the sequence of studies that followed (9.3). To end this rather heterogeneous chapter, some coments on the shortcomings and advantages of the material are offered at the light of the subsequent work (9.4).
- Chapter 10, A detailed study of Imperfeito surveys the
first and most extensive study, which followed the gathering of the
quantitative results presented in the previous chapter. Imperfeito was
chosen by its frequency and because I believed that much of the
information it conveyed had no parallel in English. I used three ways
to investigate it:
To conclude, I discussed briefly other contrastive studies focussing on Imperfeito (10.3.3), before summarizing the most relevant findings and evaluate the whole process (10.4).
- Every clause in Imperfeito in a Portuguese subcorpus
(approximately a sixth of the whole corpus) was tagged with a set of
traditional labels such as EXT (temporal extension), HAB (habitual),
IND (indirect speech), PIT ("picturesque"), etc. The distribution and
co-occorrence of these classifications was then analysed and some generalizations suggested (10.2).
- The translation of Imperfeito into English was analysed: Every
translation displaying a tense other than simple past - the default -
was looked into (10.3.1). On the one hand, this study aimed at
validating the conclusions arived at during the monplingual analysis
of Portuguese. On the other hand, it was meant to pinpoint areas where special translation difficulties might arise.
- The study of the English tenses - other than the simple past - which were translated by Imperfeito was performed (10.3.2): this study was the dual of the previous one.
- Chapter 11 analyses Perception verbs in Portuguese and
English with a double motivation (11.1): Not only do these verbs
seem to be freely added or removed by translation, but they also
express perspective, a concept which is intimately related to
aspect. First, the correlation between the Imperfeito-Perfeto
dichotomy and the presence/absence of the English modal could is
investigated (11.2). Then, the properties of the clauses involving a
perception verb whose object describes an event are noted down
(11.3), and every addition of a perception verb is looked into
(11.4). I conclude trying to establish translation rules in the two
directions. I also note that, although physical perception is unarguably common to the speakers of any natural language, the way of conceptualizing it, and referring to it, is considerably different in different languages (11.5).
- Chapter 12 goes back to grammatical phenomena and concerns
The present perfect and its expression in Portuguese,
starting from a compact description of the English present perfect
(12.1) and the parallel between its monolingual description and the
description of the Portuguese particle já (12.2). At
first the possible causes of the absence of já +
Perfeito as an instance of the crosslinguistic perfect category in
Dahl's (1985) study are discussed (12.3), and wondered why it is
almost absent in the translated text (12.4). This leads to an
in-depth study of já (12.5), of the "extended now"
concept (12.6), and of the Portuguese aspectualizer acabar
de meaning "it was a short while ago" (12.7). The resulting
conclusion is that there is no translation of English "present
perfect" in its whole. Different parts of its global meaning have
different possibilities of a wholly satisfying translation in
Portuguese (12.8). And the following point is emphasized: there is a
real difference between parallel description of two linguistic
devices in different languages and its use in practice in those very
same languages (12.9).
- In the sequence of this last remark, Chapter 13 investigates
Pluperfects and simple tenses. The goal is to find out why
there are so many translations replacing one kind of tenses by the
other, as the preliminary quantitative overview shows (13.1). This
chapter is essentially descriptive; it analyses in detail the
translations into simple tenses (13.3) and those from simple tenses
into pluperfects (13.2), ending with a description of the most clear
tendencies (13.4). The main lesson to be learned here is that to
understand a language, one must look at the performance of the
language users - in this case translators. This complements in not
trivial ways a possible parallel description of the competence of the speakers of the two languages involved.
- Finally, Chapter 14 concentrates again on Imperfeito, now
focussing on its opposition with Perfeito, in order to clarify
The Imperfeito/Perfeito distinction and its expression in
English. The distribution of the two tenses with the verbs
ser and estar is studied first, revealing that the translation into English
does not preserve their original meaning
(14.1). Then, the translations of be in the simple past to
Perfeito are looked into, uncovering two interesting phenomena (14.2):
the addition of
inchoativity (expressing the starting of the state) and the addition of
perspective. As far as other verbs are concerned, a sample of 60 cases
of translation of Perfeito to simple past and 60 cases of translation
of Imperfeito to simple past are analysed (14.3.2), as well as 50
cases of simple past to Imperfeito and to Perfeito each
(14.4.1). Finally, after dealing with a few special cases in detail, some comments on the analysis of gains/losses by translation are offered (14.5).
- The very brief Chapter 15 contains only some Final
remarks. I insist that the contrast of two languages cannot be
based on the premise that they convey the same meaning (15.1). In
addition to describing the most obvious shortcomings and achievements
of the dissertation as a whole (15.2 and 15.3), the formalization of
natural language viewed as translation is suggested (15.4), and the separation between the study of time in artificial intelligence and natural language processing is questioned (15.5).
This dissertation puts forward several proposals, theoretical or
practical, as well as presents a set of hypotheses about the languages
studied and their translation. In what follows, I describe the ones I
see as most relevant, and simply provide pointers to the others.
The order of presentation was chosen for clarity - it is not to be
taken as a ranking of the several contributions.
Functions of the tense and aspect system
I claim that the tense and aspect system of a language has the following functions (or, alternatively, can be described relative to the following parameters):
Due to my goal of comparing translations with the corresponding source texts I focussed primarily on the three first items.
- The set of aspectual classes correspond to the categorization of the language's own reality. Such a set contains of necessity complex classes ("complex" meaning here that to one expression is associated more than one entity, temporal profile, etc., provided there is a systematic coherence among the several interpretations).
- The expression of viewpoint, whose rationale is to indicate the place of the observer relative to a situation, for some kinds of situations (consequently categorized as belonging to some aspectual classes).
- The expression of the localization of an event vis-à-vis a temporal axis with at most three distinct points.
- The indication of the relationship between this temporal axis and the speech point: identity, translation, or incommensurability.
- The assignment of truth values (or of "proposition-hood") to a linguistic expression.
- The expression of temporal quantification, that is, to count (once, a definite finite number, or an indefinite number of times) the occurrence(s) of a given situation.
The aspectual network concept and its application to English and Portuguese
As far as the description of one language's tense and aspect system is concerned, the model proposed to represent the aspectual classes is based on Moens's (1987) aspectual network, revised and extended in order to encompass other kinds of information. The most significant changes to the original formulation were the restriction of coercion - too powerful in Moens's system - and the reinterpretation of the model as essentially computational, i.e., without assuming Moens's nucleus as underlying model for the network nodes.
Taken into account these changes, the aspectual networks suggested
are as follows (NB! Note that I do not claim they are complete, i.e.,
there are elements which have aspectual character and which are not
The English aspectual network:
While for English I used the wide literature available on the subject, simpy correcting some minor points (5.2), most of the analyses of Portuguese were my own, although I have tried to read everything - very little, unfortunately - that has been written on the subject.
The main contribution of Chapter 6, which goes into detail into the
aspectual behaviour of Portuguese, is to show that to start from the
linguistic behaviour of a language produces better results and sheds
more light to its - unique - system than the application of models
conceived for other languages.
The Portuguese aspectual network:
The translation network concept and its application to the pair English-Portuguese
Another contribution of the thesis is a model for the description
of translation pairs in what concerns their aspectual content: the
translation network. This model is progressively refined and improved
along the thesis:
As this proposal is original, i.e., it is not built upon well-known others I could point to, I try to summarize it here.
- it is initially presentated in (5.3), motivated by the discussion of language contrast issues presented earlier
- it is extensively used in Chapter 7, featuring more than thirty partial networks for the contrast of the two languages (7.1 and 7.2)
- the changes to the source aspectual network required by
translation are systematized in (7.3)
- the formal properties of the translation relation which underlie the model are discussed (126.96.36.199)
A translation network puts into correspondence the two aspectual
networks of two different languages. It can, therefore, be described
in broad lines as a mapping of the source language nodes (that
represent aspectual classes) onto the target language nodes, as can be
seen in the next picture of an abstract translation network:
Such mapping is not in general one-to-one (it does not correspond to an injective nor onto function), given that two languages are not isomorphic in their categorization nor in the set of operators they employ.
Furthermore, the translator is "seeing" the source language through the target language spectacles (the translator is very often native speaker of the target language, in addition). This makes the three following changes to the source aspectual network necessary:
In what follows I illustrate the three cases presenting some (sets of)
translation pairs and corresponding partial translation networks. This
way I provide concrete examples of the kind of problems handled in the
dissertation as well as of the way they are dealt with.
- New categories and/or transitions are added due to the target language interpretation. These transitions, necessarily unmarked, were dubbed "transitions coerced by translation".
- New categories are added, refining those of the source language
and consequently unfolding the mapping. This phenomenon covers the
cases in which the source language handles in the same way situations
that are (objectively) different, while the target language keeps the
- New categories are added, involving their derivation history. This is meant to model the cases of sequences of categories (possibly involving linguistic operators) being translated in a way different from their final category in isolation.
In fact, these three phenomena make the "complete" aspectual networks (those merging all partial translation networks presented - see below) insufficiently expressive, allowing nevertheless a bird's eye view. They make it clear, moreover, that the general similarity of the two systems (take the existence of the two "parallel" classes Aquisições and acquisitions, for example) conceals the fact that, in practice, due to the lack of lexical equivalents and the more complex phenomena just described, particular translation pairs featuring parallel paths are quite rare.
- While the distinction between permanent and temporary states (or rather between Qualidades and Estados) plays an important role in Portuguese (6.2) it is rather marginal in English, where it is rarely linguistically signalled. To translate an English text into Portuguese it is thus frequent that the translator must induce such distinction - about which English is vague - in the English aspectual network (7.1.4).
The two following translation pairs illustrate different choices relative to the translation network of the previous figure, in which "T" stands for transition coerced by translation (in other words, by the inability of preserving the vagueness in the translation).
Beside him on a table was a small Oriental gong and a bowl of cigarettes.
ao lado, na banca de cabeceira, havia um pequeno tantã oriental e um maço de cigarros.
And he drank a little pulque and that was breakfast
Bebeu um pouco de pulque. E foi o seu pequeno almoço.
In the first example, the situation was interpreted as permanent;
in the second as temporary, circunstancial. Now any Portuguese speaker
can confirm that equally possible translations (although with
different meaning) are estava um pequeno tantã and (este) era o seu pequeno almoço, which depict the inverse choices.
- In Portuguese there is no aspectual difference between a change of place and a change of state: both are categorized as Mudanças. In English, on the contrary, a change of place presupposes a kind of movement, whose expression is aspectually quite distinct from English change of states (acomplishments or achievements) (188.8.131.52) as the following figure shows.
The following examples display several translations of Portuguese
This last case, contrary to the previous ones, does not represent a choice of the translator (as for example between accomplishment and achievement): It depends on the kind of Mudança. (Although there can be also choice between activity and achievement, as displayed in the figure and illustrated in the next example:
- Accomplishments (that is, events with a result and having a process associated):
e a voz tornou-se-lhe amarga para acrescentar: --
and his voice grew bitter as he added, "
- Achievements (that is, events corresponding to a change of state seen as punctual):
E no momento em que Deus, Ele e a Palavra se tornaram um só
And at the moment when God, He, and the Word became one
- Activities (that is, events without inherent result which develop homogeneously in time):
Depois, foi a uma mesa próxima
He then walked to a nearby table
O escravo, tão suavemente como entrara, saiu.
The slave slipped out as quietly as he had entered.
Passei entre as mesas empilhadas.
I walked among the stacks of tables.
E Marco Semprónio subiu os degraus e penetrou no palácio.
And Marcus Sempronius climbed the stairs and entered the palace.)
- Finally, an Obra in Mais que perfeito is frequently translated by an acquisition obtained from an achievement which expresses the initial change - corresponding thus roughly to the inception of the Obra - as the next figure shows:
Examples of this kind of translation are:
onde agora, como não onde o frade se distraíra, as pedras eram tão numerosas
where now, unlike when the friar had become distracted, stones were as numerous
que se interessara demasiado pela homenagem,
who had become too interested in the commemoration
where it is conspicuous that English expresses the beginning and
continuation of a situation that might continue until the now of the
narrative, while Portuguese describes a temporally limited situation,
which is "now" over.
The contrast with the translation of other Obras is thus striking; cf. for example:
O papá discursou?
Did you give a speech, Papa?
The translation network from English to Portuguese:
The translation network from Portuguese to English:
The importance of vagueness and compactness for natural language description
Yet another feature of this thesis is its claim that the concepts of vagueness and compactness are necessary to the description of natural language, both monolingually and contrastively.
In fact, I hold it to be an essential property of human languages that a linguistic expression can, on the one hand, be interpreted in many different ways depending on the context it occurs in, and, on the other hand, accept contexts where it is compatible with more than one interpretation - this is my definition of vagueness. Conversely, compactness consists of the possibility to convey together several pieces of information that could also be expressed separately.
These two concepts are, in addition, specially relevant to the contrast of two languages as pointed out for example in Keenan (1978) and popularized in the machine translation literature as "translation mismatches".
In fact my systematization - necessarily incomplete - of the differences between English and Portuguese in Chapter 7 is organized in turn of these two concepts, as well as my attempt to identify translationese in corpus-based studies (3.8.2). And, of course, the classification of translation pairs sketched as an alternative to the translation mismatches model is specially aware of such phenomena (3.4).
Furthermore, my definition of aspectual class identifies as the object of classification the relation among linguistic expressions and kinds of situations (semantic objects) (4.4.3). Therefore, expressions that correspond to more than one type of situation (i.e., systematically vague) are on a par with those univocally related to one type of situation.
A methodological argument for contrastive studies
This thesis also argues (3.8.1) in defence of the following: The best way to do contrastive studies is basing them in existing translations (or, more technically) in parallel corpora.
This because, although they almost never preserve the total
information stated in the source text, translations are actually a
compromise between the wish to transmit the source meaning and the
need to be idiomatic in the target language. For this reason it is
often the case that, the existence of a similar form notwithstanding,
the translator employs a form more congenial to the target language.
- In fact, out of the translation context, such form would most
probably be identified as "the translation" if the only source of
judgement were the bilingual intuition of the researcher!
One should keep in mind, however, the possibility of
translationese, i.e., the translator may still be insidiously
influenced by the source language style and produce what has been
aptly called a "third code".
This dissertation had as main goal the study of the semantics of tense
and aspect in Portuguese and English using as semantic data
translations in the two directions performed by independent parties - namely, professional translators whose goal was not the study of (the translation of) temporal phenomena.
Such a study required a model to describe the translation between the two different (linguistic) systems. Such a model was developed with a view to its computational implementation, as a powerful translation browser or a module of many possible computer-assisted translation systems
The contrast of the two systems done with the help of that model also led to the formulation of several hypotheses about the two languages, about their contrast, and about the semantic of tense and aspect in general. In addition, it provided some insights about translation practice and translation complexity.
Finally, it is hoped that the detailed description of the several
contrastive studies performed may offer other researchers a methodology - to be improved - and a point of departure for new and improved hypotheses on the subjects at stake.
- Carlson 1993
Carlson, Lauri. 1993. "Aspect". Unpublished ms., Helsinki University.
- Dahl 1985
Dahl, Östen. 1985. Tense and Aspect Systems. Basil Blackwell.
- Keenan 1978
Keenan, Edward L. 1978. "Some Logical Problems in Translation". F. Guenthner & M. Guenthner-Reutter (eds.), Meaning and Translation: Philosophical and Linguistic Approaches, Duckworth, pp.157-89.
- Moens 1987
Moens, Marc. 1987. "Tense, Aspect and Temporal Reference". PhD thesis,
- Vendler 1967
Vendler, Zeno. 1967. Linguistics in Philosophy. Cornell University Press.