1.1 The Method of Hatzivassiloglou and McKeown

In the paper¹ , the authors aimed at determining the adjective polarity without taking context into account. The output of their experiments consisted of sets of labeled adjectives. They have observed that for English, in almost 78% of the cases when two adjectives are connected by conjunctions and and or, the expression connects two adjectives of the same polarity. However, this rule does not hold for a conjunction but.

Based on the above mentioned observations, Hatzivassiloglou and McKeown created a set of the following rules:

The application of these simple rules led to an overall precision of almost 82%, while the second and third rule had only minor influence on the result, improving it only marginally by a couple of percent.

2.1 Data

Two main sources of the data have been exploited in our experiment, namely the corpus SYN2010² and a Czech Subjectivity Lexicon⁶ . SYN2010 is a large scale general corpus of Czech annotated at a morphological level. We have collected all relevant sentences from the corpus, namely those which contained positive and/or negative adjectives connected by a relevant conjunction.

2.2 Conjunctions under Investigation

The paper¹ primarily concentrates on English conjunctions and and but. These conjunctions correspond to Czech words a and ale. Therefore, it is natural that we have also started our analysis with these two “core” conjunctions. Moreover, we have added the conjunction nebo (corresponding to English or)¹, in order to get a complete picture of the two most frequently used logical conjunctions. The rules for the latter had to be modified for Czech due to grammatical rules which handle the use of this conjunction.

As a starting hypothesis we have adopted the same basic rules which were proposed for English:

Let us now further investigate particular conjunctions and discuss the observations discovered in the course of the analysis of corpus data.

2.3 Conjunction a (and)

If the conjunction a (and) connected two positive or two negative adjectives, it was in a vast majority of cases (96.7 %) a connection which could be analyzed as a normal conjunction , i.e. the expected behavior of the conjunction. For example:

Tvůj otec mě obdařil nádherným darem - tebou - krásnou a dokonalou dcerou. (Your father gave me a wonderful gift - you - a beautiful and perfect daughter.)

Table 1 The conjunction a (and) and the polarity of adjectives 

The remaining cases (in which the conjunction a (and) behaved differently) included the following features:

An interesting, although also very infrequent phenomenon (0.6%) has been the gradation of the identical adjective, for example:

Pokaždé, když se dostal do toho začarovaného kruhu, byly vyjeté koleje hlubší a hlubší. (Every time, when he got into that vicious circle, the beaten tracks were deeper and deeper.)

Even more interesting situation is observed in sentences in which a positive adjective is connected with a negative one - these are in fact the counterexamples to our presumption. This happened in 14% of cases.

If we search for a connection between a really negative and a really positive adjective (in a given context), we realize that in 9% of cases the use of a pair with opposite polarity is indicated lexically, by means of words like zároveň (at the same time), současně (simultaneously) or mezi (between) (for example: Na první pohled byla mŕura krásná a hrzná zároveň. (On the first sight, the moth was beautiful and horrifying at the same time.))

Let us also mention that in the sentences containing the word mezi (between), the adjectives³ are only used in metacontext, i.e. are not intentionally evaluative⁴. From the point of view of sentiment analysis it is important that the words which are only mentioned do not take part in the evaluation of sentence polarity. They are present only as auxiliary expressions describing certain reality which is further discussed subsequently. The adjectives appearing in phrases with mezi (between) were also in majority of cases opposites - speakers wanted to point out a contrast between some conditions in reality.⁵

2.4 Conjunction ale (but)

Our initial hypothesis concerning the conjunction ale (but) supposed that this conjunction connects two expressions (adjectives) of the opposite polarity. Unfortunately, there were 37% of sentences in which this conjunction connected adjectives of the same polarity, thus contradicting our hypothesis.

Table 2 Combination of adjectives by means of the conjunction ale (but) 

2.4.1 The conjunction ale (but) and the adjectives of opposite polarity

Firstly, let us survey the connections which conform to our hypothesis. Roughly one tenth of the cases contained a connection at different syntactic levels, as, e.g., Levou pěst si tiskl k rameni, jako by mu takovíhle šupáci byli odporní, ale pravou napřahoval k seznamovacímu pozdravu. ⁶ (He pushed his left fist to his shoulder, as if he these bastards were loathsome, but he was stretching the right one in the greeting.)

In less than 10% cases one of the adjectives has been negated, as, e.g., in the sentence: Není to sport agresivní, ale čestný. (It is not an aggresive sport, but a honest one.) These sentences usually contain some kind of correction - the speaker rejects the first adjective and tries to persuade the listeners that the reality is exactly opposite. These sentences will be discussed in more detail below. For the moment, it is important that they actually represent a minor issue.

Let us suppose that we are able to recognize the connection at a different syntactic level. In that case, we would find out that the remaining 88% of sentences actually conforms to our hypothesis, i.e. they combine positive and negative adjective without additional problems or conditions.

2.4.2 The conjunction ale (but) and the adjectives of the same polarity

The combination of the two adjectives of the same polarity seems to be much more interesting. These are actually the combinations which go against our hypothesis. Similarly as in the previous case, a substantial part of sentences (39%) contains combinations at different syntactic levels, as, e.g.: Kariéra, zpívání je nádherné, ale skutečný život je priorita. (The career, singing is beautiful, but a real life has a priority.)

The constructions, in which the combination has been lexically negated, therefore it actually contained two expressions of opposite polarity, constituted one fifth of all cases, for example: Nevyhrál lepší, ale šťastnější tým. (Not the better, but the happier team won.) Let us mention that these cases are actually, due to the application of negation, conforming to our hypothesis.

The rest, i.e. slightly more than a half of cases, consisted of connections which actually contradicted our hypothesis. Let us look at the example: Je to vlídné, přesto naléhavé, je to zábavné, ale poučné, je to cíleně dojemné, ale bez klišé. (It is kind, but still urgent, it is entertaining, but educational as well, it is intentionally touching, but without clichè.) It is important that although both adjectives have the same polarity, they are used in a certain contrast. In our example, there is the contrast between entertainment and education.

If the combination of two adjectives of the same polarity is preceded by the word nejen (not only), it is actually not an adversative relation, but a synthetic one. Both adjectives have the same polarity in these cases.

2.4.3 The conjunction ale (but) and a negation

In our corpus, 9% of detected combinations containing the conjunction ale (but) also contained a negated adjective.

Let us note that in the combination of two positive or two negative adjectives combined by a conjunction ale (but) it is not possible to use all possible pairs (some of them do not make sense from the semantic point of view). For example, it is impossible to say: Byla to hezká, ale krásná žena. (She was a a handsome but beautiful woman.)⁷ or: Prodavač byl nepoctivý, ale nepříjemný. (The shop assistant was dishonest, but unpleasant)⁸). In the case of connections containing one negated expression, we are actually in a situation in which we are connecting two phrases of the opposite polarity. We should therefore not be surprised that it is possible to say: Nebyla to hezká, ale krásná žena. (She was not a handsome, but beautiful woman.). In this way we can create sentences independently of how semantically close or distant are the adjectives used. In our example they both describe the same property, both are considered to be positive, but they express a different intensity (of beauty, in this case). Nevertheless, it makes sense to use them in the combination with the conjunction ale (but) and a negation. The main reason for such construction is the fact that it actually works as some kind of refinement, correction of the speaker, rectification etc.

Even more interesting is the opposite configuration, i.e. the connection of two adjectives of opposite polarity. If we negate one of them, we actually get two phrases of the same polarity. Surprisingly, in this case we can ignore the semantic constraints applied to the combination of two adjectives of the same polarity (as they were discussed above). We can say: Nebyla to ošklivá, ale krásná žena. (She was not an ugly, but beautiful woman.) - if we negate the word ošklivá (ugly), it acquires positive polarity and, at the same time, it describes a similar property as the word krásná (beautiful). The situation is similar to the one discussed above, but now it does not contradict anything. The reason seems to be an implicit message, carried by the negation, namely the correction or rectification of some fact. The question, whether it is this kind of connection, can be answered positively either in the case that the conjunction ale (but) can be substituted by a conjunction nýbrž (not ... but), or in the case that it is a combination of opposites. The whole phrase then carries the polarity of the second adjective.

All these observations actually have a negative consequence for sentiment analysis. If the combination containing the conjunction ale (but) also contains negation, it is impossible to determine whether the adjectives should have identical or opposite polarity.

If we limit our investigation only to combinations not containing the negation, the results would be only slightly better.⁹ Even worse, although we might consider two adjectives of the same polarity with one negation as two phrases of the opposite polarity, the improvement would also be only marginal.

2.5 Conjunction nebo (or) as a Copulative Conjunction

According to its function, conjunction nebo (or) can work both as a copulative or disjunctive conjunction.¹⁰ When used as a copulative conjunction, it connects expressions of the same polarity. In structures with disjunctive meaning, nebo (or) connects expressions of the opposite polarity and is preceded by no comma. The disjunctive meaning is marked by a comma preceding nebo (or). Based on this convention, the data were divided into two separately analyzed parts.

Within the first part of the data representing adjectives connected exclusively by nebo (or) without a comma, only half of the examples complied with the similar polarity hypothesis.

Table 3 Adjectives connected by nebo (or) without a comma 

2.5.1 Conjunction nebo (or) and Adjectives of the Same Polarity

If we further investigate structures from Table [nebo_bez_carky], we realize that in two thirds of the cases, the authors used conjunction nebo (or) to express alternative or even synonymity. Sometimes, it seemed that the author intended to correct him/herself or tried to find more suitable wording, e.g. (...) je stejně chyba, abyste se označil za špatného nebo hrozného. (... it is actually a mistake to mark yourself as bad or awful.)

2.5.2 Conjunction nebo (or) and Negation

However, it is interesting to analyze in detail a use of nebo (or) in connection with negation, see the following example:

These structures follow the scheme ¬(A∨B), which can be, according to de Morgan’s Laws, converted to ¬A∧¬B. In such sentences, Czech language can employ conjunction ani (nor). When applying this conjunction, we generate a correct sentence both in terms of logic and grammar¹¹.

Whereas conjunction ani (nor) can connect also phrases of the opposite polarity, for conjunction nebo (or) in the copulative function this possibility seems rather inappropriate.

2.5.3 Conjunction nebo (or) and Adjectives of the Opposite Polarity

As we have shown at the beginning of this section, a good half of the cases does not convey with our hypothesis. In these cases, conjunction nebo (or) connects adjectives of the opposite polarity. However, in only one tenth of the cases it expresses either synonymity or alternative choice, as in the sentence Totéž se stává, když jste rozrušená nebo zamilovaná. (It also happens when you are excited or lovesick.) 43% of the cases represent structures influenced by the lexical indicators, namely expressions (or phrases) like ať (už) (no matter how), jestli (if) and zda (whether) - for particular percentage see Table [nebo_kontrast_opacna]. It is also interesting that these phenomena mostly concern opposite couples. The writer usually announces his or her own ignorance, e.g. Zda chytřejší nebo hloupější, nevím. ) (Whether cleverer or dumber, I don’t know.). This also concerns utterances expressing an indirect question: Sám posuďte, jestli je lepší nebo horší. (It is up to you whether it is better or worse.) Often, the structures implied indifference, e.g. Příroda je slepá a zná jen jednu cestu - kupředu - a to bez ohledu na to, jestli je dobrá nebo špatná. ) (Nature is blind and it admits only one way - forward - regardless of whether it is good or bad.) In sentences with the conjunction ať (už) (no matter how) the indifferent meaning is quite clear: Ať už je ale technologie dobrá nebo špatná, jsme uzamčeni v jejím středu. (No matter how good or bad the technology is, we are locked in its center.) The important feature of these structures is that they do not contain evaluative adjectives. The adjectives are used not to describe a particular entity, but in the metacontext describing a potential situation/possible world. Similarly, yet not that unequivocally, behave also the words že (that) and jako (as).

Table 4 The types of contrast in combination of opposite polarity adjectives by means of the conjunction nebo (without a comma) 

In the rest of the cases, we found a contrast of the two adjectives, while the ratio of opposites and non-opposites was 52 to 11. That means that if we were able to distinguish opposite couples, there would remain one fifth of the cases in which it makes sense to connect adjectives of the opposite polarity by the conjunction nebo (or).

2.5.4 Conjunction nebo (or) as a Disjunctive Conjunction

Conjunction nebo (or) works as a negative conjunction if it connects two sentences / two constituents and is at the same time preceded by a comma. In logic, this phenomena would be called exclusive disjunction. In the article the authors do not take this opportunity into account.¹² However, we could use this feature to determine a hypothesis for nebo (or) as a disjunctive conjunction: we suppose that it connect adjectives of the opposite polarity.

This time, the results are clear - in 89 % of the cases conjunction nebo (or) connects parts of the opposite polarity. The cases with the same polarity concerned adjectives used to express contrast - even though they have the same polarity.

Table 5 Combination of adjectives by means of the conjunction nebo (or) (and a comma) 

However, we noticed more interesting situation when it came to the adjectives of the opposite polarity. First of all, in only 2% of the cases nebo (or) was not connecting opposites, as for example in the following sentence: Rodiče si často pletou závislost s úctou a mžou být laskaví, nebo přísní, úzkostliví, nebo sebevědomí, na tom nesejde. (Parents very often confuse addiction with respect and they can be kind or strict, anxious or confident, it does not matter.)

All the other cases concerned a connection including opposites, or a connection on different syntactic levels. The opposites were used in contrastive meaning mostly in questions, both direct (questions finished by a question mark) and indirect, in which we can find expressions like jestli (if) or zda (whether), see the following examples:

Je návrh zavrženého blobu-chobotnice na Letné krásný, nebo hnusný? (Is the drawing of the rejected blob-octopus at Letná nice or ugly?) Nebyla jsem si úplně jistá, jestli je to dobré, nebo špatné znamení. (I was not sure whether it is a good or a bad sign.)

These sentences contain no evaluation either.