Figure 40. helicopter, RSL Figure 41. experiment, RSL

The situation with the phonemic handshape 34 is very similar to the one with 14 which has been described above. Handshape 34 has two selected fingers compared to one selected finger in 14. Furthermore, handshape 34 has four allophones, two of which completely reflect the situation with the phoneme 14, namely handshapes 33 (thumb in the neutral position) and 38 (selected fingers flattened). Both 33 and 38 are phonetically motivated. In addition to that, handshapes 35 and 39 are also allophones of 34. They have an extended thumb. The line of argumentation for these handshapes being allophones of 34 is the same as for handshapes 15 and 19 being allophones of 14.

As for the handshape 21, it can be sometimes iconic and sometimes phonemic. For example, in a sign to-associate (Figure 42) it is iconic and refers to two objects interconnected. On the contrary, in a sign to-be-alert (Figure 43) this handshape is phonemic. Phonemic handshape 21 has no allophones.

Figure 42. to-associate, RSL Figure 43. to-be-alert, RSL, second frame

Handshape 51 (i.e. “2”-handshape) is also phonemic. It has two allophones: 50 (thumb in the neutral position) and 55 (selected fingers flattened). Handshape 50 is explained by individual differences (the same as handshape 13 and 33), while handshape 55 is phonetically motivated.

Another phonemic handshape is handshape 75, which is also sometimes predicted from phonetics and sometimes occurs in initialized signs (handshape 75 stands for letter “Ж”). For example, it is initialized in the sign bourgeois (Figure 44) and it is phonetically predicted in the sign hygienic-pad where the fingers are flattened because of the pointing in the direction of the signer's body (Figure 45). In the sign to-promote (Figure 46) it is phonemic. The movement from the body is iconic, whereas the handshape itself does not refer to any object. One might think that flattened hands here are phonetic, because hands are facing each other. However, it is not the case here. Consider a sign door (Figure 47). It has handshape 71, but the hands are facing each other in a similar way as in the sign to-promote. So, this flexion of metacarpophalangeal joints in the case with handshapes 71 and 75 is not phonetic, but phonemic. In addition to that, handshape 75 has two allophones which differ in the position of the thumb: 73 and 74. Handshape 73 is phonetically motivated, while 74 is motivated iconically.

Figure 44. bourgeois, RSL Figure 45. hygienic-pad, RSL, frame 2

Figure 46. to-promote, RSL



Figure 47. door, RSL

Handshape 137 is also phonemic. An example of a sign with it is depicted on Figure 48. The active hand's configuration in the sign criticism is 137. Furthermore, this handshape does not have any allophones.

Figure 48. criticism, RSL Figure 49. cognac, RSL

Handshape 30 (i.e. “У”-handshape) is phonemic too. Nevertheless, sometimes it appears in initialized sign, such as to-impress. In addition to that handshape 30 can occur in iconic signs, e.g. iron (Figure 50). In this sign it stands for an iron as an object. While in signs like canteen (Figure 51) this handshape is phonemic. Handshape 30 does not have any allophones.

Figure 50. iron, RSL Figure 51. canteen, RSL, frame 1

Handshape 59 can be iconically motivated or pure phonemic. Its only allophone is a handshape 58 (thumb in the neutral position). The sign cognac (Figure 49 above) is an example of pure phonemic occurrence of handshape 59.

Handshape 68 can be phonemic or in some cases iconic. It has two allophones: handshapes 167 and 163. An example of phonemic realization of handshape 68 is shown on Figure 52 - popular. Handshape 167 is always iconic, it appears in such signs as chair, bed, and etc. 163 is also iconic. For instance, sign migraine (Figure 53) has handshape 163, where its configuration stands for a spread of pain in through the head and at the same time holding a head.

Figure 52. popular, RSL Figure 53. migraine, RSL

As for the phonemic handshape 110, it has two allophones: 108 and 109. They differ from each other and from handshape 110 with the size of the gap between fingers, 110 having the largest gap. Furthermore, handshape 110 can sometimes appear in iconic signs, e.g. river, where a river is traced with this hand configuration.

Handshape 125 also belongs to manual alphabet, where it stands for letter “Ч”. Therefore, in some cases it occurs in initialized signs, while other occurrences are phonemic. Figure 54 depicts one of its phonemic occurrences - sign chance. This handshape does not have allophones.

Figure 54. chance, RSL



Handshape 111 is an index finger's fingertip opposition with the flattened thumb. Together with its allophone 115 it appears in 2% of annotations. Handshape 111 can be sometimes iconically motivated. For instance, in the sign pizza (Figure 55) it is a part of tracing of a shape of a slice of pizza. On the contrary, handshape 115 is always iconic. We can draw an analogy here with a handshape 152 (allophone of the “O”-handshape). Handshape 115 similarly refers to holding something between fingertips.

Figure 55. pizza, RSL, frame 2

Handshape 136 (i.e. “C”-handshape) is phonemic too, although sometimes it appears in initialized signs and sometimes it can also appear in iconic signs (e.g. in the sign pipe (Figure 56) it traces a shape of a pipe). Handshape 136 has two allophones, namely 135 and 134, both of them are iconically motivated. The difference between 135, 134 and 136 is in the size of the gap between the thumb and other fingers.

Figure 56. pipe, RSL

The handshape 79 has a form of four non-spread bent fingers. The position of the thumb is not important unless the handshape is iconic or predicted from phonetics. So that this phoneme has allophones 76 and 78, with the thumb in the neutral position and with the thumb bent respectively. Crucially, there is a minimal pair for phonemes 79 and 71. The sign contract (Figure 57) is symmetric two-handed sign with a phonetic handshape 71, while the sign pie (Figure 58 The sign is not represented in Spreadthesign. The video is taken from RSL corpus (Burkova 2015)) is also symmetric two-handed, but with handshape 76. Handshapes 76 and 78 are usually iconically motivated. For example, in the aforementioned sign pie two hands with handshape 76 reflect the shape of a small pie.

Figure 57. contract, RSL Figure 58. pie, RSL

Handshape 29 has one selected finger, pinky finger, which is simply extended. It appears sometimes in iconic signs (e.g. in the sign thin it stands for something long and small somebody is very thin), however, still has a lot of cases where it is phonemic. For instance, it is phonemic in the sign bad (Figure 59). This handshape does not have any allophones.

Figure 59. bad, RSL



Figure 60. buckwheat, RSL

Another phonemic handshape is 61. It has one allophone - handshape 60 (thumb is bent). Handshape 60 is phonetically predicted. For example, in the sign buckwheat (Figure 60 above) the second hold's handshape is 60 and not 61 because the fact that the transition from extended fingers to bent fingers in selected finger, namely index and middle fingers, is easier to do when this movement applies to all possible fingers, not only to selected ones, but to the thumb too. To sum up, together with its allophone this handshape occurs only 12 times in annotations.

Handshape 53 is not very frequent, but it is phonemic in most of the signs, and iconic in some. In the sign daughter (Figure 61) handshape 53 is phonemic, while in the sign vaccination (Figure 62 below) it is iconic, and the handshape refers to a hand of a doctor pressing on a syringe. This handshape does not have any allophones.

Figure 61. daughter, RSL



Figure 62. vaccination, RSL

Handshape 83 is under question as a phoneme due to the fact that it occurs only in one sign - middle (Figure 63). In this sign it is not phonetically predicted and it is not iconically motivated. So, according to van der Kooij's (2002) Phonetic Implementation rules, we can postulate that this handshape is phonemic.

Figure 63. middle, RSL

To conclude, in this section we discussed which phonetic handshapes for RSL can be considered as phonemes, according to van der Kooij's (2002) model. Evidently, just using Phonetic Implementation rules is not enough for postulating phonemes. One should consider individual differences between signers, such as having a thumb extended or in the neutral position and etc. In addition to that, it is relevant for RSL to annotate all handshapes for whether they are relaxed or not, because this phonetic feature most likely distinguishes two phonemes, namely lax “B”-handshape (aka. lax 71) and normal “B”-handshape (aka. 71). Our preliminary analysis for now revealed that RSL has 23 phonemic handshapes, however, in the future perspective there might turn out to be less phonemic handshapes due to the aforementioned reasons.



4.5 Cross-linguistic comparison

In this section I compare main results for NGT (van der Kooij 2002) and AdaSL (Nyst 2007) with my results on RSL, firstly, from the perspective of phonetics and, secondly, from the perspective of phonology and phonemes. In addition to that, I compare phonetic inventories of YSL (Bauer 2012), IUR (Schuit 2014), and Kata Kolok (Marsaja 2008) which are rural languages with the phonetic inventory of RSL, urban sign language.

The idea of the comparison of cross-linguistic data on the most frequent phonetic handshapes was coined by van der Kooij (2002). According to her, “the relative frequency of the most frequent handshapes of the dominant hand are highly similar in unrelated sign languages (van der Kooij 2002: 92 cited by Nyst 2007: 61)”. Table 14 below represents comparison of the data on the most frequent handshapes in AdaSL, NGT, ASL, BSL, and ISL with RSL. It is obvious from this table that not all handshapes from RSL top-15 for the dominant hand are reflected in this table. Moreover, in RSL top-15 frequent handshapes have their relative frequency equal to or above 1.9%, and this is not the case for four handshapes in the sample in this table with comparison. Furthermore, the most frequent for all other languages “B”-handshape occurs in less than 20% of dominant hands in RSL, so that the numbers of relative frequencies do not really match across languages either. However, the most frequent handshapes still somewhat match. RSL has 116 phonetic handshapes, as it has been established before, and still 10 out of supposedly 14 most frequent handshapes across sign languages do belong to the most frequent in RSL. This is clearly not coincidental, although the list of the most frequent handshapes across sign languages should be checked on more data from other languages and updated.



Table 14. Relative frequency of handshapes in the dominant hand in RSL compared with AdaSL, NGT, ASL, BSL, ISL (from Nyst, 2007: 61 and from van der Kooij, 2002: 93 cited by Nyst, 2007: 61)

Handshape (HamNoSys №)	RSL	AdaSL	NGT	ASL	BSL	ISL
B (i.e. 69, 71, 70, 72)	15%	25%	22%	23%	24%	20%
1 (i.e. 14, 13, 15, 18, 19)	16%	19%	15%	14%	15%	14%
S (i.e. 2)	2.1%	14%	10%	9%	9%	10%
bO/closed bB (i.e. 137)	1.8%	6%	5%	?	3%	8%
(Lax) O (i.e. 133)	0.4%	4%	<1%	4%	2%	6%
5 (i.e. 87)	7%	3%	13%	7%	7%	8%
A (i.e. 3)	3%	2%	4%	3%	5%	3%
V (i.e. 50, 51, 55)	2.2%	2%	3%	4%	3%	4%
X (i.e. 21)	2.1%	2%	1%	4%	?	1%
F (i.e. 144, 152, 160)	4%	<1%	5%	4%	3%	6%
H (i.e. 33, 34, 35, 39)	3%	<1%	1%	4%	4%	4%
(Lax) C (i.e. 136, 135, 134)	1.1%	1%	2%	7%	2%	3%
bC (i.e. 110, 108, 109)	1.6%	0%	1%	<1%	1%	5%
C+spr (i.e. 97)	2.1%	0%	3%	?	4%	?

As for the secondary hand, neither van der Kooij (2002), nor Nyst (2007) do not postulate global cross-linguistic similarities like for the active hand. However, there are some trends too. For instance, Nyst (2007) shows that in AdaSL 83% of secondary hand handshapes have all fingers selected. This number for RSL is lower, 58.5%, although it is clear that handshapes with all fingers selected are more preferred on the secondary hand, than on the active hand in RSL. For example, “B”-handshape phoneme (69, 71, 70, and 72) occurs in 33.2% of secondary hand annotations and only in 15.4% of active hand annotations.

Probably, some of the differences in the most frequent phonetic handshapes in some language can be partially explained by how many signs of each type by handedness this particular language has. I hypothesize that due to the fact that there have been established some clear correlations between handedness and handshape type, such as: secondary hand in asymmetric two-handed signs can be of a limited number of unmarked handshapes, namely 1/3, 2, 71/69, 136, 87, 144, or 33/34 (in my notation) (Battison 1978 on the basis of ASL, true for AdaSL too (Nyst 2007)); in symmetric two-handed signs handshapes usually have less fingers selected (true for NGT (van der Kooij 2002: §2.3) and RSL); most asymmetric two-handed signs are iconic (van der Kooij 2002), thus, they should allow more complex handshapes. So, if we know how many signs of each type a sign language has, we might be able to make some predictions on the frequency of phonetic handshapes there. For example, in RSL there are more two-handed signs than one-handed (67% vs. 33%), while in AdaSL and NGT this distribution is almost half and half. So, we might expect the frequency of handshapes with all fingers selected to be higher in general in RSL. However, this is not the case, because we should also take into account high percentage of iconic signs in RSL, and also such parameters as cross-linguistic differences, language phonetic complexity (e.g. Nyst (2007) claims that AdaSL has low complexity, while NGT has high phonetic complexity), etc. The data of this research is not suitable for looking for the correlations of this kind, so, this is an open question for a future research.

For IUR (Schuit 2014), Kata Kolok (Marsaja 2008), and YSL (Bauer 2012) I have information on phonetic handshapes in general regardless of the handedness. These languages as well as AdaSL are rural sign languages, and naturally they have rather small inventories of phonetic handshapes. YSL has 33 phonetic handshapes as well as IUR, while Kata Kolok has 28 phonetic handshapes. Schuit (2014) notes that all known rural sign languages have phonetic handshape inventories of a size of approximately 30 handshapes. She also hypothesizes (Schuit 2014: 36) that it was quite unlikely for any sign language, even for an urban sign language to have a phonetic handshape inventory larger than 80 handshapes. However, RSL data in this research shows that RSL has at least 116 phonetic handshapes.

The last but not the least, I discuss phonemic inventories of RSL with AdaSL and NGT. NGT has a set of 31 phonemic handshapes (van der Kooij 2002), while AdaSL has only seven (Nyst 2007). Preliminary conclusion for RSL is a set of 23 phonemic handshapes. The sets of phonemic handshapes for these languages somewhat intersect. Phonemic inventory of AdaSL comprise of the following phonemes: “1”-handshape (i.e. 34/33), “A”-handshape (i.e. 3), “2”-handshape (i.e. 50/51), “B”-handshape (i.e. 69/70/71), “S”-handshape (i.e. 2), “O”-handshape (i.e. 144), and an empty handshape (something like lax “B” handshape, i.e. lax 71). The phonemic inventory of NGT with images can be found in (van der Kooij 2002: 154-158) See these pages in Appendix 5.. All phonemic handshapes of AdaSL except for the “S”-handshape (i.e. 2) are phonemic in RSL too. In RSL “S”-handshape (i.e. 2) occurs only in iconic contexts. On the contrary, NGT has 13 phonemes which are not phonemic in RSL, namely: №4, №8, №10 (i.e. 133), №13 (i.e. 15), №15, №16, №18, №21, №23, №26, №27, №30, and №31 See Appendix 5. 18 other phonemic handshapes of NGT are phonemic in RSL too.

In general, there are some common tendencies in what is usually phonologically distinctive in discussed sign languages. For instance, there is a lax vs. tense distinction in AdaSL. RSL also has this distinction in “B”-handshape phoneme, namely 71 and lax 71 handshapes. On the contrary, in IUR lax handshapes and curved handshapes are allophones of their tense counterparts (Schuit 2014). As for the NGT, there is no information on tenseness as a distinctive feature so far, because it has not been research yet, according to my knowledge. In addition to that, Nyst (2007) postulated that flexion of metacarpophalangeal joints of the hand does not help to distinguish any phonemes in AdaSL. This statement holds for RSL data. For example, it has been observed that phonemic handshapes 14 and 34 have allophones with flattened selected fingers, and these allophones' handshapes are always predicted from phonetics. Furthermore, in IUR the size of the gap between thumb and fingers, i.e. degree of opening, induces allophones (Schuit 2014). This is true for RSL too. Consider RSL allophonic handshapes 108-110 (i.e. one finger selected and rounded), 112-114 (i.e. one finger selected and flattened), 138-140 (i.e. four fingers selected and flattened), and 134-136 (i.e. “C”-handshape).



4.6 Applicability of the phonological frameworks

Different phonological frameworks are not equally descriptive and efficient on the material of a particular sign language, in our case RSL. The analysis of the phonological structure of RSL has revealed a number of properties which affect applicability of different phonological models.

The first quality of RSL to be discussed is complex movement in signs. In RSL there are more signs with complex movement than with simple movement, 362 (71%) and 148 (29%) signs respectively. A lot of the phonological models are based on the idea that movements can in the most of the cases be predicted from holds, therefore, movements as a separate category/node/layer in the structure can be omitted. However, complex movements cannot be predicted from the knowledge of the holds. In the situation with the complex movements many those phonological models postulate additional rules. For instance, the Movement-Hold model postulates seven additional rules to describe complex movements and other phonological processes, such assimilation, gemination, and etc. The other two models which disregard the movement status are the Hand-Tier model and the Dependency model. Both of them specify only a “manner” of movement, but under the hand configuration node (in the Hand-Tier model) or on par with the place of articulation node (in the Dependency model). On the contrary, the Prosodic model has a separate node for movement in its structure. Since in RSL signs with the complex movement outnumber signs with the simple movement, it is more reasonable to apply to it such model which perceives movement as an important node in the structure and allows complex movements in the structure without overcomplicating it with additional rules. If a model postulates additional rules to describe something very frequent, it is not descriptive enough. Therefore, in this sense the Prosodic model is the most suitable theoretical framework for RSL.

However, annotation of all signs in the Prosodic model is time consuming. RSL has a lot of two-handed signs, to be precise 67% of the dataset. This means that for each hand all phonological features need to be specified separately. Fortunately, an annotation with the help of HamNoSys is much faster in a sense that it does not require to describe everything about handshape separately. Therefore, the most suitable framework for RSL is the Prosodic model integrated with HamNoSys for handshape (and potentially place of articulation and orientation) annotation, and with more detailed prosodic features (i.e. movement features) subtree with respect to Mak & Tang's (2011) theory (see the end of section 2.1.5 for the discussion of this theory).



5. Discussion

This section is devoted to describing what can be developed in the algorithm of hold extraction and which directions of the analysis of the RSL phonology can be explored in future research.

Firstly, annotation can be extended to other phonological features, such as location, orientation, and movement. Location and orientation features can also be annotated with the help of HamNoSys and analyzed in a similar way as handshapes were analyzed in this work. It would be interesting to search for interactions between these three phonological features. For example, this has already been done for AdaSL by Nyst (2007) for handshape and location phonological features, and it could be compared with the RSL data in the future. Furthermore, adding more phonological features into the annotation can give us an opportunity to see whether phonological features have equal status with respect to each other and form a feature hierarchy. For instance, in the Prosodic model, orientation does not have the same status as handshape, location, and movement. Orientation is specified on a lower level, under the handshape and place of articulation nodes. If the data are annotated for all of these features, it will be possible to quantitatively test this Prosodic model's proposal about the orientation feature status in the phonological structure. In addition to that, the work on movement types can be extended to the whole dataset. I hypothesize that sign languages have cross-linguistic preferences for movement types. Furthermore, the movement feature can also have some interactions with other phonological features and interactions of inner movement features, such as on which joints the full repeat or return features occur more often.

Secondly, this work aimed at discussing the inventory of handshapes, both phonetic and phonemic, but one should keep in mind that the inventory of phonetic handshapes which was made with the help of automatic hold extraction algorithm might not be exhaustive. One thing that could have caused this is the fact that the algorithm did not have perfect accuracy, so that in the 23.3% of signs holds were estimated erroneously. So, the information on handshapes in holds of these signs is missing. The other thing that could have had an impact on retrieving the inventory of phonetic handshapes is the fact that the dataset has not been annotated for tense/lax distinction. However, as I showed in section 6.4, there is a chance that this distinction is phonological for the “B”-handshape. In addition to that, the phonetic inventory of handshapes would have been larger too if there was an annotation for tenseness. Finally, the Spreadthesign dictionary itself is not an exhaustive collection of the RSL lexis, which also leads to a loss of some data on the RSL handshapes. As a result, the methodology used in this work cannot ensure that the inventory of phonetic handshapes is exhaustive. However, it still covers much of the RSL lexis, and, consequently, a lot of holds and a lot of phonetic handshapes.

Thirdly, the phonemic inventory of RSL established in this work is quite preliminary. It was estimated only on the basis of van der Kooij's (2002) Phonetic Implementation Rules, namely on the basis of whether a handshape can be phonetically predicted or iconically motivated. Evidently, this is not enough due to the fact that there are also individual differences between signers. In the future perspective it is essential to collect videos of the same signs produced by different signers in order to account for the individual differences. This way it would be possible to establish allophones with more precision. In addition to that, minimal pairs should be discussed where possible.

Finally, the algorithm could be developed in order to reach higher accuracy level. For example, it could be more efficient to use most of the previous methods as preprocessing (e.g. manually deleting compounds, cutting out frames with setting) and then apply OpenPose python module for an actual hold extraction. However, this method requires more time, because OpenPose is based on machine learning. This work put a particular emphasis on achieving hold extraction on the small-scale data, namely the Spreadthesign dictionary, and employing time-efficient linguistically-driven methods. In the future perspective the hold extraction algorithm can help with the sign language recognition (SLR) for RSL and potentially for other sign languages. This kind of linguistically-driven SLR can be implemented the following way. First, one needs to make a list of correspondences between a list of meanings and the phonemes which correspond with those meanings in RSL. If each sign has a corresponding known beforehand phonological structure, this structure can be recognized in a new video where this sign comes in the citation form. The only remaining problem is to teach the SLR program to do two things: 1) to divide a video of signing into separate signs and sentences; 2) to `understand' how each type of phonological processes like assimilation changes the citation form of a sign. These two problems are left for future research, and after they are implemented, the sign recognition of RSL can go in the following steps: first, the video of signing is divided into sentences, then signs; then these videos with separate signs are cleaned from frames with setting images, then holds are extracted, and then the correspondences between holds and meanings are established.

All in all, the technique for the hold extraction presented in this work can both help the phonological analysis and be useful for SLR. It can also be upgraded in a number of ways. In particular it can be developed in order to get higher accuracy on RSL, it can also be extended to other sign languages, and it can be used for further analysis of other phonological features, such as place of articulation, orientation, and movement type, and the interactions between these features.



Conclusions

This research achieved two main goals. First of all, it at least partially automated a process of the phonological analysis of a sign language's lexis. Hold extraction makes the process of data annotation much faster. All previous research (van der Kooij 2002; Nyst 2002; Schuit 2014; etc) have exploited only manual methods of data annotation. The accuracy of hold extraction achieved by this algorithm on the whole RSL dataset is 76.7%. Although the algorithm developed in this work applies only to RSL, it still can be updated in the future to make it applicable to potentially any sign language.

Secondly, this research coined the first extensive analysis of RSL phonetic handshapes and produced preliminary conclusions on the phonemic handshape inventory of RSL. RSL turned out to have an unusual from the typological point of view size of the phonetic handshape inventory - 116 phonetic handshapes in total for both hands. This could be explained by the higher degree of iconicity in RSL compared with other investigated sign languages, such as NGT, Auslan, AdaSL, YSL, IUR, and Kata Kolok. Previously, Schuit (2014) proposed that the probability of finding a sign language with more than 80 phonetic handshapes is very low. So, RSL is the first discovered sign language with such a large phonetic handshape inventory. Surprisingly, the size of the RSL phonemic handshape inventory is quite standard for an urban sign language, having at most 23 phonemic handshapes. For instance, NGT, also an urban sign language, has 31 phonemic handshapes. In addition to that, 10 phonemic handshapes out of the 15 most frequent handshapes in RSL belong to the most cross-linguistically frequent handshapes (Nyst 2007).

To sum up, the results of this work can be useful for further research on RSL phonology, its phonological features interactions, and phonological processes. It already provides a description and an analysis of handshape inventory and its interaction with handedness and preliminary conclusions on the movement type frequency in RSL. Moreover, this work can have an impact on RSL automatic recognition research, because the knowledge of phonemes and their correspondences to meanings can be used for translation into the spoken language after chunking signing into sentences and words in their citation form.



References

1. Battison, Robbin (1978). Lexical borrowing in American sign language. Silver Spring, MD: Linstok Press.

2. Bauer, Anastasia (2012). The use of signing space in a shared sign language of Australia. PhD Dissertation, Universitдt zu Kцln.

3. Brentari, Diane (1998). A prosodic model of sign language phonology. Cambridge, MA: MIT Press.

4. Brentari, Diane (2002). Modality Differences in Sign Language Phonology and Morphophonemics. In: R. P. Meier, K. Cormier & D. Quinto-Pozos (eds.), Modality and Structure in Signed and Spoken Languages. Cambridge: Cambridge University Press, p. 35-64.

5. Brentari, Diane (2012). Phonology. In: R. Pfau, M. Steinbach & B. Woll (eds.), Sign Language. An international handbook. Berlin: Mouton de Gruyter, p. 21-54.

6. Bross, Fabian (2015). Chereme. In: T. A. Hall & B. Pompino-Marschall (eds.), Dictionaries of Linguistics and Communication Science (Wцrterbьcher zur Sprach- und Kommunikationswissenschaft, WSK). Volume: Phonetics and Phonology. Berlin, New York: Mouton de Gruyter.

7. Channon, Rachel & Harry van der Hulst (2011). Introduction: Phonetics, Phonology, Iconicity and Innateness. In: R. Channon & H. van der Hulst (eds.), Formational Units in Sign Languages. Berlin: de Gruyter, p. 1-17.

8. Crasborn, Onno, Harry van der Hulst & Els van der Kooij (1998). Phonetic and phonological distinctions in sign languages. Paper presented at InterSign workshop on phonology, December 1998, Leiden, Manuscript.

9. Crasborn, Onno & Han Sloetjes (2008). Enhanced ELAN functionality for sign language corpora. In: O. Crasborn, T. Hanke, O. Efthimiou, I. Zwitserlood, H. Thoutenhoofd (eds.), Proceedings of the Workshop on Representation and Processing of Sign Languages. Paris: ELRA, p. 39-43.

10. Filimonova, Elizaveta (2015). Funkcional?no-semantiиeskaв kategoriв aspektual?nosti v russkom ћestovom вzyke [Functional-semantic Category of Aspect in Russian Sign Language]. PhD Dissertation, Novosibirsk State Technical University.

11. Hanke, Thomas (2004). HamNoSys - representing sign language data in language resources and language processing contexts. In: O. Streiter & C. Vettori (eds.), Proceedings of the Workshop on Representation and Processing of Sign Languages. Paris: ELRA, 2004, p. 1-6.

12. Hulst, Harry van der (1995). The Composition of Handshapes. In: Working papers in linguistics, 23. Trondheim: University of Trondheim, p. 1-17.

13. Johnston, Trevor & Adam Schembri (2007). Australian Sign Language (Auslan): An introduction to sign language linguistics. Cambridge: Cambridge University Press.

14. Kimmelman, Vadim, Anna Klezovich, & George Moroz (2019). Quantitative analysis of lexical iconicity in sign languages. In: Proceedings of the Conference 41. Jahrestagung der Deutschen Gesellschaft fьr Sprachwissenschaft.

15. Koller, Oscar, Hermann Ney, & Richard Bowden (2016). Deep Hand: How to Train a CNN on 1 Million Hand Images when your Data is Continuous and Weakly Labelled. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, p. 3793-3802.

16. Kooij, Els van der (2002). Phonological categories in sign language of the Netherlands: the role of phonetic implementation and iconicity.

17. Liddell, Scott (1984). think and believe: Sequentiality in American Sign Language. Language, 60, p. 372-392.

18. Liddell, Scott & Robert E. Johnson (1994). American Sign Language: The Phonological Base. Sign Language Studies, 64, p. 195-278.

19. Mak, Joe & Gladys Tang (2011). Movement types, repetition and feature organization in Hong Kong Sign Language. In: R. Channon & H. van der Hulst (eds.), Formational Units in Sign Languages. Berlin: de Gruyter, p. 315-337.

20. Marsaja, I. Gede (2008). Desa Kolok. A deaf village and its sign language in Bali, Indonesia. Nijmegen: Ishara Press.

21. Metaxas, Dimitris, Mark Dilsizian, & Carol Neidle (2018). Linguistically-driven Framework for Computationally Efficient and Scalable Sign Recognition. In: N. Calzolari, K. Choukri, Ch. Cieri, Th. Declerck, S. Goggi, K. Hasida, H. Isahara, B. Maegaard, J. Mariani, H. Mazo, A. Moreno, J. Odijk, S. Piperidis, T. Tokunaga (eds.), Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018). Paris: ELRA.

22. Nyst, Victoria (2007). A Descriptive Analysis of Adamorobe Sign Language (Ghana). PhD Dissertation, University of Amsterdam.

23. Nyst, Victoria (2012). Shared sign languages. In: R. Pfau, M. Steinbach, & B. Woll (eds.), Sign Language. An international handbook. Berlin: Mouton de Gruyter, p. 552-574.

24. Цstling, Robert, Carl Bцrstell, & Servane Courtaux (2018). Visual iconicity across sign languages: Large-scale automated video analysis of iconic articulators and locations. Frontiers in Psychology, 9, 725.

25. Pan Du, Warren A. Kibbe, & Simon M. Lin (2006). Improved peak detection in mass spectrum by incorporating continuous wavelet transform-based pattern matching. Bioinformatics, 22, 17, p. 2059-2065.

26. Prillwitz, Siegmund, Regina Leven, Heiko Zienert, Thomas Hanke, & Jan Henning (1989). Hamburg notation system for sign languages. An introductory guide, HamNoSys version 2.0. Hamburg: Signum Press.

27. Rakowski, Alexander & Lukasz Wandzik (2018). Hand Shape Recognition Using Very Deep Convolutional Neural Networks. In: Proceedings of the 2018 International Conference on Control and Computer Vision (2018), ACM, p. 8-12.

28. Sandler, Wendy (1989). Phonological Representation of the Sign: Linearity and Nonlinearity in American Sign Language. Dordrecht: Foris Publications.

29. Chomsky, Noam & Morris Halle (1968). The sound pattern of English. In: Studies in language (Harper & Row eds.).

30. Schuit, Joke (2014). Signs of the arctic: Typological aspects of Inuit Sign Language. PhD Dissertation, University of Amsterdam.

31. Stokoe, William (1960/2005). Sign Language Structure: An Outline of the Visual Communication Systems of the American Deaf. Journal of Deaf Studies and Deaf Education, 10, 1, p. 3-37.

32. Tang, Gladys (2007). Hong Kong Sign Language: A trilingual dictionary with linguistic descriptions. Hong Kong: the Chinese University Press.

33. Vainio, Martti, Antti Suni, & Daniel Aalto (2013). Continuous wavelet transform for analysis of speech prosody. In: B. Bigi & D. Hirst (eds.), Proceedings of the Conference TRASP 2013 - Tools and Resources for the Analysis of Speech Prosody. P. 78-81.



Appendix

Dactyl or number-based basic handshapes in RSL

A-handshape B-handshape

1-handshape 2-handshape

O-handshape П-handshape



C-handshape У-handshape

5-handshape

All code used in the chronological order

Pictures for 13 signs from RSL used for testing (`absolutely nothing', `to adapt', `to fall in love', `choice', `to breathe out', `hygienic pad', `rage', `debt', `daughter', `comet', `love', `sun', `flower') absolutely-nothing



to-adapt

to-fall-in-love

choice

to-breathe-out



hygienic-pad

rage



Daughter

comet

love

Sun



flower

Handshapes chart (Hanke 2010) numbered



Additional handshapes which does not appear in the original handshapes chart 161 162 163

164 165 166

167 168 169

170 171



Pages 154-158 from van der Kooij (2002)



Размещено на Allbest.ru

...