Comprehensible legal texts – utopia or a question of wording? On processing rephrased German court decisions

This paper presents a study on the comprehensibility of rephrased syntactic structures in German court decisions. While there are a number of studies using psycholinguistic methods to investigate the comprehensibility of original legal texts, we are not aware of any study looking into the effect resolving complex structures has on the comprehensibility. Our study combines three methodological steps. First, we analyse an annotated corpus of court decisions, press releases and newspaper reports on these decisions in order to detect those complex structures in the decisions which distinguish them from the other text types. Secondly, these structures are rephrased into two increasingly simple versions. Finally, all versions are subjected to a self paced reading experiment. The ﬁ ndings suggest that rephrasing greatly enhances the comprehensibility for the lay reader.


Introduction
In the course of the last thirty years we have seen many contributions to the question of how to make the legal language more comprehensible.In the Anglophone world the plain language movement is quite infl uential, in Germanic research -on which we will mainly concen trate here since we are concerned with the German language -the work of the interdisciplinary working group "Analyse der Juristischen Sprache" (cf.Rave et al. 1971) from the beginning of the 1970s on wards has to be mentioned and most recently the working group at the Berlin-Brandenburgische Akademie der Wissenschaften (cf.Lerch 2004a).Lex i cal and syntactic specifi cities of legal language are described such as nominal style, longer than average sentences which are said to contribute to the incomprehensibility of the language of law (cf.Wagner 1981, Oksaar 1988).In this paper we concentrate on the language of court decisions whose linguistic characteristics are de scribed in an example-based way by Altehenger (1983), contrasting German and Danish decisions by Engberg (1997) and using corpus-linguistic meth ods by Hansen-Schirra & Neumann (2004).This latter study confi rms the specifi cities of longer sentences, nominalisations and highly complex noun phrases.These syntactic elements previously only described on an exemplary basis were analysed in a quantitative way in com pa rison to a reference corpus consisting of samples from 15 text types.
Building on work like this, several studies investigated legal language with the help of methods for researching comprehensibility.Basedow (1999) reports on a word-based study using the Flesch-test (Flesch 1948).This readability formula developed in the 1940s allows quantifying criteria like word and sentence length.Nevertheless, the read ability approach is problematic because it is limited to counting syllables and words per sentence and completely ignores the semantic content of the text (a short sentence may still be diffi cult to understand because it consists of short, but rare words).In the framework of research on comprehensibility it is therefore regarded as obsolete (cf.Rickheit 1995; see also Lerch 2004b for a review of the readability approach for measuring the comprehensibility of the law).
The above mentioned working group "Language and Law" at the Berlin-Brandenburgische Akademie der Wissenschaften did a psycholin guistic experiment aiming at the investigation of the general knowledge background of the participant.The complex research design used various methods like Thinking Aloud Protocols to look into the transparency rule in the German legislation on standard terms and conditions.It investigated deeper coherence structures in standard terms and conditions of insurance contracts.Unfortunately, the results of this highly interesting study are not published yet.Therefore it is not possible to draw any consequences from the fi ndings.Neumann & Hansen-Schirra (2004) used an acceptability judgment experiment (Gernsbacher 1994) in a pilot study to test the comprehensibility of syntactic peculiarities in German court decisions.The German verb mood Konjunktiv1 was chosen as a register-specifi c feature of court decisions as described by Altehenger (1983).This pilot study showed the use of the combination of corpus-linguistic methods establishing the grammatical characteristics of a given register with psycholin guistic experiments testing the comprehensibility of the specifi c features.The described results served as a basis for the research discussed in the present paper.
So far, we have discussed legal language as such.However, this is a cover term which has to be specifi ed for our purposes.We have to distin guish between at least legislative texts, administrative texts and court decisions.These text types differ in view of the authors, the addressees as well as the function of the texts.Jaspersen (1998) discusses the com prehensibility of legislative texts.There are a number of linguistic guidebooks for German administrations (cf."Bürgernahe Ver wal tungssprache" by the Bundesverwaltungsamt2 and publications re sult ing from the project "Bürgerfreundliche Verwaltungs sprache"3 ).Fur thermore, Grönert (2004) reports on research geared towards improv ing the com muni cation between administrations and citizens.
Court decisions are written by jurists and are not addressed at a clearly identifi able recipient but -apart from the parties involved in the case -rather at unspecifi c recipients subdivided into two groups.Court decisions are addressed both at legal experts who work with the texts and at the lay citizen who is supposed to accept the decision and abide by it.The needs of the two recipient groups partly diverge: Experts expect to be supplied with concise and precise information effi ciently packed into typical terms and syntactic constructions.On the other side, the lacking specialised knowledge of citizens not legally trained requires a more elaborate -in the view of the expert probably lengthy -presentation of informa tion.As this type of legal texts particularly draws public atten tion (cf.Jaspersen 1998) it should be com prehensible to a large group of recipients, i.e. both, experts and citizens.For this reason, our main interest lies in investigating the comprehensibility of court decisions by tracking down their linguistic complexity.
Our study focuses on how readers process syntactic specifi cities found to be typical of German court decisions.The aim of the study is to quantify syntactic specifi cities of German court decisions, to rephrase the complex structures into simplifi ed versions and fi nally to test these version psycholingu istically.Consequently, the research setup com bines corpus-linguistic and psycholinguistic methods against the back ground of legal knowledge.This interdisciplinary research design is refl ected in the authors' affi liations with not only linguistics but also law and psychology.
The remainder of the paper is organised as follows.In the following section 2, we will present the design of the study with the preparatory steps for the psycholinguistic test, namely the corpus-linguistic analysis of the texts as well as the process of rephrasing three types of syntactic specifi cities conforming to the legal content of the original wording.In section 3, we will present the psycholinguistic experiment in detail and discuss the results.Finally, we draw conclusions from this study and look at consequences for future research in section 4.

Corpus design
As mentioned above, the fi rst step to analysing the syntactic complexity of German legalese consists in a corpus-linguistic analysis.We intend to identify and quantify the syntactic features causing this complexity.For this purpose, we compare intralingual versions of German legal texts.Hence, our corpus of investigation consists of decisions of the German Federal Constitutional Court and press releases and newspaper reports on these decisions.One main interest for looking into the three versions resides in the assumption that, serving different purposes, they may display different linguistic properties, in particular, that they vary in their syntactic complexity: We assume the decisions to be the most complex version as they contain language for specifi c purposes.They con stitute a written version of the oral pronouncement in court.They display a complexity of legal content in combination with lin guistic complexity that may manifest on every linguistic level.
The newspaper reports are assumed to be the version displaying the least complex syntactic structures.They are expected to exhibit more general language features as they are designed to be the citizen's everyday reading.Further on, we expect the press releases to be of medium linguistic complexity as this version may be infl uenced by both, court decisions and newspaper reports: On the one hand, as they were written by legal experts, the press releases may contain specifi c features of German legalese.On the other hand, as they are considered to be rephrased variants of the court decisions for specialised journalists they may show general language features of newspaper texts.

Research design
Complex grammatical structures are major indicators of the linguistic complexity of texts.Therefore, we focus on the empirical analysis of the syntactic properties of the above mentioned intralingual versions of German legal texts.We apply a range of corpus-linguistic methods to verify and intensify the hypothesis about properties of legal and/or administrative texts stated in example-based studies (cf.Wagner 1981 for administrative texts and Altehenger 1983 for court decisions).
We intend to gain quantitative results about the syntactic properties respons ible for the varying degrees of complexity in the three intralingual versions.In addition, the annotation results are supposed to refl ect which features on which syntactic level are unique to the court decisions and which features are modifi ed or deleted in the press releases and newspaper reports respectively.We interpret syntactic complexity by investigating the following features: -Sentence length -Embedding in sentences -Length of noun phrases and prepositional phrases -Embedding of noun phrases and prepositional phrases -Nominalisations For the purpose of obtaining more information about the syntactic complexity on sentence and phrase level, we use the notion of fi eld topology which was especially developed for investigating the structure of the German sentence.It constitutes a relatively theory-neutral description of German syntax and considers its fl exible constituent structure.The German sentence is split up into smaller parts with regard to the distribu tional properties of the verb complex -so called topological fi elds: 'Vorfeld', 'linke Satzklammer' (fi nite part of the verb complex), 'Mittelfeld', 'rechte Satzklammer' (non-fi nite part of the verb complex) and 'Nachfeld'.
We combine automatic parsing and fi ne-grained manual annotation to get detailed information about the linguistic complexity on the syn tactic levels of the three versions.We analyse sentences from the three corpora using a topological parser (Braun 1999).It is based on the notion of fi eld topology for German syntax as mentioned above.The parser struc tures the sentence into a series of neighbouring and embedded topolo gical fi elds.The noun phrases and prepositional phrases are analysed man ually using an XML editor.For practical reasons, only the 'Vorfeld' is annotated: The 'Vorfeld' constitutes the leftmost part of the German sen tence, in front of the fi nite verb, and is expected to contain mainly noun and prepositional phrases.Furthermore, we annotate only phrases which have a minimum length of seven tokens (newspaper reports) or ten tokens (decisions and press releases) as we expect that phrases of this length already have a certain complexity.Both the automatic output of the parser and the output of the manual annotation are doublechecked by the annotators.
We query nominalisations with the help of a concordance tool which displays all matching constructions.We count all tokens which are classifi ed as nouns by a part-of-speech tagger and which contain the suffi xes "-ung", "-ion", "-ismus", "-heit", "-keit", "-ität", "-schaft" as well as the respective plural forms.

Findings
In the following, we present some of the results of our corpus-linguistic analysis before we move on to a general interpretation.We start by looking into the syntactic complexity on sentence level.Considering the average number of tokens per sentence in the three subcorpora we fi nd longer sentences in the court decisions than in the press releases.The newspaper reports by far have the shortest sentences.In addition, more than every second sentence in the decisions also contains a subordinate clause, whereas in the press releases every second, and in the newspaper reports every third sentence contains a subordinate clause.The higher number of subordinate clauses in the court decisions indicates that the sentences contain more embedding than the press releases, and even far more embedded clauses than the newspaper reports.This tendency can be confi rmed when looking at the percentage of embedded clauses on the respective levels of embedding.

Decisions
The lower percentage of clauses on Level 0 in the court decisions tells us that there are more clauses to be found on deeper levels of embedding.
In most instances, the percentage of embedded subordinate clauses on deeper levels is higher in the court decisions than in the press releases, and the newspaper reports clearly contain less embedded clauses on the deeper levels.Generally speaking, the fi ndings from our analysis of the sentences refl ect an extreme syntactic complexity to be found in the court decisions.The newspaper reports show a tendency towards much simpler sentence constructions, whereas the press releases are of medium complexity on sentence levels.We can conclude that extreme syntactic complexity on sentence level is one prominent feature of court decisions, especially compared to newspaper reports.
We now consider the syntactic complexity on phrase level.Comparing the length of phrases in the three intralingual versions as shown in Table 2, we see that the court decisions have the longest phrases and that the newspaper reports have the shortest phrases.The aver age phrase length in the press releases lies between the other two ver sions.
In order to obtain the depth of embedding, we count all head nouns on the respective levels of embed ding.The results for the newspaper re ports correspond to our expectations: Most of the noun phrases and pre po sitional phrases appear on level 0. On level 1 the percentage of em bed ded phrases is much higher than on level 2 and level 3. Apart from that, they contain only four levels of embedding, whereas the press releases have fi ve levels and the court decisions even six levels of embedding.Comparing the press releases and the court decisions, two peculiarities have to be noted: First, the percentage of phrases on level 0 in the press releases is lower than in the court decisions.Second, on level 1, the percentage of embedded phrases is higher in the press re leases than in the decisions and the difference from level 0 to 1 is even bigger than in the decisions.This may be due to terminology and para graphs in the court decisions being explained in the press releases by using embedded phrases.Apart from that, the higher percentage of phrases on level two in the decisions and press releases may be caused by coordinations within phrases.This tendency towards organising phrases in extremely compact structures may result from the need to pre sent the information in a concise way: Explanations and further informa tion on the court decisions may be packed into phrases rather than being realised by verbal constructions which would require more space.It thus leads to a high nominal density.
Finally, the amount of nominalisations as deverbal derivations to be found in each intralingual version is also an indicator for more or less syntactic complexity.

Decisions
Press releases Newspaper reports All nominalisations 7.15 % 7.30 % 4.54 % "ung"-nominalisations 5.31 % 5.56 % 3.30 % Table 3: amount of nominalisations Table 3 shows that 7.15 % of all nouns in the court decisions are no mina lisations.The percentage of nominalisations in the press releases is even higher (7.30 %).This higher proportion of nominalisations may also be an indicator of the above described tendency to compress information: press releases have to inform the recipient in a very con dens ed way.The newspaper reports exhibit a much lower amount of nomi nalisations (4.54 %) than the two other intralingual versions.
Our data clearly show that derivations on "-ung" are the most frequent German nominalisations in all three intralingual versions.For this reason, we concentrate on these forms when elaborating rephrased ver sions of nominalisations as will be explained in section 2.2.
Summarising our fi ndings, we can say that the annotation results mostly confi rm our hypotheses concerning the syntactic complexity of German legal texts.Comparing the court decisions, press releases and newspaper reports the following important tendencies have to be outlined: -The court decisions display more complexity on most syntactic levels than the press releases and the reports: They have the longest sen tences, they contain the highest number of subordinate clauses and the highest depth of embedding.They also contain the longest noun phrases and prepositional phrases.Furthermore, on most lev-els, the decisions show more embedded phrases than the press releases and the newspaper reports and they contain a high number of nominalisations.We assume that this high complexity on all syntac tic levels has a negative effect on the comprehensibility of the court decisions.It complicates processing and risks destructing the recipient's capacity of storing information in the short term memory.
-The press releases' complexity lies between that of the court decisions and the newspaper reports.In some cases, they approximate the court decisions or are even more complex: They contain more embedded noun phrases on level two and have slightly more nomi nali sations than the court decisions.These factors indi cate a tendency to informational density similar to that in the court de cisions: Information is packed into heavy noun phrases and no mi nalisations rather than being distributed onto larger gram ma tic al units like sentences.Therefore, we assume that the syntac tic struc tures in the press releases, on most levels, are still too complex for lay per sons.And this complexity still affects the reception process.For this reason, the press releases cannot be used as a starting point to resolve the complexity of the court decisions.-The newspaper reports are less complex than the court decisions, and are still clearly less complex than the press releases on all investigated syntactic levels.As stated in our assump tions above, they mostly contain general language features with which lay persons as readers are confronted every day.We assume that the newspaper reports are much easier to process than the two other intralingual versions.There fore, they can serve as a yardstick for elabo rating methods to rephrase the court decisions, as will be explained in the following section.

Rephrasing register-specifi c syntactic constructions
For the rephrases, it is necessary to restrict the number of syntactic fea tures as well as the number of versions to be elaborated.We thus con centrate on the three syntactic features of deeply embedded sentences (abbreviated "S"), deeply embedded phrases ("P") and "-ung"nominalisations ("N").With a view to the feasibility of the psycho lingui stic experiment, we restrict ourselves to two rephrases (versions B and C) of the original version A retrieved from the corpus of court decisions.Our goal is to work out three degrees of complexity: a highly complex version A, a medium complex version B and a simple version C. All versions have to have the same legal contents.This is guaranteed by the interdisciplinary team consisting of both jurists and linguists.
Facing the highly complex structures of the court decisions, the question arises how to delimit the rephrased versions.The medium complex version follows the idea of an optimal approach to comprehensibility as proposed by Groeben & Christmann (1989).This approach claims that texts which conform to the reader's expectations -for instance by means of maximally simple structures (see below) -do not offer any cognitive stimulus to the reader.Extremely simplifi ed texts are said to destruct the recipient's motivation to keep reading (cf.Groeben & Christmann 1989:175).We apply this optimal approach to comprehensibility to the medium version by taking the corpus results of the newspaper reports as a benchmark.As previously mentioned, we assume that newspapers use a range of language well adapted to the reading habits of the lay reader4 -or, to put it the other way round -newspapers form their reading habits.Thus, structures typical for newspaper reports should be familiar to lay readers of court decisions and still challenge them enough to pay attention to the unfolding text.
The simple version C realises a maximum strategy: the complex sentence, for instance, is broken down to one clause per sentence.This corresponds to the maximum approach to comprehensibility as advocated by Langer et al. (1974).They report on a study in which experts score texts in view of the four dimensions 'linguistic simplicity', 'structureorganisation', 'brevity-shortness' and 'interest-liveliness'.They argue that the higher a text is scored with respect to these dimensions the better the text will be memorised.We apply their dimension of simplicity to the three syntactic features under investigation in our study.

The rephrased syntactic dimensions in detail
As shown in the corpus analysis, the authors of court decisions make use of more complex and embedded sentence structures than the au-thors of the other text types in the corpus.Initially, those structures were rephrased which displayed the highest level of embedding on the sentence level.They turned out to be rather exceptional examples with non-representative peculiarities, so we broadened the scope for extract ing sentences from the corpus to slightly less striking examples.
Example 1 shows the rephrasing process for the syntactic dimension S, with 1a showing the original sentence (version A), 1b the medium version B and 1c the maximum version C: As this example already shows, the logical relation within the sen tence poses a problem for the rephrasing process.Transferring the intrasenten tial logical relation to a sequence of logically related sentences may in volve major restructuring, because sentence splitting can lead to ambiguous intersentential reference and to a loss of textual coherence.
In Example 1 restructuring is realised by postponing the clause containing the postmodifi cation of "Einzelfallprüfung" to the end of the newly created sequence and adding an introductory anaphoric prepositional phrase.
Embedding on the phrase level is resolved fi rst by transferring a nominal structure into a verbal structure.This is typically realised by expressing the nominal meaning in a subordinate clause with the formerly only nominal meaning distributed on nominal and verbal elements.For our rephrasing process this means that version B is more complex on the sentence level than version A. Therefore, the second re phrase leading to version C consists in a break up of the complex sentence structure as for the embedded sentences (see above With respect to nominalisations, we concentrate on deverbal derivations with the suffi x "-ung", as they have the highest frequency of all nomi nalisations in the three intralingual versions (see section 2.1).The high frequency of this feature, however, raises the question, which instances we want to rephrase.If we rephrased all instances, the sentence structure would become overly complex.Furthermore, this would not match the frequency in the newspaper reports.We therefore have to make a selection.One should think that nominalisations created on an ad-hoc basis to condense clausal meaning into a nominal structure qualify for rephrases.However, most instances of nominalisations are lexicalised through frequent use.Often both, the nominal and the verbal form of the lemma, are equally frequent.The most plausible candidates therefore are accumulations of "-ung"-nominalisations within the same noun phrase construction.The fi rst rephrasing step consists in using the verbal form of the lemma, thus creating a clausal structure. In

Limitations of the rephrasing process
Originally, we had expected to be able to develop rules for rephrasing each syntactic dimension.However, the instances proved too hetero gene ous to maintain the same strategy for each sentence in one syntactic dimension.In order to be able to systematise the procedure, it may help to categorise the structures from a functional perspective and then develop rules for each function identifi ed.Within the given limitations of our stu dy this has to be left for future research.
As previously mentioned, the rephrasing process has consequences for the coherence structure of the text.When we break up complex sentences into simple sentences without or almost without embedding -as we do in the maximum version C -, we risk causing a loss of textual co herence.As it is no longer possible to insert local intrasentential cohe sive ties, we can only create coherence by establishing cohesive ties on intersentential level.However, adding cohesive elements to every simple sentence often produces an awkward style inappropriate for written language.Therefore, in this study we try to minimalise the elements ad ded in the rephrases, thus inserting cohesive elements only where absolute ly necessary.A future study focussed on textual characteristics of legal texts may provide further insight into the requirements for improv ed comprehensibility regarding cohesion and coherence.

Preparation of the experiment
We restrict our experimental data to those sentences which clearly display one of the three syntactic dimensions mentioned above.This is done in order to avoid phenomena assumed to impede the understanding of the sentence in question, which could distort the results of the test.
Where necessary, we therefore slightly change the original sentences (ver sion A) in order to discard any possible interfering elements.Ex ample 4 below shows a sentence from the corpus used for testing the syntactic dimension P. The target phrase is underlined in the test sen tence in 4b.Compared to the original sentence in 4a, it becomes clear that the changes affect the length of other phrases which are not in the fo cus of the present stimulus.The third part of the study consists in testing the rephrases in a psycholin guistic experiment in order to determine whether rephrasing the complex structures of the court decisions improves the comprehensi bility for lay persons.As this is the main focus of the current paper, it will be de scribed in more detail in the following section.

Method
The underlying assumption of our psycholinguistic experiment is that long er reading times equal either deeper processing or more complex texts.Reading times can be measured in a self paced reading experiment (Mitchell 1987) which is chosen as the optimal testing method given the limited resources of the study.Measuring the reading times for the different versions gives an indication of the processing effort the participants need for these versions.Furthermore, self paced reading represents a good indica tor of comprehensibility when combined with compre hension questions and measuring the response latencies, i.e. the time, participants need to answer the comprehension questions.
The 45 participants consist of 36 lay persons, mainly students of Saarland University, and 9 experts, i.e. legal experts and advanced law students.As we intend to test these subjects on the three versions A, B and C, they are split up into three groups consisting of 15 participants (12 lay persons and 3 experts), respectively (see below).We exclude students or experts of linguistics and law from the group of lay persons.The subjects were paid 4 Euros for participating in the experiment.
In the experiment, the texts are presented on two portable computers using the software DMDX5 .The participants' task is to read the texts on the screen.The words of a sentence appear on request by mouse click.The programme logs all mouse clicks and computes the time from one mouse click to the next, thus recording the time the participant takes to pro cess one word.
We use 30 sentences from the rephrasing process as stimuli.For each syn tactic dimension (S, P, N) 10 sentences are chosen.Each of the sentences is realised in the versions A, B and C described above in section 2.2.In order to give the sentences some context, we include fi ller sentences where appropriate.These sentences serve to introduce the circum stances and remain unchanged in all three versions.They are part ly taken from the corpus and partly individually written adapted to the stimuli.Example 5 shows that the fi ller sentences are simpler in their structure and introduce some content which is then elaborated on in the stimulus.
The stimuli (and fi ller) appear in randomised order and are each fol lowed by a comprehension question, which is proofread by a jurist checking the legal content of the question.The question corresponding to the sti mulus in example 5 is given here as example 6.
Like the fi llers these questions also remain unchanged in all versions and have to be answered with yes or no.A short break is included before the question to avoid inadvertent clicking instead of answering the que stion.
We thus obtain the following design.Three complexity conditions (ver sions A, B, C) and two expertise conditions (lay persons, experts) re sult in a 3x2 factorial design.15 participants -12 lay persons and 3 experts -are assigned to each of the three complexity conditions at random.
We do not follow the moving windows paradigm where a word disappears as soon as the next is requested.We expect that the participants might loose track of the highly complex sentences from the court decisions if they cannot move back within one sentence.However, we analyse the reading times with regard to so called windows, i.e. recurring areas in a sentence which indicate additional cognitive load.The analysis does not suggest any noticeable windows.Probably, the parti cipants request the words until they see a complete sentence on the screen and then process the whole sentence.Therefore we only consider aggre gated reading times for complete sentences.The reading times for all individual words of a sentence as logged by DMDX are summed and divided by the number of tokens of each complexity condition as the versions differ in length.The time span from appearance of the comprehension question to its answering, the response latency, is interpreted as logged by the programme.The number of correct responses is summed.Any outliers are eliminated.

Hypotheses
The psycholinguistic test aims at identifying the differences in the cognitive processing of the three rephrased versions of legal texts de scribed above.We review processing by measuring three dependent variables: 1) reading times, 2) response latencies and 3) correctness of res ponses given by the subjects.Independent variables are 1) degree of complexity and 2) expertise.Combining these two types of variables in hypo theses allows us to draw conclusions about the comprehensibility of the three rephrased versions A, B and C.
The group of experts only serves as an explorative group, since it is too small to yield any signifi cant results.In order to permit a com parison between lay persons and experts both groups should be of comparable size.Therefore, we do not formulate any hypotheses in con nection with the comparison.However, we will briefl y discuss the results for this group in section 3.3.We concentrate on detecting dif fe rences in the comprehensibility of the three rephrased versions A, B and C by varying the degree of complexity along three syntactic di men sions S, P and N, as has been explained above.However, we do not intend to compare the difference in comprehensibility between the three grammatical dimensions as this comparison is problematic from a lin guistic point of view.While comparing the effect of varying different syntactic struc tures on the comprehensibility constitutes an interesting research ques tion, we cannot analyse it with our research design.This question would require a design focussed on the different structures and leaving aside the study of legal language.
In consideration of the above mentioned conditions we can formulate the general assumption that rephrasing version A, which displays a very high complexity on all syntactic levels, will increase comprehensibility.The expected effect of the two rephrases on the comprehensibility for lay readers can be described as follows.Maximally simplifying the syntac tic structures should result in the shortest reading times in version C. Simple sentences do not require tracing back complicated structures and should therefore be the version requiring less reading time than the other two versions.However, cognitive processing is not stimulated by these simple structures.According to Groeben & Christmann (1989) this should destruct the reader's motivation to read carefully and then memorise what he/she read.We thus expect that participants read ing version C take longer to answer the comprehension questions and answer them less well than those reading version B which have a cog nitive stimulus in the medium complex sentences.Version A is ex pect ed to perform worst because the complex structures ask too much from the lay persons not familiar with these structures.More precisely, we can establish the following hypotheses for the interpretation of the fi ndings about the three above mentioned dependent variables:

Findings
Our main focus in connection with the fi ndings is the performance of the lay participants.The results for this group's reading times in the three versions A, B and C as displayed in Figure 1 are partially signi ficant, i.e. they partially confi rm our hypothesis H1.In more detail, the dif ference between version A and B is not signifi cant.The difference between version B and version C is signifi cant.This means that the test per sons read version C much more quickly than the other two versions.How ever, reading times for version B which is of medium syntactic com plexity almost equal those for version A as the most complex version.While being much less complex, version B still seems to cost much processing effort.Our hypothesis H2 regarding the response latencies is partly confi rmed as depicted in Figure 2.While the mean latency for lay readers of ver sion A is signifi cantly higher than for readers of version B, the dif ference between version B and C is not signifi cant.Participants thus have most diffi culties in processing version A, therefore they need much more time to answer the comprehension questions than in the other two versions.In the rephrased versions, the processing effort seems clear ly reduced.However participants do not need signifi cantly more time to process the question in version C than in version B. This result in combination with the result for the reading times suggests that the me dium complex version B is not processed clearly better than version C.This picture changes when looking at the fi ndings for correctness of responses for the three rephrased versions A, B and C. The difference between versions A and B is signifi cant as is the difference between versions B and C. The participants reading version B are able to answer much more often correctly than are those reading ver-sion A or C. The difference between A and C is not signifi cant, meaning that both groups of participants have similar diffi culties responding to the comprehension questions, albeit for different reasons.The dif fi culties of participants assigned to version A most likely begin with pro cessing the complex sentences, a problem manifested in the long reading times and continue with considering a response to the comprehen sion question.Participants assigned to version C seem to simply run through the sentences without really memorising what they are reading and feel capable of answering the questions quickly, a fact refl ected in the short response latencies.Because of its medium complexity version B provides enough incentive for the subjects to read the legal text and work towards understanding it correctly.
The results compared by degree of expertise show that the jurists parti cipating in our experiment have longer reading times and response la tencies in all three versions.This is probably due to the fact that versions B and C contradict the reading habits of legal experts: the expert re cog nises the sentences as originating from a legal text and has to incorporate the differences to what he/she expects.Jurists give more correct responses than the lay participants.This does not come as a surprise as the experts compre hend the legal content regardless of the syntactic form.A cautionary note is in place, since these results only constitute tendencies not based on statistical tests.They show the difference between the two recipient groups of court decisions and have to be kept in mind when varying the syntactic structure of court decisions.

Interpretation
We are able to show that, broadly speaking, syntactic rephrases improve pro cessing with lay readers, thus confi rming our general assumption.The experiment leaves no doubt that version A, i.e. sentences from court decisions, massively impedes the comprehensibility for lay readers.The lay persons assigned to this version read longer, consider their re sponses longer and still answer the comprehension questions less well than those assigned to the rephrased versions.We can say that the syn tactic structures of version A are too complex for the participants to understand them correctly.
While reading times do not support our hypothesis that the B-version is processed faster by the lay readers, the combined interpretation of all three dependent variables -reading times, response latencies and correct ness -show that version B leads to an optimal improvement of the com prehensibility.Compared to version A, the participants assigned to version B do not read faster in this version but take signifi cantly faster to answer the comprehension questions and -what is most important -are in the position to answer the questions more often correctly.This shows that, in comparison to version A, the comprehensibility has improv ed.
The C-version does not perform better in this overall interpretation.Although the reading times are signifi cantly shorter in this version than in the other two versions and the participants answer the questions as quickly as the readers of the B-version, they are not able to give as many correct responses as in the B-version.The simple syntactic structures seem to lead to a loss of cognitive motivation for the subjects to read the texts properly in order to understand them correctly.There are two possible explana tions for this: fi rst, the overly simple structures of this version may induce the readers to not start cognitive processing of what they are reading.This explanation is in line with Groeben & Christmann's (1989) motivation-and cognition-based approach.The se cond explanation is that the structures are resolved beyond what the lay reader expects.The logical relation between clauses is shifted to the sentence level, leading to a text structure which is less cohesive.Version B creates enough textual coherence to linguistically link parts of the text in a meaningful way.This is not the case in the maximal simplifi cation of version C. In this version information is spread so spar se ly in short and simple sentences that it partially lacks cohesive ties to establish meaningful relations in the text as discussed in section 2.2.This lack of textual coherence may complicate processing and decrease compre hensibility and, fi nally, result in a low amount of correct responses to comprehension questions.Thus, both the A-and the C-version do not con form to the reading habits of the lay reader, leaving version B which fol lows the degree of syntactic complexity found in the newspaper reports as the version understood best by the lay readers.

Conclusions and outlook
The study presented here gives us an in-depth look at the workings of three syntactic features of court decisions.Each of the three metho dol-ogical steps contributes its part to the overall picture.First, we built a cor pus annotated with syntactic information on German court decisions and related text types.The interpretation of this corpus in itself may yield valuable results for the analysis of German legal language.We used the cor pus to investigate how syntactic specifi cities of court decisions are varied in the related text types of press releases and newspaper reports on the decisions.The corpus analysis also enabled us to retrieve the most distinctive instances from the subcorpus of court decisions, which were then used for the rephrasing process.
The second step showed the possibilities and limitations of the rephras ing process.On the basis of the instances retrieved from the corpus, we elaborated two rephrases: a medium complex one with syn tactic structures demanding some cognitive effort from the recipient and a simple one which builds on the assumption that maximally simple structures are memorised best.We identifi ed limitations with each of the three syntactic features varied in this process.The specifi cities of each instance rephrased made it obvious that there is a need for more fi negrained (functional) analyses of complex structures in a future study.This may ultimately lead to rules for automatically rephrasing complex syn tactic structures.
Finally, the results of the psycholinguistic experiment help us understand what a comprehensible court decision could look like.The study showed that rephrases with a medium degree of syntactic complexity similar to that of newspaper reports score better when tested with lay per sons.This clearly shows that the language of court decisions is not adapt ed to the needs of lay citizens as one of two recipient groups of court decisions.However, the comparison of the two groups of expertise, i.e. the two recipient groups, indicated that legal experts will not easi ly accept rephrases from the type scoring best with lay persons as they contradict the jurists' expectations.This may change in the future if research fi ndings such as those presented here are used for teaching legal writing to law students.If they learn a plain writing style at an early stage in their formation it will become natural to them to put complex facts in a simple way.
In a broader view, a cautionary note is in place.Syntactic changes can not remedy the inherent incomprehensibility of the law itself.They on ly operate on the surface level.Furthermore, on this surface level, the changes should not be limited to the level of syntax on which we have con centrated in this study but have to be combined with changes in the lexis used as well as in the overall cohesive structure of the text.For this purpose, continuative studies on these levels have to be conducted.It remains to be seen, however, what these combined linguistic efforts can achieve towards the goal of improving the comprehensibility of legal language.

H1
Reading times: A > B > C The reading times for version A are longer than the reading times for version B, which is of medium complexity.The reading times for version B are longer than the reading times for version C, which constitutes a maximum simplifi cation.H2 Response latencies: A > C > B The response latencies for version A are longer than the response latencies for version C. The response latencies for version C are longer than the response latencies for version B. H3 Correctness of responses: A < C < B In version A less correct responses are given than in version C. In version C less correct responses are given than in version B.

Figure 1 :
Figure 1: Lay persons' reading times for the three versions

Figure 3 :
Figure 3: Correctness of lay persons' responses in the three versions Table 2 dis plays the annotation results for the complexity of the phrases in the court decisions, press releases and newspaper reports, respectively.

Table 2 :
Complexity on phrase level